Metrics details Smart contracts are changing many business areas with blockchain technology but they still have vulnerabilities that can cause major financial losses Because deployed smart contracts (SCs) are irreversible once deployed fixing these vulnerabilities before deployment is critical This research introduces a new method that combines code embedding with Generative Adversarial Networks (GANs) to find integer overflow vulnerabilities in smart contracts we can vectorize the source code of smart contracts while keeping all of the important contract characteristics and going beyond what can be achieved with conventional textual or structural analysis Synthesizing contract vector data using GANs alleviates data scarcity and facilitates source code acquisition for training our detection system The proposed method is very good at finding vulnerabilities because it uses both GAN discriminator feedback and vector similarity measures based on cosine and correlation coefficients Experimental results show that our GAN-based proactive analysis method achieves up to 18.1% improvement in accuracy over baseline tools such as Oyente and sFuzz Obtaining numerous suitable source codes is labor- and resource-intensive because of the Ethereum network and node limitations The generator generates data; the discriminator compares that generated data with real data Addressing these shortcomings is critical to improving SC security Researchers are exploring vectorization and graph-based techniques with Graph Neural Networks (GNNs) showing the potential to capture contract features Limited access to high-quality datasets due to privacy and legal concerns remains challenging Researchers are exploring data augmentation and few-shot learning to enhance SC vulnerability detection This article introduces a method for detecting vulnerabilities in SCs that combines GAN with code embedding We train a GAN model on integer overflow vulnerabilities by transforming SC source codes into vector representations using code2vec The GAN discriminator detects vulnerabilities and performs vector similarity analysis while the GAN generator expands the dataset this approach preserves contract properties through Abstract Syntax Tree (AST) vectorization and enables deep learning on limited data Dual similarity detection accuracy is enhanced using GAN feedback WE have developed a code embedding and GAN-based vulnerability detection methodology for integer overflow vulnerabilities AST-based representation retains contract properties We are using GAN data augmentation for small-sample deep learning and efficiency tests on 150 public Ethereum contracts and there is no need for human involvement Since most blockchain nodes are required to verify the validity of the SC Adjust it to suit the various settings you are working with These vulnerabilities must be addressed to secure SCs and prevent financial losses SCs were developed with the goals of automating contract execution removing the need for dependable third parties and enhancing transaction security and efficiency Researchers have focused a lot of attention on SC security because of the widespread use of these contracts in industries including finance which shaped blockchain technology and security testing SCs must be verified and maintained for blockchain ecosystem security and growth GANs can greatly enhance SC vulnerability detection We need research and development to overcome these obstacles and maximize their potential Preprocessing ensures data consistency for all SCs The GAN architecture consists of a generator and discriminator each implemented with three fully connected layers ReLU activation functions are applied to all hidden layers and a sigmoid function is used at the output layer of the discriminator The models were trained using the Adam optimizer with a learning rate of 0.0002 AST pathways are structured program syntax tree node associations They capture syntactic and structural interdependence by connecting code parts with nodes These routes describe code logic and flow to help machine learning models extract features neural network-based code embedding model that converts source code to vectors It generates distributed embeddings from AST routes to capture code semantics and syntax These embeddings map code snippets into a continuous vector space for pattern recognition the generator and discriminator will compete to create synthetic contracts that look more realistic and to separate actual contracts from fake ones on the processed training set the generator can distribute target data accurately while producing high-quality synthetic samples The trained generator will increase the size of the test set Identifying an integer overflow problem in a SC necessitates the initial vectorization of the source code The trained discriminator receives the vector to construct the security label we compare the contract vector with the expanded test set The detecting mechanism determines the contract’s susceptibility using the similarity threshold coefficient The contract is at risk when the similarity coefficient goes beyond the cutoff Preprocessing the source code: Preprocess source code and build vulnerability-specific contract vectors Model training: GAN generator and discriminator training The discriminator separates actual and synthetic contracts and the generator generates high-quality synthetic contracts Finding security holes: Using the discriminator you may determine the target contract’s vector Determine vector similarity and use the similarity threshold coefficient to check for vulnerability if the label is affirmative The generation of susceptible contract data involves code preprocessing, code embedding, and code generation, as shown in Fig. 1b. Source code for generating an Abstract Syntax Tree (AST) Preprocessing makes sure that the data for AST-based analysis is clean and standardized There is a possibility that private user and transaction information is included in the SCs source code Improper processing during model training could lead to a breach of data protection regulations Solidity’s SC programming language allows customising identifiers The naming conventions and programming techniques of programs and developers differ in their coding styles For GAN modelling and similarity judgement vectorisation of source code allows for translating semantically identical code segments into separate vectors This means that source code preprocessing is required Maintain integer overflow vulnerability features The following two aspects will be standardized The preprocessing logic is illustrated in the following pseudocode: Fundamental concepts of integer overflow with examples. The AST produced by the Solidity parser ANTLR has to be handled in line with the code2vec embedding criteria the traversable AST has to fit the following definition we are using various parameters like Non-Terminal (NT) One can depict the AST of an SC as \(\langle NT where \(NT\) represents non-terminal nodes and \(TN\) represents terminal nodes The function \(MNT: NT \rightarrow (NT \cup TN)^{*}\) maps non-terminal nodes to their respective child nodes, and the function \(MTN: TN \rightarrow A\) associates terminal nodes with values. All child node listings list each node once except the root. Figure 4 describes the AST smart contract solution. An AST path is a directed sequence of nodes that represents a syntactic relationship between two terminal elements in the AST of a program. It captures the structure and direction of traversal (upward or downward) between nodes and characterizes the relationships between code tokens. These paths form the building blocks for constructing code semantics in later stages. AST Paths: AST paths are defined as sequences of length \(L :\) where \(x_1, x_{L+1} \in TN\) (terminal nodes). Non-terminal nodes are \(x_j \in NT\) for \(j \in [2..L].\) The AST movement direction is represented by \(m_j \in \{ \uparrow , \downarrow \}\): - \(\uparrow\) (up) denotes that \(x_j\) is a child of \(x_{j+1}\) (rootward movement). - \(\downarrow\) (down) denotes that \(x_{j+1}\) is a child of \(x_j\) (going away from the root). The starting and ending nodes of a path \(p_T\) are \(S(p_T)\) and \(E(p_T)\). Define Path Context for AST path \(p_T\); the context is the triplet: The beginning node value is represented by \(A = MTN(S(PT)) .\) The terminal node value is represented by \(A_{TN} = MTN(E(PT)).\) code2vec is a neural embedding technique that converts source code into continuous vector representations by learning from syntactic paths (AST paths) and their contexts It enables learning about code structure and semantics in a way suitable for machine learning applications such as vulnerability detection CE represents the final code embedding vector for a given smart contract. Here n is the total number of AST paths extracted from the source code. Illustration of code embedding and synthetic code generation process Each \(PT_j\) is the jpath context in the Abstract Syntax Tree and is transformed into a path vector \(VR_j\) using a neural network function (NNF) Eq. (4) This function takes as input three components: \(S_j\) (the vector representation of the starting token) and \(Node_j\) (the sequence of node types along the path) The output \(VR_j\) captures the semantic and structural features of the code path \(AW_j\) is the attention weight assigned to the jth path It reflects the importance or relevance of that path in the overall context of the code Paths that contribute more to the code’s functional meaning are given higher weights during aggregation The final embedding vector CE is thus a weighted sum of all the individual path vectors where each path’s influence is modulated by its attention weight This method captures both local and global code semantics and enables the detection of subtle patterns related to vulnerabilities which we adapted to support Solidity syntax and distributions: DIS\(_data\)(r) and Dis(n) As generator and discriminator achieve Nash equilibrium The generator may produce realistic-looking Solidity code vectors after training we can augment the vulnerable dataset with numerous synthetic contract vectors Vector similarity detection will use the updated vulnerable contract dataset only vectors that have integer overflow vulnerabilities are used This means the trained discriminator can tell the difference between actual and fake contracts and identify those with integer overflow vulnerabilities Vector similarity analysis is a fundamental criteria for automated detection Contract vectors including structural and semantic information of the source code are produced using Code2vec while r and c represent b’s size and a’s dimensionality We have already addressed the procedures necessary to enhance the vulnerable contract dataset utilizing the GAN model and the methodology for converting the source code of SCs into vectors that exhibit structural and semantic attributes we showcased the process of evaluating SC integer overflow vulnerabilities using the GAN discriminator in conjunction with vector similarity This part presents the proposed method for finding SC integer overflow vulnerabilities at their paces we will provide the experimental setup and dataset to establish the appropriate vector similarity cutoff coefficient and compare our results to other tools The investigations used a Windows 10 PC with an Intel Core CPU (2.30 GHz) Code2vec (Version 2020 release) [23] derives feature vectors from SC source code whereas Solidity-parser-antlr (Version Version 0.4.11)[102] produces abstract syntax trees (ASTs) We ran two experiments to see if the suggested technique might find integer overflow issues in SC code This method averages cosine and Pearson correlation coefficients to determine vector similarity We tested numerous weights and thresholds to discover the best parameters for finding SC source code vulnerabilities and assessing recall and accuracy We built training and testing subsets from our core dataset (enhanced-smart-contracts-dataset.CSV) using open-source SCs with security classifications This enabled us to assess the effectiveness of vulnerability detection Fifty SCs with integer overflow flaws were incorporated into the training set for GAN models and vector similarity investigations There are 150 SCs in the testing set; 80 are secure The testing set compares our and other detection methods Table 3 offers an Enhanced Dataset Summary the analytical dataset for vector similarity identification comprising 50 genuine contracts in the training set we also produced 1,950 counterfeit contracts we discovered SC vulnerabilities following dataset generation The following ideas directed the selection of these two instruments: Many vulnerability detection programs use this tool to test their performance we test the detection model using Accuracy (ACC): we evaluated our approach against sFuzz and Oyente using the same test set of 150 contracts Our GAN-based method achieved an overall detection accuracy improvement of 12.4% over sFuzz and 18.1% over Oyente it demonstrated higher F1-score and precision values confirming its superior balance of sensitivity and specificity in detecting integer overflow vulnerabilities To validate the specific contribution of GAN-generated synthetic data to model performance we conducted an ablation study by training the system without synthetic contracts confirming the critical role of GAN-based data augmentation in addressing the data scarcity challenge These results empirically support the claim that the proposed method benefits from the synthetic vector generation process To offer a more reliable assessment beyond accuracy, we computed precision, recall, and F1-score for each threshold and weight configuration. The model achieved an F1-score of 0.91 when W = 0.74 and T = 0.9, indicating strong balance between sensitivity and specificity. Precision was 0.89 and recall was 0.94 in this optimal configuration. Semantics and code structure: cosine similarity is a good measure of these. Integer overflow vulnerabilities share the same semantics and code design. We can find these vulnerabilities by computing the cosine similarity between the codes. Resilience: Cosine similarity can withstand outliers and noise. Modifications to the code, such as comments, whitespace, etc., may be incorporated into practical applications. To some extent, cosine similarity can mask these distinctions, making the model more robust. To evaluate statistical significance, we performed repeated trials (n = 10) for each parameter setting. A paired t-test on detection accuracy across weight values showed statistically significant differences (p< 0.05), confirming that the chosen configuration (W = 0.74) improves detection performance in a meaningful way. Vector similarity threshold experiments we must identify the threshold T after determining W Model sensitivity to vector similarity depends on threshold T the model demands a higher degree of vector similarity when the threshold is raised False positives may grow with a low threshold and false negatives with a high one the threshold can have a negative impact on model performance if set too high or too low We also examined the potential risk of overfitting in the GAN-generated synthetic contracts Since these contracts are derived from a small training set there is a chance that the generator could produce overly similar instances we injected noise variability into the generator’s latent space and applied dropout regularization in the discriminator during training we plan to adopt adversarial validation techniques and external datasets to further test the robustness of the model against synthetic overfitting bias Figure 8 displays the lab’s concluding findings We attained detection accuracy and generalisability by calibrating the model threshold to 0.9 Our methodology preserves necessary structural and semantic information while converting SCs into small vector representations via a code embedding method This method improves the efficiency and potency of vulnerability identification This research has proved that a one-of-a-kind method for locating integer overflow vulnerabilities in SCs is both valuable and empirically validated This was accomplished through the use of code embedding and GANs It can be concluded that the strategy’s effectiveness has been demonstrated throughout this work The suggested way to get around the big problem of not having enough data in SC security research is by using GANs to make fake contract vector data that maintains real-world contracts’ structural and semantic properties The structure and semantic parts of real-world contracts are kept This indicates that the technique can help address the challenge of data scarcity in smart contract vulnerability detection By combining discriminator feedback with vector similarity analysis the proposed approach can uncover vulnerabilities even with limited training data While the results demonstrate promising accuracy further validation using additional tools such as Mythril and Slither will be necessary to comprehensively assess and benchmark the method’s performance our method improves detection accuracy by 12.4% over sFuzz and 18.1% over Oyente These gains are accompanied by stronger F1-score and precision values To validate the specific contribution of GAN-generated synthetic data we performed an ablation study comparing detection results with and without synthetic vectors The inclusion of synthetic data improved the F1-score from 0.84 to 0.91 demonstrating the effectiveness of GANs in mitigating data scarcity this method offers a valuable substitute to improve the security of SCs and lower the risk of significant financial losses The datasets generated and/or analyzed during the current study are provided as supplementary files with this article (enhanced-smart-contracts-dataset.csv) Rachad, A. et al. A smart contract architecture framework for insurance industry using blockchain and business process management technology. IEEE Eng. Manage. Rev. 52, 55–68. https://doi.org/10.1109/EMR.2023.3348431 (2023) Praitheeshan, P., Pan, L., Yu, J., Liu, J., & Doss, R. Security analysis methods on ethereum smart contract vulnerabilities: a survey. arXiv:1908.08605 (2019) Mazhar, T. et al. Generative AI, IoT, and blockchain in healthcare: application, issues, and solutions. Discov. Internet Things 5, 5. https://doi.org/10.1007/s43926-025-00095-8 (2025) Eth2Vec: learning contract-wide code representations for vulnerability detection on ethereum smart contracts Hu, C. et al. Smart contract assisted privacy-preserving data aggregation and management scheme for smart grid. IEEE Trans. Depend. Secure Comput. 21(4), 2145–2161. https://doi.org/10.1109/TDSC.2023.3300749 (2023) Li, C. et al. Smart contract-based decentralized data sharing and content delivery for intelligent connected vehicles in edge computing. IEEE Trans. Intell. Transp. Syst. 25, 14535–14545. https://doi.org/10.1109/TITS.2024.3388422 (2024) Yao, P. et al. Security-enhanced operational architecture for decentralized industrial internet of things: a blockchain-based approach. IEEE Internet Things J. 11, 11073–11086. https://doi.org/10.1109/JIOT.2023.3329352 (2023) Das, D. et al. A secure blockchain enabled v2v communication system using smart contracts’’. IEEE Trans. Intell. Transp. Syst. 24, 4651–4660. https://doi.org/10.1109/TITS.2022.3226626 (2022) Ngo, D. M. et al. A scalable security approach in IoT networks: smart contracts and anomaly-based IDS for gateways using hardware accelerators. IEEE Access 12, 145. https://doi.org/10.1109/ACCESS.2024.3486605 (2024) An efficient smart contract vulnerability detector based on semantic contract graphs using approximate graph matching Code2vec: learning distributed representations of code In Proceedings of the ACM on Programming Languages Robust vulnerability detection in solidity-based ethereum smart contracts using fine-tuned transformer encoder models Jiang, F. et al. Enhancing smart-contract security through machine learning: a survey of approaches and techniques. Electronics 12, 2046. https://doi.org/10.3390/electronics12092046 (2023) Shyamasundar, R.K. et al. Enhancing robustness of smart contracts through declarations. In ICBC 2024. Lecture Notes in Computer Science, vol. 15425 (Springer, 2025). https://doi.org/10.1007/978-3-031-77095-1-4 A New Approach to Prevent Reentrant Attack in Solidity Smart Contracts Blockchain Technology and Application 83–103 (Springer Bond, F. A solidity parser for JS built on top of a robust ANTLR4 grammar. https://github.com/solidity-parser/parser (2019) A Dynamic Scalable Blockchain Based Communication Architecture for IoT Smart Blockchain 159–166 (Springer Hwang, S.-J. et al. CGGNet: compiler-guided generation network for smart contract data augmentation. IEEE Access 12, 56. https://doi.org/10.1109/ACCESS.2024.3427829 (2024) A novel neural source code representation based on abstract syntax tree In 2019 IEEE/ACM 41st International Conference on Software Engineering (ICSE) 783–794 (IEEE The role of blockchain to secure the internet of medical things Wan, J. et al. A blockchain-based solution for enhancing security and privacy in smart factory. IEEE Trans. Ind. Inf. 15, 3652–3660. https://doi.org/10.1109/TII.2019.2894573 (2019) Hasan, H. R. et al. Blockchain-enabled telehealth services using smart contracts. IEEE Access 9, 56. https://doi.org/10.1109/ACCESS.2021.3126025 (2021) Haritha, T. & Anitha, A. Multi-level security in healthcare by integrating lattice-based access control and blockchain-based smart contracts system. IEEE Access 11, 114322–114340. https://doi.org/10.1109/ACCESS.2023.3324740 (2023) Krichen, M. Strengthening the security of smart contracts through the power of artificial intelligence. Computers 12, 107. https://doi.org/10.3390/computers12050107 (2023) of the 2016 ACM SIGSAC Conference on Computer and Communications Security 254–269 (2016) Ayub, M. F. et al. Secure consumer-centric demand response management in resilient smart grid as industry 5.0 application with blockchain-based authentication. IEEE Trans. Consumer Elect. 70, 145. https://doi.org/10.1109/TCE.2023.3320974 (2023) Securify: practical security analysis of smart contracts In ACM SIGSAC Conference on Computer and Communications Security (CCS) 2018 Measuring the control flow immutability of all smart contracts deployed on ethereum Cryptocurrencies and Blockchain Tech.: ESORICS 357–372 (Springer ContractCheck: checking ethereum smart contracts in fine-grained level PaVM: a parallel virtual machine for smart contract execution and validation Sfuzz: an efficient adaptive fuzzer for solidity smart contracts ACM/IEEE 42nd International Conference on Software Eng.) 778–788 (2020) Download references Department of Computer Science and Engineering International Institute of Technology and Management (IITG) Author 1(Dileep Kumar Murala): Conceptualization Author 3 (K Vara Prasada Rao): Literature review The authors declare no competing interests Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations Download citation DOI: https://doi.org/10.1038/s41598-025-99267-3 Anyone you share the following link with will be able to read this content: a shareable link is not currently available for this article Sign up for the Nature Briefing newsletter — what matters in science All News Infineon and Marelli enter new era of automotive cockpit design with MEMS laser beam scanning at 2025 Auto Shanghai Trump’s Trade Bombshell: Tariffs on China Hit 245% India Aims to Capture 10% of Global Chip Demand by 2030 Cost of Semiconductor Chips per Vehicle to Double by 2030: NITI Aayog Infineon expands its GaN power portfolio with EasyPACK CoolGaN power modules for high-voltage applications Rohde & Schwarz pioneers the future of automotive Ethernet using Analog Devices’ 10BASE-T1S solutions Keysight Awarded NATO FORACS Contract to Enhance Operational Readiness Power and Thermal Management Concerns in AI: Challenges and Solutions Vishay Intertechnology 1 Form A Solid-State Relays Offer Continuous Load Current to 5 A in Compact SOP-4 Package Rohde & Schwarz presents its advanced solutions for power electronics testing and characterization at PCIM Expo 2025 Vishay Intertechnology Thick Film Power Resistor With Optional NTC Thermistor and PC-TIM Simplifies Designs and Saves Board Space Security by Design in Electronics: A Proactive Approach to Cybersecurity Keysight Introduces Comprehensive LPDDR6 Solution for End-to-End Memory Design and Test Workflows System-level considerations for EVSE design Keysight Introduces System Designer for PCIe and Chiplet PHY Designer for Digital Standards-Driven Simulation Workflows TI pioneers new magnetic packaging technology for power modules, cutting power solution size in half The 3D Printing Industry in India: Growth, Opportunities, and Challenges Top 10 3D Printing Companies in India Optimize MES intelligence in semiconductor advanced packaging Amtech Shares Key Trends Shaping the Future of Electronics Manufacturing in 2025 Exporting Power: India’s Growing Influence In Global Transformer Supply Chains Engineering the Future of Compact Audio: A Deep Dive into the NAU82110YG Filter-Free Class-D Amplifier Unlocking the Potential of 6G FR3 ST’s Automotive MCU technology for next-generation vehicles Steel Soldering: Definition, Process, Working, Uses & Advantages Understanding Metal Soldering: Definition, Process, Working, Uses & Advantages Stealth Technology: Definition, Types, Working & Applications Top 10 eSIM Manufacturers in India Powering a Sustainable Future: How STMicroelectronics is Leading by Design Rewiring the Future of India’s Power Grid with Wirepas Pushing the Boundaries of Miniaturization with Texas Instruments’ New MCU Delta Electronics Fuels India’s Digital Ambitions with Scalable, Sustainable ICT Solutions the increasing adoption of electric vehicles and the ongoing trends in global digitalization and reindustrialization global electricity demand is expected to surge Infineon Technologies AG is introducing the EasyPACK CoolGaN Transistor 650 V module the module has been specifically developed for high-power applications such as data centers It is designed to meet the growing demand for higher performance while providing maximum ease of use helping customers accelerate their design processes “The CoolGaN based EasyPACK power modules combine Infineon’s expertise in power semiconductors and power modules,” says Roland Ott Senior Vice President and Head of the Green Energy Modules and Systems Business Unit at Infineon “This combination offers customers a solution that meets the increasing demand for high-performance and energy-efficient technologies in applications such as data centers The EasyPACK CoolGaN module integrates 650 V CoolGaN power semiconductors with low parasitic inductances achieved through compact die packing enabling fast and efficient switching Delivering up to 70 kW per phase with just a single module the design supports compact and scalable high-power systems by combining Infineon’s .XT interconnect technology with CoolGaN options the module enhances both performance and reliability The .XT technology is implemented on a high-performance substrate which in turn translates to higher system efficiency and lower cooling demands This results in increased power density and excellent cycling robustness With support for a broad range of topologies and customization options the EasyPACK CoolGaN module addresses diverse requirements in industrial and energy applications ELE Times provides a comprehensive global coverage of Electronics In addition to providing in depth articles Navitas Semiconductor is aiming at motors up to 600W with a GaN motor drive IC “Specifically designed for motor drive applications this integrated solution combines two GaN FETs in a half-bridge configuration with drive sensing and autonomous protection,” according to the company it is a 650V part with two 120mΩ GaN HEMTs in an 8 x 10mm PQFN 8×10 package Bi-directional ‘loss-less’ current sensing is included to measure both forward current and recirculating currents between switching coil phases – an autonomous freewheeling function switches on the appropriate GaN transistor when reverse current is detected to reduce conduction losses Turn-on and turn-off slew rates are adjustable to trade EMI performance for efficiency Protections include high-side short-circuit The part is actually listed on the company selection table as a 700V part with an 800V dynamic capability No other information is available at the time of writing Two lower power devices were announced at the same time: the 6 x 8mm 2x 170mΩ NV6257 and the 8 x 10mm 2x 170mΩ NV6287 as is use in domestic white goods and even hair dryers The devices will be on display at PCIM in Nuremberg: Hall 9 Only three weeks ago, Infineon listed motor drives among application when it announced 100V GaN transistors with integrated Schottky diodes Tagged with: gallium nitride GaN HEMT Δdocument.getElementById( "ak_js_1" ).setAttribute( "value" Electronics Weekly is excited to launch it's own podcast brought to you by our editor Caroline Hayes You can hear interesting insights from a range of industry figures Catch all the episodes » Sign up for the Electronics Weekly newsletters: Mannerisms Gadget Master and the Daily and Weekly roundups Find out more » Read our special supplement celebrating 60 years of Electronics Weekly and looking ahead to the future of the industry Read the first ever Electronics Weekly online: 7th September 1960 We've scanned the very first edition so you can enjoy it Read the very first edition » Keep up with developments relating to space technology - satellite technology View our busy aerospace section » Keep up with developments relating to the Internet of Things (IoT) - Industrial IoT View our popular Internet of Things section » Keep up with developments relating to Power electronics - MOSFETs View our busy Power section » While it won’t please gamers or power users the Belkin 6-in-1 GaN USB-C dock is ideal for workplaces and home offices this neat little dock is a great way to tidy up your workstation The Belkin 6-in-1 USB-C dock is designed for home office and workplaces providing an all-in-one solution for charging which means it uses Gallium Nitride chips instead of silicon Gallium Nitride allows more efficient power transfer than silicon which enables manufacturers to produce high powered charging devices with smaller form factors than equivalent silicone devices making it ideal for busy desks or minimalist setups Headline specs are relatively unimpressive it’s a good little dock that I’ve found ideal for decluttering my desk and hooking up my laptop when working from home Find out in my full Belkin USB-C 6-in-1 Core GaN Dock $139 /£129 Thanks to its Gallium Nitride power supply the Belkin 6-in-1 delivers strong charging performance in a very compact body making it perfect for tidying up busy desks while extending your number of ports The Belkin 6-in-1 dock uses a Gallium Nitride (GaN) power supply Gallium nitride transfers power more efficiently than silicon allowing manufacturers to shrink the size of power supply units versus silicon units while delivering an equivalent power delivery rate This means that the 6-in-1 dock is pleasantly compact, at just over 3-inches square. It’s taller than the dock I use for my photography workstation, the Belkin Connect 11-in-1 dock but the 6-in-1’s footprint is much smaller It’s quite heavy — surprisingly so given its diminutive size — but I was nevertheless able to mount it on the side of my laptop stand using some 3M double-sided tape I’ll cover the ports in more detail later but this has effectively extended my ports when using my work laptop by 1x USB-C and 2x USB-A. and its battery is not long left for this world It’s been useful having such a powerful charging dock to keep the battery juiced up and prevent my laptop from cutting out The 6-in-1 dock’s 130W power supply offers 96W pass-through to a laptop via its side USB-C host port which is more than enough for most laptops The frontal USB-C port and USB-A ports offer 7.5W each but is fine to keep phones and small peripherals charged during or between uses the Belkin 6-in-1 GaN dock offers decent enough transfer speeds for general productivity and office use At the rear of the dock sits a 1 gigabyte per second (Gbps) Ethernet port for relatively quick internet connectivity This is as much as the Anker 675 USB-C dock though — for example if you’re uploading and downloading large files like videos to/from the cloud regularly — the Belkin 11-in-1 USB-C offers 2.5 Gbps speeds 5 GBps transfer speeds through the USB ports are relatively good for general workplace or home office use This will allow you to transfer everyday files between devices quickly The Belkin 6-in-1 isn’t designed (or indeed priced) for hardcore creatives the Belkin 6-in-1 can support a single 4K UHD monitor but it does mean you can continue to use your beautiful 4K monitor for work by simply switching over inputs The Belkin USB-C 6-in-1 Core GaN dock is designed primarily with compactness and general productivity in mind which may prove frustrating even to home office users I’ve touched on a few issues already in the ‘Ups’ section above but so far these have just been buying considerations to tidy up the desks of general productivity users in the workplace or in a home office So it kinda figures that it doesn’t have 80Gbps transfer speeds 2.5Gbps Ethernet or +60Hz refresh rate support Given one of my laptops’s USB-C ports is going to be taken up by either way for charging either via its adapter or the dock’s host cable but it isn’t extending my USB-C ports otherwise so I’ve found it really limiting having only a single such port on the dock but I’ve genuinely had to root around in my cable bag to find compatible USB-C cables at times I know plenty of people who run multiple monitor setups for work so it would’ve been nice to see a second display hookup such as an SD/MicroSD reader or headphone jack The Belkin 11-in-1 USB-C dock offers all of the above a VGA port (for older monitors) and two USB-C ports (one PD charging only The Belkin USB-C 6-in-1 Core GaN dock isn’t a bad product It primarily serves to clean up cluttered desks by utilizing GaN technology to offer powerful 96W charging in a compact form Its data transfer and Ethernet speeds are fine making it a decent product for single-monitor workplace or home office setups It’s really difficult to recommend it over the Belkin 11-in-1 USB-C dock The 11-in-1 dock offers more of everything: faster transfer and Ethernet speeds more expansion ports and multiple monitor hookup As long as you have the extra $60 to spare if all you want is a simple dock for basic productivity needs you can do a lot worse than the Belkin 6-in-1 Reviews & CamerasPeter is a Senior Editor at Tom's Guide heading up the site's Reviews team and Cameras section specialising in architectural and portrait photography When he's not snapping away on his beloved Fujifilm camera he can usually be found telling everyone about his greyhounds squeezing as many FPS as possible out of PC games you will then be prompted to enter your display name Funding moves GF closer to large-scale production of next-generation gallium nitride chips for a range of RF and high-power applications  2024 — GlobalFoundries (Nasdaq: GFS) (GF) has received an additional $9.5 million in federal funding from the U.S government to advance the manufacturing of GF’s essential gallium nitride (GaN) on silicon semiconductors at its facility in Essex Junction The funding moves GF closer to large-scale production of GaN chips With the ability to handle high voltages and temperatures GaN chip technology is essential for enabling higher performance and greater energy efficiency across a range of RF and high-power control applications including automobiles equipment and prototyping capabilities to its market-leading GaN IP portfolio and reliability testing as the company moves closer to full-scale manufacturing of its 200mm GaN chips in Vermont GF is committed to creating a fast and efficient path for customers to realize new innovative designs and products that leverage the unique efficiency and power management benefits of GaN chip technology.  “GF is proud of its leadership in GaN chip technology which is positioned to make game-changing advances across multiple end-markets and enable new generations of devices with more energy-efficient RF performance and faster-charging longer-lasting batteries,” said Nicholas Sergeant vice president of IoT and aerospace and defense at GF government’s partnership and ongoing support of our GaN program Realizing full-scale GaN chip manufacturing will be a catalyst for innovation for both our commercial and government partners and will add resilience and strengthen the semiconductor supply chain.”  Department of Defense’s Trusted Access Program Office (TAPO) represents the latest federal investment to support GF’s GaN program in Vermont.   “This strategic investment in critical technologies strengthens our domestic ecosystem and national security and ensures these assets are readily available and secure for DoD utilization empowering resilience and responsiveness,” said Dr Director at Defense Microelectronics Activity.  GF has received more than $80 million since 2020 from the U.S development and advancements to pave the way to full-scale GaN chip manufacturing.  Vermont is a U.S.-accredited Trusted Foundry and the global hub of GF’s GaN program with longstanding leadership in 200mm semiconductor manufacturing GF acquired Tagore Technology’s Gallium Nitride Power portfolio and created the GF Kolkata Power Center in Kolkata The center is closely aligned with and supports GF’s facility in Vermont development and leadership in GaN chip manufacturing.  the GF logos and other GF marks are trademarks of GlobalFoundries Inc All other trademarks are the property of their respective owners.  This news release may contain forward-looking statements Readers are cautioned not to place undue reliance on any of these forward-looking statements These forward-looking statements speak only as of the date hereof GF undertakes no obligation to update any of these forward-looking statements to reflect events or circumstances after the date of this news release or to reflect actual outcomes Michael Mullaney [email protected]  Join GF and help us shape what’s essential GlobalFoundries (GF) is a leading manufacturer of essential semiconductors the world relies on to live We innovate and partner with customers to deliver more power-efficient high-performance products for high-growth markets With our global manufacturing footprint spanning the U.S. we are a trusted and reliable source for customers around the world '+n.escapeExpression("function"==typeof(o=null!=(o=r(e,"eyebrowText")||(null!=l?r(l,"eyebrowText"):l))?o:n.hooks.helperMissing)?o.call(null!=l?l:n.nullContext||{},{name:"eyebrowText",hash:{},data:t,loc:{start:{line:28,column:63},end:{line:28,column:78}}}):o)+" \n '+(null!=(o=c(e,"if").call(r,null!=l?c(l,"cta2PreText"):l,{name:"if",hash:{},fn:n.program(32,t,0),inverse:n.noop,data:t,loc:{start:{line:63,column:20},end:{line:63,column:61}}}))?o:"")+"\n"+(null!=(o=(c(e,"ifAll")||l&&c(l,"ifAll")||n.hooks.helperMissing).call(r,null!=l?c(l,"cta2Text"):l,null!=l?c(l,"cta2Link"):l,{name:"ifAll",hash:{},fn:n.program(34,t,0),inverse:n.noop,data:t,loc:{start:{line:64,column:20},end:{line:70,column:30}}}))?o:"")+" (CR Micro) reported steady gains in its 2024 annual and 2025 first-quarter results underscoring resilience in a volatile chip market Full-year revenue rose 2.2% to CNY10.119 billion (US$1.39 billion) Some subscribers prefer to save their log-in information so they do not have to enter their User ID and Password each time they visit the site check the 'Save my User ID and Password' box in the log-in section This will save the password on the computer you're using to access the site Convolutional neural networks (CNNs) and generative adversarial networks (GANs) are examples of neural networks -- a type of deep learning algorithm modeled after how the human brain works CNNs, one of the oldest and most popular of the deep learning models, were introduced in the 1980s and are often used in visual recognition tasks. GANs are relatively newer. Introduced in 2014, GANs were one of the first deep learning models used for generative AI CNNs are sometimes used within GANs to generate and discern visual and audio content "GANs are essentially pairs of CNNs hooked together in an 'adversarial' way so the difference is one of approach to output or insight creation albeit there exists an inherent underlying similarity," said computer pioneer John Blankenbaker retired principal data scientist at global management consulting firm SSA CNNs might try to determine if a picture contains a cat -- a recognition task -- while GANs will try to make a picture of a cat the networks are building up a representation of what makes a picture of a cat distinctive a professor at New York University and chief AI scientist at Meta invented CNNs in the 1980s when he was a researcher at the University of Toronto His aim was to improve the tools for recognizing handwritten digits by using neural networks Although his work on optical character recognition was seminal it stalled due to limited training data sets and computing power Interest in the technique exploded after 2010 following the introduction of ImageNet -- a large labeled database of images -- and the launch of its annual ImageNet Large Scale Visual Recognition Challenge (ILSVRC) One of the most promising entries in the inaugural year of the competition was the AlexNet model based on CNNs Its success demonstrated that CNNs could efficiently scale to achieve good performance on even the largest image databases "CNNs are designed to use data with spatial structure such as images or video," said Donncha Carroll a partner at Lotis Blue Consulting who leads the firm's Data Science Center of Excellence The convolutional neural network is composed of filters that move across the data and produce an output at every position a convolutional neural network designed to recognize animals in an image would activate when it recognizes legs It's also important to note that CNNs are designed to recognize the lines edges and textures in patterns near each other which means that we are processing something where the idea of neighborhood is important -- such as pixels around a given pixel or signal values slightly before and after a given moment." GANs were invented by American computer scientist Ian Goodfellow currently a research scientist at DeepMind when he was working at Google Brain from 2014 to 2016 are a type of deep learning model used to generate images of numbers and realistic-looking faces The field exploded once researchers discovered it could be applied to synthesizing voices GANs and their variations were heralded by CNN inventor LeCun as the most interesting idea of the last 10 years in machine learning The term adversarial comes from the two competing networks creating and discerning content -- a generator network and a discriminator network the generator network creates new images that look like faces the discriminator network tries to tell the difference between authentic and generated images The discriminator performance data then helps to train the overall system These other models might improve results in areas outside image generation RNNs might be a better fit for generating realistic speech or music and VAEs for synthetic streaming sensor data One important distinction between CNNs and GANs is that the generator in GANs reverses the convolution process "Convolution extracts features from images while deconvolution expands images from features." Here is a rundown of the chief differences between CNNs and GANs and their respective use cases Although GANs are getting a lot of the attention lately CNNs continue to be used under the hood -- that is within GANs for generating and discerning authenticity a supply chain planning and optimization firm considers the two neural networks to be complementary in terms of function "Since CNNs are so effective at image processing both the generator and discriminator networks are by default CNNs," he said It is important to note that CNNs and GANs only tend to be combined in one way "GANs typically work with image data and can use CNNs as the discriminator But this doesn't work the other way around meaning a CNN cannot use a GAN," Mead said Blankenbaker cautions against getting caught up in the latest model rather than focusing on specific goals and the underlying data "We see too many companies getting excited about the buzzwords and trying to fit a square peg into a round hole resulting in overspending on overkill solutions," Blakenbaker said "One of the biggest challenges is always the data quality itself for training the models especially when we're talking about business-specific solutions instead of something as generic as a cat," he said Editor's note: This article was updated to include additional information about the specific roles of CNNs and other generative AI models in GANs George Lawton is a journalist based in London he has written more than 3,000 stories about computers How to detect AI-generated content What is a neural radiance field (NeRF)? What is synthetic data? GAN vs. transformer models: Comparing architectures and uses Assessing different types of generative AI applications As part of an effort to enable its employees to make data-informed decisions Virgin Media O2 is using the GenAI-powered insight .. The longtime BI vendor is the latest to add generative AI-powered natural language query capabilities aimed at enabling more .. The new platform features agents addressing needs such as data preparation and natural language-based analysis Organizations must take steps to ensure compliance with emerging rules targeting foreign adversaries' access to U.S A former federal deputy CIO testified that reauthorizing the Technology Modernization Fund will help the government continue IT .. Providing detailed background information is one way to help data privacy regulators during investigations of potential .. Data cleansing corrects errors to improve data quality while data transformation changes data format and structure to support .. The data integration specialist's purchase provides users with complementary capabilities that enable them to send transformed .. The database vendor plans to add support for Model Context Protocol and automated vector embedding capabilities to make it faster.. Many companies are already carrying out demand planning and predictive analytics can potentially help improve the results of .. RFID tags can help improve inventory management Problems with suppliers can affect a company's supply chain and Metrics details The development of photoelectrochemical cells for reduction of nitrate to ammonia under solar light is of significant interest for the production of clean chemicals and fuels but has remained a daunting challenge we investigate various metal catalysts supported on GaN nanowires grown on n+-p Si wafer – an emerging functional platform for scalable artificial photosynthesis – and demonstrate highly stable and efficient photoelectrochemical nitrate reduction reaction We find that Co and Ni catalysts on GaN/Si exhibit the best performance with an onset potential >0.3 VRHE and a faradaic efficiency of NH3 of 99% at 0.2 VRHE These results highlight the advantage of photoelectrochemical system in achieving efficient nitrate reduction under more positive potentials In-situ measurements and theoretical calculations reveal that the binding modes of the \({{{\rm{NO}}}}_{2}^{{-}}\) intermediate play a key role in the NH3 synthetic process These results demonstrate that the rational design of catalysts on photoelectrodes can construct synergistic metal-semiconductor interactions for efficient and stable photoelectrochemical NH3 synthesis This synergistic relay catalysis or intermediate spillover improved the performance compared to their single phases respectively) and low onset potentials (0.62 VRHE and 0.43 VRHE) highlighting their ability to optimize reaction kinetics and improve charge transfer efficiency These studies illustrate the importance of catalyst design in achieving high performance in PEC \({{{\rm{NO}}}}_{3}^{{-}}\) RR systems This suggests that GaN NWs can be an ideal supporting materials to promote the proton-coupled electron transfer reaction including \({{{\rm{NO}}}}_{3}^{{-}}\) reduction to NH3 If the integration of metal catalyst on the semiconductor nanostructures can demonstrate synergetic effects by reducing the thermodynamic energy barriers controlling the binding modes of adsorbates and providing a sufficient number of protons with effective transfer of photogenerated electrons the photoelectrodes will further reduce the overpotential and enhance the productivity of NH3 compared to electrocatalysts rational design of photoelectrodes by selecting a suitable catalyst on the semiconductor surface and the understanding of reaction pathways are urgently required to achieve more efficient and productive solar-driven \({{{\rm{NO}}}}_{3}^{{-}}\) conversion to NH3 we investigated the metal catalysts supported on GaN NWs vertically grown on n+-p Si photoelectrodes for PEC \({{{\rm{NO}}}}_{3}^{{-}}\) RR to NH3 to demonstrate the synergistic effects of metal catalysts and GaN NWs We first screened various noble metals (Pt Co-loaded GaN/Si (Co/GaN/Si) showed the highest faradaic efficiency of NH3 (FENH3 > 90%) at the widest potential window (0.2 to −0.7 VRHE) Ni/GaN/Si achieved the NH3 production rate of 83.3 μmol/h/cm2 and the maximum value of 201.6 μmol/h/cm2 at potentials of 0 and −0.4 VRHE We characterized the performance of photoelectrodes based on the presence of GaN NWs and found that the catalysts loaded on GaN NWs showed much lower onset potential and higher productivity of NH3 than those on planar n+-p Si photoelectrodes This is due to the synergetic metal-GaN interaction which enhances water dissociation and catalytic activity of nitrogen-containing intermediates providing a sufficient number of protons near the surface which in turn activates \({{{\rm{NO}}}}_{3}^{{-}}\) and facilitates its efficient conversion to \({{{\rm{NO}}}}_{2}^{{-}}\) following \({{{\rm{NO}}}}_{2}^{{-}}\) to NH3 conversion is further promoted by catalysts A series of in-situ experiments and theoretical calculations were performed to interpret reaction mechanisms and revealed that the binding modes of the NO2− intermediate (*ONO−* or *NO2−) on the metal catalysts were the key to the further deoxygenation process This work provides insights into the design of efficient photoelectrode for green NH3 production a 45 degree-tilt-view SEM image of Co/GaN/Si STEM-HADDF images and EDS elemental maps of b Co/GaN and c Ni/GaN Ga (blue) and N (cyan) were uniformly detected on GaN NWs while Co (yellow) O (green) elements were segregated on the surface of GaN NWs since the catalysts were deposited under aqueous conditions and exposed to ambient air prior to the analysis a Energy band diagram of metal/GaN/Si photoelectrode under light illumination b Faradaic efficiency of NH3 for metal catalysts loaded on GaN/Si The measurement was conducted in 0.1 M K2SO4 with 0.5 M KNO3 at various cathodic potentials (0.3 to −0.8 VRHE) under 1 sun light illumination Co and Ni catalysts showed the highest FENH3 (>98%) among the tested samples Within the precursor concentration range in this work there was no significant difference in catalytic performance indicating that the determining factor for catalyst characteristics is the type of metal catalyst rather than the loading amount of metal catalysts a LSV curves and b Nyquist impedance plots of Si and Ni/GaN/Si measured in 0.1 M K2SO4 with 0.5 M KNO3 under 1 sun light illumination c Faradaic efficiency and d production rate of NH3 plotted as a function of cathodic potentials Co/GaN/Si exhibited the widest potential window (−0.2 to −0.7 VRHE) for high FENH3 > 90% and Ni/GaN/Si revealed the best YNH3 = 201.6 μmol/cm2/h at −0.4 VRHE The results revealed that Co/GaN/Si exhibited superior charge retention and reduced recombination compared to GaN/Si suggesting the critical role of Co catalyst in improving charge transfer efficiency and sustaining PEC activity during \({{{\rm{NO}}}}_{3}^{{-}}\)RR the KIE value for Co/GaN/Si significantly dropped to 1.15 Once protons became readily available at the surface through water dissociation by GaN NWs it appears that Co catalysts efficiently reduce \({{{\rm{NO}}}}_{3}^{{-}}\) to NH3 through deoxygenation and hydrogenation processes due to the intrinsic selectivity for producing NH3 rather than hydrogen evolution using GaN nanowires as a support for metal catalysts presents a promising approach to demonstrating the synergy between proton supply and the catalysis of nitrogen-containing intermediates and YNH3 of Co or Ni catalysts on GaN NWs were comparable to those of recent photoelectrodes and superior to state-of-the-art electrocatalysts for \({{{\rm{NO}}}}_{3}^{{-}}\)RR (Table S1) where the absence of the \({{{\rm{NO}}}}_{2}^{{-}}\) band highlighted the catalytic role of Co catalyst in facilitating the PEC \({{{\rm{NO}}}}_{3}^{{-}}\)RR to NH3 Photocurrent density of a Co/GaN/Si and b Ni/GaN/Si in 0.1 M K2SO4 with 0.5 M KNO3 under 1 sun light illumination Measurements were performed 5 times at 0 VRHE for 1 h each and then 10 times at −0.6 VRHE for 30 min each Insets show SEM images of photoelectrodes after 10 h reaction FENH3 and YNH3 for c Co/GaN/Si and d Ni/GaN/Si a Full mechanism of \({{{\rm{NO}}}}_{3}^{{-}}\) reduction on the catalyst surfaces including reduction to NH3 and incomplete reduction to \({{{\rm{NO}}}}_{2}^{{-}}\) and subsequent desorption The key intermediate and its two binding modes determining the outcome are encircled b Reduction from \({{{\rm{NO}}}}_{3}^{{-}}\) to \({{{\rm{NO}}}}_{2}^{{-}}\) and subsequent desorption on Cu catalyst c Thermodynamics of desorption and further reduction through change of binding mode on Cu catalyst a Full mechanism of \({{{\rm{NO}}}}_{3}^{{-}}\) reduction on Co3O4 Defect formation leads to N-O bond dissociation b Free energy of the reduction from \({{{\rm{NO}}}}_{3}^{{-}}\) to NH3 on Co3O4 a Mechanism of \({{{\rm{NO}}}}_{3}^{{-}}\) reduction to NH3 on Ni(OH)2 Defect formation allows for attachment of nitrate further reduction regenerates the Ni(OH)2 surface and dissociates the N-O bond The mechanism proceeds similarly with HNO2 and NO b Free energy of the \({{{\rm{NO}}}}_{3}^{{-}}\) reduction on Ni(OH)2 we have investigated various metal catalysts for the PEC \({{{\rm{NO}}}}_{3}^{{-}}\)RR to NH3 we found that first-row earth-abundant transition metals Co and Ni catalysts outperformed noble metals (Pt In-situ IR measurement showed that a large number of \({{{\rm{NO}}}}_{2}^{{-}}\) intermediates were desorbed from Cu catalysts but further reduced to NH3 on Co catalysts Theoretical calculations confirmed that the deoxygenation process of the *NO2− intermediate is feasible on Co3O4 and Ni(OH)2 catalysts via reaction with oxygen defects \({{{\rm{NO}}}}_{2}^{{-}}\) molecules desorbed from the Cu surface due to the *ONO−* binding mode with a lower barrier for desorption than for switching to the *NO2− mode that is needed for further deoxygenation processes This leads to incomplete reduction of nitrate to nitrite on Cu while the Co and Ni containing catalysts are capable of full reduction of \({{{\rm{NO}}}}_{3}^{{-}}\) to NH3 Owing to the strong catalytic activities of Co and Ni and the synergistic effects of metal-GaN interactions including a low onset potential Vonset = 0.3 VRHE high faradaic efficiency of NH3 (~99% at 0.2 VRHE) and high production rate of NH3 (201.6 μmol/h/cm2 at −0.4 VRHE) the photoelectrodes were stable and performed for 10 h of reaction without any noticeable degradation which consist of the two most popular semiconductors (Si and GaN) and inexpensive transition metals (Co and Ni) have the potential to be used in industry-scale conversion of \({{{\rm{NO}}}}_{3}^{{-}}\) pollutant to value-added NH3 under solar light Starting p-type Si (100) wafer was purchased from Purewafer and has a resistivity of 1–10 Ω·cm The n+-p Si was demonstrated by spin-coating of phosphorus dopant on the front side of the polished p-type Si wafer and then thermal annealing at 950 °C under nitrogen atmosphere for 4 h N-type GaN NWs were grown on n+-p Si wafer by plasma-assisted molecular-beam epitaxy under nitrogen-rich condition to promote the formation of an N-terminated surface The growth temperature was 790 °C and the duration was ~2 h The forward plasma power was 350 W with Ga flux beam equivalent pressure of 5 × 10−8 Torr Metal catalysts were loaded on GaN/Si by a photodeposition method GaN/Si wafer on a Teflon holder was placed in the bottom of a quartz reactor containing 66 mL of 20 vol% methanol aqueous solution 97%) solutions were added into the chamber to deposit Pt The reactor was evacuated for 5 min using a rotary pump PE300BUV) was used as a light source which irradiated the chamber for 30 min When light illuminated the GaN/Si wafer in the precursor solution electrons are photoexcited to the conduction band creating holes in the valence band of the GaN NWs The photoexcited electrons then migrated to the GaN surface participating in the reduction of metal ions photogenerated holes in the valence band migrated to the surface oxidizing the methanol sacrificial agent in the solution Since the bandgap of GaN (~3.4 eV) was suitable to absorb ultraviolet light and drive the redox reactions there was no need for an external bias in the photodeposition process the samples were rinsed with deionized water and dried by air blowing All photoelectrochemical measurements were performed in an H-type cell separated by a Nafion membrane with a three-electrode system using a potentiostat (Bio Logic SAS SP-200) Ag/AgCl filled with 3 M KCl was used for the reference electrode and Pt wire was used for the counter electrode GaN/Si as well as other metal-loaded photoelectrodes were used as the working electrodes The measured potentials (VAg/AgCl) (V) were converted to the reversible hydrogen electrode (VRHE) (V) by using the Nernst function: VRHE = VAg/AgCl + 0.197 + 0.0591 × pH The electrolytes were aqueous solutions of 0.1 M K2SO4 (Sigma-Aldrich ≥99%) or 0.1 K2SO4 with 0.5 M KNO3 (Sigma-Aldrich ≥99%) prepared by dissolving the solid salts in deionized water pH values of electrolytes were measured using a pH meter (Mettler Toledo) 8 ml of electrolyte was filled in the cathodic compartment of H-type cell and purged with Ar for 30 min before the measurement The light source for the photoelectrochemical reaction was LCS-100 (ORIEL) and the intensity with AM 1.5G filter was calibrated by adjusting the distance from the sample to the light source Linear sweep voltamnetry curves were obtained at a rate of 10 mV/s Chronoamerometic curves were recorded every 1 s All measurements were conducted at ambient pressure and room temperature gas products were analyzed using gas chromatography (Shimadzu GC-8A) equipped with a thermal conductivity detector the electrolytes were collected to quantify the NH3 and N2H4 products using UV-Vis spectrophotometry The concentration of NH3 product was spectrocolorimetrically determined by the indophenol blue method25 1 ml of electrolyte was taken out from the cathodic compartment in H-typce cell (and diluted if needed) ≥97%) solution containing 5 wt% salicylic acid (Sigma-Aldrich ≥99%) and 5 wt% sodium citrate (Sigma-Aldrich ≥99%) was added to the 1 ml of electrolyte followed by the addition of 0.5 ml of 0.05 M NaClO (Sigma-Aldrich available chlorine 4.00–4.99%) and 0.1 ml of 1 wt% C5FeN6Na2O (Sigma-Aldrich The solution was stored in dark for ~2 h at room temperature before the UV-vis measurement The concentration of NH3 was determined by absorbance at a wavelength of 655 nm ≥99.5%) standard solutions with known concentration was used for the calibration 2.5 ml of the diluted electrolyte were mixed with 0.05 ml of color reagent and stored ~30 min prior to the measurement The absorbance at a wavelength of 540 nm was collected by UV-vis spectrometer to quantify the concentration of \({{{\rm{NO}}}}_{2}^{{-}}\) ≥96%) standard solutions with known concentration was used for the calibration 1 ml of the electrolyte was mixed with 1 mL of color reagent and stored for ~1 h before the measurment The absorbance at a wavelength of 455 nm was collected by UV-vis spectrometer to determinte the concentration of N2H4 80% in water) standard solutions with known concentration was used for the calibration Electrocatalytic nitrate reduction for sustainable ammonia production Drinking water from private wells and risks to children State-of-the-art and perspectives of the catalytic and electrocatalytic reduction of aqueous nitrates Electrochemical synthesis of ammonia as a potential alternative to the Haber–Bosch process Breaking scaling relations to achieve low-temperature ammonia synthesis through LiH-mediated nitrogen transfer and hydrogenation Advances in electrocatalytic N2 reduction—strategies to tackle the selectivity challenge Tandem electrocatalytic nitrate reduction to ammonia on MBenes Efficient ammonia electrosynthesis from nitrate on strained ruthenium nanoclusters Efficient electroreduction of nitrate into ammonia at ultra‐low concentrations via enrichment effect Enhanced nitrate-to-ammonia activity on copper–nickel alloys via tuning of intermediate adsorption Efficient electrochemical nitrate reduction to ammonia with copper‐supported rhodium cluster and single‐atom catalysts Interfacial engineering enhances the electroactivity of frame‐like concave RhCu bimetallic nanocubes for nitrate reduction Electrochemical reduction of nitrate to ammonia via direct eight-electron transfer using a copper–molecular solid catalyst Unveiling the activity origin of a copper‐based electrocatalyst for selective nitrate reduction to ammonia 3.4% solar‐to‐ammonia efficiency from nitrate using fe single atomic catalyst supported on MoS2 nanosheets Solar-driven electrochemical synthesis of ammonia using nitrate with 11% solar-to-fuel efficiency at ambient conditions Splicing the active phases of copper/cobalt-based catalysts achieves high-rate tandem electroreduction of nitrate to ammonia Governing interlayer strain in bismuth nanocrystals for efficient ammonia electrosynthesis from nitrate reduction mechanism and application of selective electrosynthesis of ammonia from nitrate on cobalt phosphide Interfacial reduction nucleation of noble metal nanodots on redox-active metal–organic frameworks for high-efficiency electrocatalytic conversion of nitrate to ammonia Ambient ammonia synthesis via electrochemical reduction of nitrate enabled by NiCo2O4 nanowire array Coordination symmetry breaking of single atom catalysts for robust and efficient nitrate electroreduction to ammonia Built‐in electric field triggered interfacial accumulation effect for efficient nitrate removal at ultra‐low concentration and electroreduction to ammonia Electrochemical ammonia synthesis via nitrate reduction on Fe single atom catalyst A single-site iron catalyst with preoccupied active centers that achieves selective ammonia electrosynthesis from nitrate Sub-nm RuO x clusters on Pd metallene for synergistically enhanced nitrate electroreduction to ammonia Photoelectrochemical nitrate reduction to ammonia on ordered silicon nanowire array photocathodes Enhanced charge‐carrier dynamics and efficient photoelectrochemical nitrate‐to‐ammonia conversion on antimony sulfide‐based photocathodes Enhanced charge carrier dynamics on Sb2Se3 photocathodes for efficient photoelectrochemical nitrate reduction to ammonia Control of metal-support interactions in heterogeneous catalysts to enhance activity and selectivity Tailoring electronic structure of bifunctional Cu/Ag layered electrocatalysts for selective CO2 reduction to CO and CH4 Rational catalyst design for N2 reduction under ambient conditions: strategies toward enhanced conversion efficiency High-throughput identification of highly active and selective single-atom catalysts for electrochemical ammonia synthesis through nitrate reduction Boosting CO2 hydrogenation via size-dependent metal–support interactions in cobalt/ceria-based catalysts Support and interface effects in water‐splitting electrocatalysts Pt nanoclusters on GaN nanowires for solar-asssisted seawater hydrogen evolution Oxynitrides enabled photoelectrochemical water splitting with over 3000 hrs stable operation in practical two-electrode configuration Crystallographic effects of GaN nanostructures in photoelectrochemical reaction X-ray photoelectron spectroscopy and x-ray diffraction study of the thermal oxide on gallium nitride Interfacial chemistry and valence band offset between GaN and Al2O3 studied by X-ray photoelectron spectroscopy Jr Handbook of X-ray photoelectron spectroscopy X-ray photoelectron spectroscopic investigation of plasma-enhanced chemical vapor deposited NiO x and CoNiO x (OH) y: influence of the chemical composition on the catalytic activity for the oxygen evolution reaction High efficiency Si photocathode protected by multifunctional GaN nanostructures Silver halide catalysts on GaN nanowires/Si heterojunction photocathodes for CO2 reduction to syngas at high current density Unraveling the rate-limiting step of two-electron transfer electrochemical reduction of carbon dioxide Electrocatalytic reduction of Nitrate on Copper single crystals in acidic and alkaline solutions Hydrogel protection strategy to stabilize water-splitting photoelectrodes Accelerating 1H NMR detection of aqueous ammonia Spectrophotometric method for determination of hydrazine Download references This work was supported by United States Army Research Office Award W911NF2110337 Peng Zhou for discussions on the catalytic reactions These authors contributed equally: Wan Jae Dong Department of Electrical Engineering and Computer Science Department of Integrative Energy Engineering Graduate School of Energy and Environment (KU-KIST Green School) performed initial calculations and discussed possible catalytic mechanisms All authors discussed the results and commented on the manuscript Some IP related to this work was licensed to NS Nanotech The University of Michigan and MI have a financial interest in the company The remaining authors declare no competing interests reviewers for their contribution to the peer review of this work Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations Download citation DOI: https://doi.org/10.1038/s41467-025-58372-7 Metrics details The stacking sequence of two-dimensional hexagonal boron nitride (hBN) is a critical factor that determines its polytypes and its distinct physical properties Although most hBN layers adopt the thermodynamically stable AA′ stacking sequence achieving alternative stacking configurations has remained a long-standing challenge Here we demonstrate the scalable synthesis of hBN featuring unprecedented AA stacking where atomic monolayers align along the c axis without any translation or rotation This previously considered thermodynamically unfavourable hBN polytype is achieved through epitaxial growth on a two-inch single-crystalline gallium nitride wafer using a metal–organic chemical vapour deposition technique Comprehensive structural and optical characterizations evidence the formation of AA-stacked multilayer hBN and reveal that hBN nucleation on the vicinal gallium nitride surface drives the unidirectional alignment of layers Here electron doping plays a central role in stabilizing the AA stacking configuration Our findings provide further insights into the scalable synthesis of engineered hBN polytypes characterized by unique properties such as large optical nonlinearity Prices may be subject to local taxes which are calculated during checkout The data supporting the findings of this study are available within this article and its Supplementary Information Additional data are available from the corresponding authors upon reasonable request Electronic structure of few-layer graphene: experimental demonstration of strong dependence on stacking sequence Stacking-dependent electronic structure of trilayer graphene resolved by nanospot angle-resolved photoemission spectroscopy Observation of Van Hove singularities in twisted graphene layers Intrinsic quantized anomalous Hall effect in a moiré heterostructure Tunable strongly coupled superconductivity in magic-angle twisted trilayer graphene Designed growth of large bilayer graphene with arbitrary twist angles Alternative stacking sequences in hexagonal boron nitride Hexagonal boron nitride is an indirect bandgap semiconductor Hexagonal boron nitride for next-generation photonics and electronics Flexible and transparent MoS2 field-effect transistors on hexagonal boron nitride-graphene heterostructures Boron nitride substrates for high-quality graphene electronics Field-effect tunneling transistor based on vertical graphene heterostructures Transport properties of pristine few-layer black phosphorus by van der Waals passivation in an inert atmosphere Van der Waals heterostructure of hexagonal boron nitride with an AlGaN/GaN epitaxial wafer for high-performance radio frequency applications Quantum emission from hexagonal boron nitride monolayers Initialization and read-out of intrinsic spin defects in a van der Waals crystal at room temperature Far-ultraviolet plane-emission handheld device based on hexagonal boron nitride Deep-ultraviolet electroluminescence and photocurrent generation in graphene/hBN/graphene heterostructures A flexible solar-blind 2D boron nitride nanopaper-based photodetector with high thermal resistance Bright luminescence from indirect and strongly bound excitons in h-BN Flat bands and giant light-matter interaction in hexagonal boron nitride Stacking-engineered ferroelectricity in bilayer boron nitride Enhanced tunable second harmonic generation from twistable interfaces and vertical superlattices in boron nitride homostructures Bernal boron nitride crystals identified by deep-ultraviolet cryomicroscopy Determination of the optical bandgap of the Bernal and rhombohedral boron nitride polymorphs Highly ordered boron nitride/epigraphene epitaxial films on silicon carbide by lateral epitaxial deposition Bevel-edge epitaxy of ferroelectric rhombohedral boron nitride single crystal Structural and electronic transitions in few layers of isotopically pure hexagonal boron nitride Epitaxial growth of a 100-square-centimetre single-crystal hexagonal boron nitride monolayer on copper Wafer-scale single-crystal hexagonal boron nitride monolayers on Cu(111) Epitaxial single-crystal hexagonal boron nitride multilayers on Ni(111) Large-area two-dimensional layered hexagonal boron nitride grown on sapphire by metalorganic vapor phase epitaxy Wafer-scale and selective-area growth of high-quality hexagonal boron nitride on Ni(111) by metal-organic chemical vapor deposition Wafer-scale single-crystal hexagonal boron nitride film via self-collimated grain formation Atomic-level defect modulation and characterization methods in 2D materials Structural and electronic properties of h-BN Synthesis of large-area multilayer hexagonal boron nitride for high material performance Stacking in bulk and bilayer hexagonal boron nitride Probing symmetry properties of few-layer MoS2 and h-BN by optical second-harmonic generation Monolayer boron nitride: hyperspectral imaging in the deep ultraviolet Hunting for monolayer boron nitride: optical and Raman signatures A low frequency Raman-active vibration of hexagonal boron nitride Overtones of interlayer shear modes in the phonon-assisted emission spectrum of hexagonal boron nitride Low frequency Raman spectroscopy of few-atomic-layer thick hBN crystals Nanometric resolved luminescence in h-BN flakes: excitons and stacking order Resonant Raman scattering in cubic and hexagonal boron nitride Layer-dependent second-order Raman intensity of MoS2 and WSe2: influence of intervalley scattering Phonon dispersions and piezoelectricity in bulk and multilayers of hexagonal boron nitride Spatially-resolved UV-C emission in epitaxial monolayer boron nitride and imperfect circular dichroism in monolayer h-BN Atomic-scale investigation of the lattice-asymmetry-driven anisotropic sublimation in GaN Effect of step edges on adsorption behavior for GaN(0001) surfaces during metalorganic vapor phase epitaxy: an ab initio study Native point defects and impurities in hexagonal boron nitride Probe: a software for high-resolution STEM image simulation VESTA 3 for three-dimensional visualization of crystal Deep ultraviolet hyperspectral cryomicroscopy in boron nitride: photoluminescence in crystals with an ultra-low defect density Qudi: a modular Python suite for experiment control and data processing QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials Generalized gradient approximation made simple Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu Effect of the damping function in dispersion corrected density functional theory Hybrid functionals based on a screened Coulomb potential Direct observation of the band structure in bulk hexagonal boron nitride Download references This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MSIT) (nos and the Technology Innovation Program (grant no Development of sub-micrometer blue color light source for modular display) funded by the Ministry of Trade are grateful for financial support from Samsung Electronics (nos IO210520-08650-01 and IO210707-08808-01) and Samsung Funding & Incubation Center of Samsung Electronics under project no acknowledge support from the internal fund/grant of Electronics and Telecommunications Research Institute (ETRI) (no Development of nano-sheet transfer-based ultra-thin GaN transistor technology) are grateful for financial support from the BONASPES projects (no acknowledge support from a Korea Basic Science Institute (National Research Facilities and Equipment Center) grant funded by the Ministry of Education (no for helpful discussions on the investigations of the band structure of hBN polytypes Department of Materials Science and Engineering Pohang University of Science and Technology Daegu Gyeongbuk Institute of Science and Technology (DGIST) Ulsan National Institute of Science and Technology Graduate School of Semiconductor Materials and Devices Engineering School of Semiconductor and Chemical Engineering International Center for Materials Nanoarchitectonics School of Materials Science and Engineering conducted the growth of the AA-stacked hBN films reflectivity and second-order Raman scattering measurements carried out the theoretical calculations of optical properties prepared samples for the STEM and TEM measurements carried out the scanning electron microscopy NEXAFS and ultraviolet photoelectron spectroscopy measurements All authors contributed to the writing of the paper Nature Materials thanks the anonymous reviewers for their contribution to the peer review of this work a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law Download citation DOI: https://doi.org/10.1038/s41563-025-02173-2 Metrics details We fabricated GaN/BN double heterostructure light-emitting diodes (LEDs) where the BN layer exhibited an amorphous-like short-range order and facilitated the in-situ epitaxial lateral overgrowth (ELOG) of GaN films Using an identical metal-organic chemical vapor deposition the BN layer was reliably formed on the GaN film and then served as a growth mask during the high-temperature growth of the GaN overlayer The BN layers were well dispersed over the entire surface with a partial coverage of 40–60% and a thickness of a few nm The laterally overgrown GaN was epitaxially related to the initial GaN film exhibiting single crystallinity with flat and smooth surface morphology the in-situ-formed BN layer effectively blocked the threading dislocations where its density reductions were comparable to those of typical ex-situ ELOG processes the BN-assisted ELOG reduced the mosaic of the practical single crystalline GaN grains and drastically improved crystallographic alignment and internal quantum efficiency the BN-assisted ELOG yielded high device performance of the GaN LEDs demonstrating that the benefits of ELOG were fully achieved with the fast and instant fabrication process we employed an extremely thin SRO BN layer to demonstrate the in-situ epitaxial lateral overgrowth (ELOG) of GaN light-emitting diodes (LEDs) The formation of the BN mask layer only requires in-situ interruption during the growth of the GaN layer which greatly simplifies and accelerates the ELOG process The BN layer exhibited high-temperature tolerance and served as dislocation blocks to markedly improve the structural and optical qualities of the overgrown GaN for fabricating high-efficiency devices To preserve the advantages of heteroepitaxy aimed at the monolithic integration of a substrate comprising different functional materials with the semiconductor overlayers the ELOG process should ideally be seamless to enable high manufacturing scalability Taking advantage of these favorable properties we fabricated a GaN/BN/GaN heterostructure monolithically where the SRO BN interlayer functioned as a growth mask to effectively prevent the threading dislocations in the bottom GaN layer from extending any further The BN mask remains on the sample and ELOG is initiated from the openings of the mask This enabled the subsequent growth of highly crystalline GaN overlayers The fabricated heterostructure was systematically investigated to demonstrate the growth reliability and cost-effectiveness of SRO BN for achieving high crystalline GaN on heterogeneous substrates we fabricated the p − n junction LEDs using the BN-assisted ELOG to show our approach can be further utilized to improve the practical device performance efficiently A vertical-type homemade MOCVD reactor was employed for the in-situ ELOG of the GaN films contained a SiC-coated graphite susceptor that was remotely heated by an induction heater The susceptor continually rotated at 40 rpm during sample growth for superior growth uniformity For the deposition of both the GaN and BN layers trimethylgallium (TMGa) and borazine were introduced into the same MOCVD reactor and NH3 (99.9995%) gases were installed in the MOCVD SRO BN can be formed to enable the in-situ ELOG The BN-coated GaN film showed a root mean square roughness (RMS) of 3.25 nm while the RMS roughness of the initial GaN film was 0.34 nm The thickness variation of the BN layer showed a quite large range from 1.4 nm to 13.2 nm we are using the BN deposition condition before its full coalescence where the initial growth stage typically involves multiple nucleation and coalescence processes to result in thickness variations we re-grew the GaN layers on the BN-coated GaN film for 1 h The pressure in the chamber was maintained at 100 Torr throughout the entire growth process TMGa and NH3 were supplied into the chamber at flow rates of 54 µmol min−1 and 1 l min−1 we employed 7.5 µmol min−1 of borazine with 1 l min−1 of N2 ambient gas a Si-doped n-GaN layer was grown on the BN-coated GaN film by ELOG and a Mg-doped p-GaN layer was coated on the n-GaN layer to make the p−n junction The p-GaN layer was thermally activated in N2 ambient at 850 °C composed of 200 μm × 200 μm square arrays of 20 nm/20 nm Ni/Au bilayer then thermally annealed in air ambient at 500 °C The selective-area ICP-RIE was performed to expose the n-GaN layer Indium pellets were used to form n-contact we additionally fabricated p−n junction GaN LEDs without inserting the BN mask the growth and fabrication processes were identical to those of the BN-inserted GaN LED The morphological analyses were carried out using field emission scanning electron microscopes (FE-SEM JEOL) operated in secondary electron measurement mode The main operating parameters were 8.5 mm (working distance) Chemical information of the BN layers was acquired using X-ray photoelectron spectroscopy (XPS The spectrometer was equipped with a monochromatic Al Kα x-ray excitation source (~1486.6 eV) and the performance of the system was first assessed by using a clean silver reference sample and the XPS core level scan was fixed at 50 eV and 0.1 eV The hemispherical electron analyzer was located normal to the thin film surface (i.e a 90° electron take-off angle) and dedicated software (Thermo Avantage v5.962 Thermo Fisher Scientific) was used to control the instrument as well as for data acquisition/processing The analyzed area has an average diameter of 900 µm and an ultra-high vacuum level of ~10−10 torr was maintained in the analysis chamber as a base pressure The diffusion of boron into the epitaxial laterally overgrown GaN layers was investigated by a time-of-flight secondary ion mass spectrometry (TOF-SIMS A 1 keV O2+ sputter ion beam etched the sample of the GaN film epitaxial laterally overgrown on c-Al2O3 with BN mask interlayers a 25 keV Bi+ primary ion gun was employed to scan the center region of the etching spot where the scanning area was 50 × 50 μm2 Specimens for cross-sectional TEM were prepared by a focused ion beam (Helios NanoLab 450 JEOL) with acceleration at 200 kV equipped with a high-sensitivity CMOS camera (OneView Gatan) was used to acquire the TEM images and selective-area electron diffraction (SAED) patterns Samples were mounted on a double-tilt holder and analyzed and images were obtained under bright-field or dark-field conditions All TEM experiments were conducted at room temperature The synchrotron radiation X-ray diffraction (SR-XRD) analyses including reciprocal space mapping for our thin film samples were carried out at the 5A MS-XRS beamline of the Pohang Acceleration Laboratory The sample was mounted directly on a six-circle diffractometer which allows flexible orientation of the sample in six different rotational axes The photon source energy and beam spot size were 17.9 keV and 600 µm × 500 µm respectively The expected resolution of the system is 0.032(5) degrees The optical characteristics were investigated by measuring photoluminescence (PL) Andor Technology) linked with a monochromator (Omni-λ2002i was employed for the optical characterization A Keithley 2400 source meter was used to measure the electrical characteristics of the LEDs The electroluminescence (EL) measurement system consists of a probing stage capturing the light emission images and the EL spectra To compare EL characteristics of the LEDs with and without BN-assisted ELOG including the identical distance between the sample and the optical fiber connected to the spectrometer The output power was measured by a power meter (Newport 1919-R) to derive the external quantum efficiency (EQE) of the LED All the EL characteristics are obtained from the backside of the LEDs where the front side is covered by the Ni/Au contact pads b FE-SEM image of the grown GaN islands on the BN-coated GaN film c \({\rm{g}}=0\bar{1}12\) DF TEM image of the GaN island/BN mask/GaN film structure and SAED patterns of the GaN island (top) and the GaN film (bottom) XPS core-level spectra for d B1s and e N1s f FE-SEM image of the fully merged GaN film by ELOG The identified Bragg peaks on the SAED patterns as well as their corresponding zone axes indicated the epitaxial relation between the GaN islands and the underlying GaN film (GaN island (\(0\bar{1}12\)) || GaN film (\(0\bar{1}12\)) and GaN island (\(10\bar{1}1\)) || GaN film (\(10\bar{1}1\))) The horizontal dislocation lines in the overgrown GaN islands imply its lateral overgrowth The XPS results strongly suggest the presence of the BN mask and its durability during the high-temperature growth of the GaN overlayer at temperatures of 1060–1100 °C the TEM analysis of the thick BN film suggests the amorphous-like crystallinity of the BN localized nanocrystalline grains were identified in the BN layer the BN layer exhibited a partial regular lattice array with a vertical interlayer spacing of 3.3 Å and a lateral lattice spacing of 2.1 Å These values correspond to the out-of-plane lattice constant and the lattice interval along the armchair direction of h-BN The structure includes the main component is amorphous but also has other nanocrystalline structures with short-range order a Magnified g = 0002 DF image near the GaN/c-Al2O3 heterojunction The BN mask was located 140 nm above the substrate The white arrow marks a spot that is uncovered by the BN mask b g = 0002 and c g = \(11\bar{2}0\) DF images displaying another region where the BN mask exists d BF image of the GaN film grown by ELOG with 20s-coated BN mask The threading dislocations are marked as red arrows XRD rocking curves of the a GaN(0002) and b GaN(\(10\bar{1}0\)) diffraction peaks The solid black lines and red dotted lines show GaN films grown on c-Al2O3 with and without using the BN interlayer RSMs of the GaN films were grown by c in-situ ELOG with the BN mask and by d typical heteroepitaxy without using the BN interlayer a PL spectra of the GaN films obtained by BN-assisted ELOG (black solid line) and by typical heteroepitaxy without using the BN interlayer (red dotted line) The power density of the laser was 39 mWcm−2 and the PL intensity of the GaN film grown by typical heteroepitaxy was multiplied 10 times for better visibility b Integrated PL intensity as a function of the power density of the laser a Room-temperature EL spectra of the LEDs at various applied currents in the range of 10−50 mA with 10 mA intervals The EL intensity increased as increasing the currents The inset shows the light emission images at 50 mA which suggests the improved device performance by the in-situ ELOG two heterogeneous materials of crystalline GaN and SRO BN were combined by a monolithic growth to demonstrate the in-situ ELOG Using a few-nm-thick SRO-BN growth mask that provides partial surface coverage the benefits of ELOG were fully demonstrated including the prevention of the propagation of edge-type and screw-type threading dislocations enlargement of the crystal grains of the semiconductor thin films the BN-assisted in-situ ELOG improved the GaN crystallinity comparable to the typical ex-situ ELOG processes while our approach considerably reduced the fabrication labor as no off-site growth interruptions or lithography was involved the in-situ ELOG yielded high device performance of the GaN LEDs The SRO BN can be prepared by various apparatus with a wide range of growth temperatures which will allow us to employ our in-situ ELOG method to prepare other semiconductor and/or functional oxide materials with greater crystallinity Especially for the conventional growth system where tens or hundreds of wafers are loaded in one chamber in-situ ELOG can become significantly more effective Remote heteroepitaxy of GaN microrod heterostructures for deformable light-emitting diodes and wafer recycle Vertical full-colour micro-LEDs via 2D materials-based layer transfer Application of two-dimensional MoSe2 atomic layers to the lift-off process for producing light-weight and flexible bifacial Cu(In Chip-less wireless electronic skins by remote epitaxial freestanding compound semiconductors ZnO nanotube waveguide arrays on graphene films for local optical excitation on biological cells Combining freestanding ferroelectric perovskite oxides with two-dimensional semiconductors for high performance transistors Transferable single-crystal GaN thin films grown on chemical vapor-deposited hexagonal BN sheets Quasi Van der Waals epitaxy nitride materials and devices on two-dimension materials Atomic evolution mechanism and suppression of edge threading dislocations in nitride remote heteroepitaxy The origin of high activity of amorphous MoS2 in the hydrogen evolution reaction Unlocking more potentials in two-dimensional space: disorder engineering in two-dimensional amorphous carbon Ultra-low threshold field emission from amorphous BN nanofilms Realization of continuous Zachariasen carbon monolayer Lateral epitaxial overgrowth of nitrogen polar GaN on smooth nitrogen polar GaN templates by metalorganic chemical vapor deposition Heteroepitaxial growth of high-quality GaN thin films on Si substrates coated with self-assembled sub-micrometer-sized silica balls Epitaxial lateral overgrowth of GaN on nano-cavity patterned sapphire substrates One-step graphene coating of heteroepitaxial GaN films Epitaxial GaN microdisk lasers grown on graphene microdots Flexible GaN light-emitting diodes sing GaN microdisks epitaxial laterally overgrown on graphene dots Epitaxial lateral overgrowth of GaN on sapphire substrates using in-situ carbonized photoresist mask Low-temperature direct growth of amorphous boron nitride films for high-performance nanoelectronic device applications Oxide two-dimensional electron gas with high mobility at room-temperature Self-separation of thick two-inch GaN layers grown by HVPE on sapphire using epitaxial lateral overgrowth with masks containing tungsten Study on nucleation and growth mode of GaN on patterned graphene by epitaxial lateral overgrowth Microscopic lateral overgrowth by physical vapour transport of GaN on self-organized diamond-like carbon masks Optical and structural microanalysis of GaN grown on SiN submonolayers Effects of dislocations on the carrier transport and optical properties of GaN films grown with an in-situ SiNx insertion layer Thermal annealing effect on the structural and the optical properties of nano CdTe films Optical and structural study of GaN nanowires grown by catalyst-free molecular beam epitaxy Measurement and analysis of photoluminescence in GaN Stress-induced piezoelectric field in GaN-based 450-nm light-emitting diodes Diode ideality factor in modern light-emitting diodes Download references This work was financially supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (NRF-2021R1A5A1032996 NRF-2021R1C1C1010924) and Korea Institute for Advancement of Technology (KIAT) grant funded by the Korea Government (MOTIE) (P0023703 These authors contributed equally: Heesoo Kim Ulsan National Institute of Science and Technology (UNIST) Research Center for Novel Epitaxial Quantum Architectures designed and progressed the whole experiment; B.K contributed to the TEM characterizations; H.J contributed to the XRD characterizations; A.A contributed to the XPS characterization; H.B contributed to the PL characterization; I.R. contributed to the LED fabrication and characterization; A.A. analyzed the data and wrote the manuscript All authors have given approval to the final version of the manuscript This study does not involve human participants All experiments and analyses were conducted using publicly available data/materials or in compliance with standard laboratory protocols Download citation DOI: https://doi.org/10.1038/s41427-025-00594-8 Metrics details This mixed study presents the optical properties of gallium nitride material at different conditions and thicknesses over a very wide wavelength range for optical simulations The availability of optical properties of gallium nitride over a very wide wavelength range facilitates the use of gallium nitride in different simulated optical devices such as laser diodes and MATLAB codes for refractive indices will be presented to reduce mistakes within simulations Providing a database for material RIs is crucial for enabling accurate simulations across various fields A database containing the RIs of different materials helps in determining their optical properties Zincblende and wurtzite are two different crystal structures adopted by GaN The lattice structures of wurtzite and zincblende GaN are hexagonal close-packed and cubic lattice structures Wurtzite GaN is the most common and commonly used in optical devices zincblende GaN has been investigated for possible uses in specific electronic devices The distinct characteristics and behaviors of zincblende and wurtzite GaN may impact the functionality of GaN-fabricated devices The particular needs of the gadget being produced determine the crystal structure to be used If these results are fitted to simple equations it may help in different theoretical applications the RI of GaN will be fitted and studied at different conditions Fitting and grouping the rRIs of GaN over a wide range of wavelengths in the same paper may eliminate the mistakes in simulation studies and the plotted curve with initial conditions helps researchers (maybe beginners) verify their codes and results step by step We fitted the wavelength in all the equations in this manuscript to be microns This connection between experimental and fitted RIs facilitates the fabrication of theoretical studies We fitted the experiment data using the Sellmeier formula as follows: For wurtzite \(\:\text{G}\text{a}\text{N}\): For zincblende \(\:\text{G}\text{a}\text{N}\): and 515 °C) over the 0.367 –1.000 μm wavelength range (\(\:\lambda\:\) in µm) The experimental data are fitted using the Sellmeier formula as follows: The RI of \(\:\text{G}\text{a}\text{N}\) versus wavelength at different temperatures can be investigated as a function of the thermo-optic coefficient using the following equation: The deduced experimental and fitted indices of refraction are presented in Fig. 6. where \(\:{\text{n}}_{0}^{\:}\) is the pores’ RI where \(\:{A}_{1}=2.637\:+\:0.004\text{i}\); \(\:{A}_{2}=-5.842\:+\:0.007\text{i}\); \(\:{A}_{3}=0.004\:+\:2.176\text{i}\); \(\:{A}_{4}=5.370\:-\:0.002\text{i}\); \(\:{A}_{5}=0.327\:+\:1.314\times\:{10}^{-5}\text{i}\); \(\:{A}_{6}=54264698.942\:-\:8537608.451\text{i}\); \(\:{A}_{7}=5052995.106\:-\:829099.109\text{i}\) and 506 °C) on (0001) oriented sapphire substrate over 0.370 –1.700 μm wavelength range (\(\:\lambda\:\) in µm) We fitted the experimental data using the Sellmeier formula over the 0.390 –1.700 μm wavelength range (\(\:\lambda\:\) in µm) as follows: Thermo-optic coefficient of \(\:\text{G}\text{a}\text{N}\) versus the wavelength according to Eq. (20) The RI of \(\:\text{G}\text{a}\text{N}\) versus wavelength at different temperatures from 24 to 506 °C can be investigated as a function of the thermo-optic coefficient over 0.390–1.700 μm wavelength range using the following equation: Calculated extinction coefficient of cubic \(\:\text{G}\text{a}\text{N}\) versus a wide range of wavelengths the RI of GaN has been studied a wide wavelength from 0.35 μm to 3099.5 μm the RI of GaN strongly decreases and slightly decreases in near IR spectra the RI of GaN can be considered a constant value (ex the effect of different parameters such as the temperature from 21 to 515 °C and pressure from 0 GPa to 40 GPa are studied and fitted Requests for materials or code should be addressed to Zaky A Wu, F. et al. Tunable large omnidirectional photonic bandgap in a one-dimensional photonic crystal comprising elliptical metamaterials based on elasto-optic effect. Eur. Phys. J. Plus. 139, 410. https://doi.org/10.1140/epjp/s13360-024-05226-8 (2024) Zaky, Z. A., Amer, H. A., Suthar, B. & Aly, A. H. Gas sensing applications using magnetized cold plasma multilayers. Opt. Quant. Electron. 54, 217. https://doi.org/10.1007/s11082-022-03594-y (2022) Zaky, Z. A., Alamri, S., Zhaketov, V. & Aly, A. H. Refractive index sensor with magnified resonant signal. Sci. Rep. 12, 13777. https://doi.org/10.1038/s41598-022-17676-0 (2022) She, Y. et al. Tunable wide-angle high-efficiency polarization selectivity based on a one-dimensional photonic crystal containing elliptical metamaterials. Phys. Lett. A. 494, 129299. https://doi.org/10.1016/j.physleta.2023.129299 (2024) Zaky, Z. A., Hanafy, H., Panda, A., Pukhrambam, P. D. & Aly, A. H. Design and Analysis of Gas Sensor Using Tailorable Fano Resonance by Coupling Between Tamm and Defected Mode Resonance. Plasmonics 17, 2103–2111. https://doi.org/10.1007/s11468-022-01699-4 (2022) Zaky, Z. A., Al-Dossari, M., Zohny, E. I. & Aly, A. H. Refractive index sensor using fibonacci sequence of gyroidal graphene and porous silicon based on Tamm Plasmon Polariton. Opt. Quant. Electron. 55, 6. https://doi.org/10.1007/s11082-022-04262-x (2023) Zaky, Z. A., Singh, M. R. & Aly, A. H. Tamm resonance excited by different metals and graphene. Photonics Nanostructures-Fundamentals Appl. 49, 100995. https://doi.org/10.1016/j.photonics.2022.100995 (2022) Zaky, Z. A., Panda, A., Pukhrambam, P. D. & Aly, A. H. The impact of magnetized cold plasma and its various properties in sensing applications. Sci. Rep. 12, 3754. https://doi.org/10.1038/s41598-022-07461-4 (2022) Wu, F. et al. Strong polarization-dependent photonic bandgap and giant photonic spin Hall effect in a one-dimensional photonic crystal entirely composed of all-dielectric metamaterials. Phys. B: Condens. Matter. 670, 415348. https://doi.org/10.1016/j.physb.2023.415348 (2023) She, Y. et al. Large angle-insensitive photonic band gap in a one-dimensional photonic hypercrystal in the visible range based on the phase-variation compensation theory. Phys. Lett. A. 521, 129752. https://doi.org/10.1016/j.physleta.2024.129752 (2024) Zaky, Z. A., Al-Dossari, M., Matar, Z. & Aly, A. H. Effect of geometrical and physical properties of cantor structure for gas sensing applications. Synth. Met. 291, 117167. https://doi.org/10.1016/j.synthmet.2022.117167 (2022) Ameen, A. A., Al-Dossari, M., Zaky, Z. A. & Aly, A. H. Studying the effect of quantum dots and parity-time symmetry on the magnification of topological edge state peak as a pressure sensor. Synth. Met. 292, 117233. https://doi.org/10.1016/j.synthmet.2022.117233 (2023) Xu, K., Zhang, J., Yang, B., Wu, F. & Yin, C. Non-reciprocal optical Tamm state in a photonic crystal heterojunction containing Weyl semimetals. Phys. B: Condens. Matter. 691, 416329. https://doi.org/10.1016/j.physb.2024.416329 (2024) Al-Dossari, M., Zaky, Z. A., Awasthi, S. K., Amer, H. A. & Aly, A. H. Detection of glucose concentrations in urine based on coupling of Tamm–Fano resonance in photonic crystals. Opt. Quant. Electron. 55, 484. https://doi.org/10.1007/s11082-023-04621-2 (2023) Zaky, Z. A., Al-Dossari, M., Sharma, A. & Aly, A. H. Effective pressure sensor using the parity-time symmetric photonic crystal. Phys. Scr. 98, 035522. https://doi.org/10.1088/1402-4896/acbcae (2023) Wu, F., She, Y., Zhou, T., Cheng, Z. & Huang, J. Hybrid one-dimensional photonic crystals containing anisotropic metamaterials: Angle-driven photonic band gaps and angle-driven Tamm Plasmon polaritons. Phys. Rev. A. 110, 023503. https://doi.org/10.1103/PhysRevA.110.023503 (2024) Zaky, Z. A. et al. Theoretical optimization of Tamm Plasmon Polariton structure for pressure sensing applications. Opt. Quant. Electron. 55, 738. https://doi.org/10.1007/s11082-023-05023-0 (2023) Zaky, Z. A. et al. Theoretical analysis of porous silicon one-dimensional photonic crystal doped with magnetized cold plasma for hazardous gases sensing applications. Opt. Quant. Electron. 55, 584. https://doi.org/10.1007/s11082-023-04907-5 (2023) Zaky, Z. A., Al-Dossari, M., Hendy, A. S., Zayed, M. & Aly, A. H. Gamma radiation detector using Cantor quasi-periodic photonic crystal based on porous silicon doped with polymer. Int. J. Mod. Phys. B. 2450409. https://doi.org/10.1142/S0217979224504095 (2024) Zaky, Z. A., Al-Dossari, M., Hendy, A. S. & Aly, A. H. Studying the impact of interface roughness on a layered photonic crystal as a sensor. Phys. Scr. 98, 105527. https://doi.org/10.1088/1402-4896/acfa4a (2023) Zaky, Z. A., Al-Dossari, M., Zhaketov, V. & Aly, A. H. Defected photonic crystal as propylene glycol THz sensor using parity-time symmetry. Sci. Rep. 14, 23209. https://doi.org/10.1038/s41598-024-73477-7 (2024) Huo, K., Zhang, J., He, M., Wu, F. & Yin, C. Tunable Fano resonance in a one-dimensional photonic crystal containing a Weyl Semimetal. Opt. Commun. 561, 130518. https://doi.org/10.1016/j.optcom.2024.130518 (2024) Usman, Z. et al. First-principle electronic, elastic, and optical study of cubic gallium nitride. J. Phys. Chem. A. 115, 6622–6628. https://doi.org/10.1021/jp201495e (2011) Lin, M., Sverdlov, B., Zhou, G. & Morkoc, H. A comparative study of GaN epilayers grown on sapphire and SiC substrates by plasma-assisted molecular‐beam epitaxy. Appl. Phys. Lett. 62, 3479–3481. https://doi.org/10.1063/1.109026 (1993) Prabaswara, A. et al. Review of GaN thin film and nanorod growth using magnetron sputter epitaxy. Appl. Sci. 10, 3050. https://doi.org/10.3390/app10093050 (2020) Song, H. & Lee, S. Red light emitting solid state hybrid quantum dot–near-UV GaN LED devices. Nanotechnology 18, 255202. https://doi.org/10.1088/0957-4484/18/25/255202 (2007) Jafar, N., Jiang, J., Lu, H., Qasim, M. & Zhang, H. Recent Research on Indium-Gallium-Nitride-Based Light-Emitting Diodes: Growth Conditions and External Quantum Efficiency, Crystals, vol. 13, p. 1623, (2023). https://doi.org/10.3390/cryst13121623 Zaky, Z. A. et al. Photonic crystal with magnified resonant peak for biosensing applications. Phys. Scr. 98, 055108. https://doi.org/10.1088/1402-4896/accbf1 (2023) Pezzagna, S., Brault, J., Leroux, M., Massies, J. & De Micheli, M. Refractive indices and elasto-optic coefficients of GaN studied by optical waveguiding. J. Appl. Phys. 103, 123112. https://doi.org/10.1063/1.2947598 (2008) Wang, S. et al. Gallium Nitride crystals: Novel Supercapacitor Electrode materials. Adv. Mater. (Deerfield Beach Fla). 28, 3768–3776. https://doi.org/10.1002/adma.201600725 (2016) Kim, Y. J. et al. Efficient light absorption by GaN truncated nanocones for high performance water splitting applications. ACS Appl. Mater. Interfaces. 10, 28672–28678. https://doi.org/10.1021/acsami.8b09084 (2018) Nie, B., Duan, B. K. & Bohn, P. W. Nanoporous GaN–Ag composite materials prepared by metal-assisted electroless etching for direct laser desorption-ionization mass spectrometry. ACS Appl. Mater. Interfaces. 5, 6208–6215. https://doi.org/10.1021/am401132s (2013) Castelletto, S. & Boretti, A. Gallium Nitride nanomaterials and Color Centers for Quantum Technologies. ACS Appl. Nano Mater. 7, 5862–5877. https://doi.org/10.1021/acsanm.3c05304 (2024) Wu, S. H., Chen, Z. G., Liu, H. S., Chen, S. H. & Wang, P. H. Fabrication of gallium nitride waveguide resonators by high-power impulse magnetron sputtering at room temperature. APL Photonics. 9, 016106. https://doi.org/10.1063/5.0169455 (2024) Taha, I. & Anjum, D. H. A review on Gallium Nitride for Liquid sensors: fabrications to applications. ACS Appl. Electron. Mater. https://doi.org/10.1021/acsaelm.4c00006 (2024) Lv, S. et al. Wafer Scale Gallium Nitride Integrated Electrode toward Robust High Temperature Energy Storage. Small 2310837. https://doi.org/10.1002/smll.202310837 (2024) Yu, G. et al. Optical properties of wurtzite structure GaN on sapphire around fundamental absorption edge (0.78–4.77 eV) by spectroscopic ellipsometry and the optical transmission method. Appl. Phys. Lett. 70, 3209–3211. https://doi.org/10.1063/1.119157 (1997) Watanabe, N., Kimoto, T. & Suda, J. The temperature dependence of the refractive indices of GaN and AlN from room temperature up to 515 C. J. Appl. Phys. 104, 106101. https://doi.org/10.1063/1.3021148 (2008) Yang, T. et al. Optical properties of GaN thin films on sapphire substrates characterized by variable-angle spectroscopic ellipsometry. Jpn. J. Appl. Phys. 37, L1105. https://doi.org/10.1143/JJAP.37.L1105 (1998) Lin, M., Sverdlov, B., Strite, S., Morkoç, H. & Drakin, A. Refractive indices of wurtzite and zincblende GaN. Electron. Lett. 29, 1759–1760. https://doi.org/10.1049/el:19931172 (1993) El-Naggar, A., El-Zaiat, S. & Hassan, S. M. Optical parameters of epitaxial GaN thin film on Si substrate from the reflection spectrum. Opt. Laser Technol. 41, 334–338. https://doi.org/10.1016/j.optlastec.2008.05.022 (2009) Torvik, J. T., Pankove, J. I., Iliopoulos, E., Ng, H. M. & Moustakas, T. D. Optical properties of GaN grown over SiO 2 on SiC substrates by molecular beam epitaxy. Appl. Phys. Lett. 72, 244–245. https://doi.org/10.1063/1.120698 (1998) Du, J. et al. Growth and Characteristics Analysis of the Thermal Oxide Grown on Gallium Nitride. In 2009 Symposium on Photonics and Optoelectronics, pp. 1–3. https://doi.org/10.1109/SOPO.2009.5230090 (2009) Alevli, M. et al. Substrate temperature influence on the properties of GaN thin films grown by hollow-cathode plasma-assisted atomic layer deposition. J. Vacuum Sci. Technol. A. 34, 01A125. https://doi.org/10.1116/1.4936230 (2016) Zhang, Y. et al. The fabrication of large-area, free-standing GaN by a novel nanoetching process. Nanotechnology 22, 045603. https://doi.org/10.1088/0957-4484/22/4/045603 (2010) Xi, X. et al. GaN nanocolumns fabricated by self-assembly ni mask and its enhanced photocatalytic performance in water splitting. Appl. Surf. Sci. 462, 310–315. https://doi.org/10.1016/j.apsusc.2018.08.113 (2018) Han, Y. et al. Large-area surface-enhanced Raman spectroscopy substrate by hybrid porous GaN with Au/Ag for breast cancer miRNA detection. Appl. Surf. Sci. 541, 148456. https://doi.org/10.1016/j.apsusc.2020.148456 (2021) Alvarez-Puebla, R. A. et al. Gold nanorods 3D-supercrystals as surface enhanced Raman scattering spectroscopy substrates for the rapid detection of scrambled prions. Proc. Natl. Acad. Sci. 108, 8157–8161. https://doi.org/10.1073/pnas.1016530108 (2011) Amador-Mendez, N. et al. Porous nitride light-emitting diodes. ACS Photonics. 9, 1256–1263. https://doi.org/10.1021/acsphotonics.1c01729 (2022) Huang, K. P. et al. InGaN light-emitting diodes with multiple-porous GaN structures fabricated through a photoelectrochemical etching process. ECS J. Solid State Sci. Technol. 3, R185. https://doi.org/10.1149/2.0131410jss (2014) Hayilesilassie, R. B. et al. Detection and photothermal inactivation of Gram-positive and Gram-negative bloodstream bacteria using photonic crystal biosensor and plasmonic core–shell. RSC Adv. 14, 11594–11603. https://doi.org/10.1039/D4RA01802H (2024) Aly, A. H., Mohamed, B., Awasthi, S., Abdallah, S. A. O. & Amin, A. MATLAB simulation based study on poliovirus sensing through one-dimensional photonic crystal with defect. Sci. Rep. 13, 9422. https://doi.org/10.1038/s41598-023-35595-6 (2023) Chowdhury, A., Ng, H. M., Bhardwaj, M. & Weimann, N. G. Second-harmonic generation in periodically poled GaN. Appl. Phys. Lett. 83, 1077–1079. https://doi.org/10.1063/1.1599044 (2003) Lee, K. J. et al. Enhanced optical output in InGaN/GaN light-emitting diodes by tailored refractive index of nanoporous GaN. Nanotechnology 30, 415301. https://doi.org/10.1088/1361-6528/ab31d0 (2019) Zaky, Z. A., Ahmed, A. M. & Aly, A. H. Remote Temperature Sensor Based on Tamm Resonance. Silicon 14, 2765–2777. https://doi.org/10.1007/s12633-021-01064-w (2021) Ahmed, A. M. & Mehaney, A. Ultra-high sensitive 1D porous silicon photonic crystal sensor based on the coupling of Tamm/Fano resonances in the mid-infrared region. Sci. Rep. 9, 6973. https://doi.org/10.1038/s41598-019-43440-y (2019) Zaky, Z. A. & Aly, A. H. Theoretical study of a tunable low-temperature Photonic Crystal Sensor using Dielectric-Superconductor Nanocomposite Layers. J. Supercond. Novel Magn. 33, 2983–2990. https://doi.org/10.1007/s10948-020-05584-1 (2020) Zaky, Z. A., Ahmed, A. M., Shalaby, A. S. & Aly, A. H. Refractive index gas sensor based on the Tamm state in a one-dimensional photonic crystal: theoretical optimisation. Sci. Rep. 10, 9736. https://doi.org/10.1038/s41598-020-66427-6 (2020) Khardani, M., Bouaïcha, M. & Bessaïs, B. Bruggeman effective medium approach for modelling optical properties of porous silicon: comparison with experiment. Phys. Status Solidi C. 4, 1986–1990. https://doi.org/10.1002/pssc.200674420 (2007) Banerjee, S. et al. Composite GaN–C–Ga (GaCN) layers with tunable refractive index. J. Phys. Chem. C. 122, 29567–29576. https://doi.org/10.1021/acs.jpcc.8b09142 (2018) Bolat, S., Tekcan, B., Ozgit-Akgun, C., Biyikli, N. & Okyay, A. K. Electronic and optical device applications of hollow cathode plasma assisted atomic layer deposition based GaN thin films. J. Vacuum Sci. Technol. A. 33, 01A143. https://doi.org/10.1116/1.4903365 (2015) Lee, Y. S. et al. Formation of GaN film on Si for microbolometer, in Infrared Technology and Applications XXXVIII, pp. 435–441. https://doi.org/10.1117/12.919072 (2012) El-Nahass, M. & Farag, A. Structural, optical and dispersion characteristics of nanocrystalline GaN films prepared by MOVPE. Opt. Laser Technol. 44, 497–503. https://doi.org/10.1016/j.optlastec.2011.08.021 (2012) Watanabe, N., Kimoto, T. & Suda, J. Thermo-optic coefficients of 4H-SiC, GaN, and AlN for ultraviolet to infrared regions up to 500 C. Jpn. J. Appl. Phys. 51, 112101. https://doi.org/10.1143/JJAP.51.112101 (2012) Gungor, N. & Alevli, M. Role of film thickness on the structural and optical properties of GaN on Si (100) grown by hollow-cathode plasma-assisted atomic layer deposition. J. Vacuum Sci. Technol. A. 36, 021514. https://doi.org/10.1116/1.5003154 (2018) Gungor, N. & Alevli, M. Visible/infrared refractive index and phonon properties of GaN films grown on sapphire by hollow-cathode plasma-assisted atomic layer deposition. J. Vacuum Sci. Technol. A. 37, 050901. https://doi.org/10.1116/1.5097467 (2019) Bowman, S. R., Brown, C. G. & Taczak, B. Optical dispersion and phase matching in gallium nitride and aluminum nitride. Opt. Mater. Express. 8, 1091–1099. https://doi.org/10.1364/OME.8.001091 (2018) Bowman, S. R. et al. Broadband measurements of the refractive indices of bulk gallium nitride. Opt. Mater. Express. 4, 1287–1296. https://doi.org/10.1364/OME.4.001287 (2014) Gasmi, F., Chemam, R., Graine, R., Boubir, B. & Meradji, H. Structural, electronic, and optical properties of the gallium nitride semiconductor by means of the FP-LAPW method. J. Mol. Model. 26, 356. https://doi.org/10.1007/s00894-020-04614-y (2020) Singh, B. P., Kumar, V. & Tripathy, S. First-principle calculations of optical and phonon properties of wurtzite gallium nitride. In AIP Conference Proceedings. https://doi.org/10.1063/1.4929262 (2015) Download references The authors are thankful to the Deanship of Graduate Studies and Scientific Research at University of Bisha for supporting this work through the Fast-Track Research Support Program Department of Computational Mathematics and Computer Science Institute of Natural Sciences and Mathematics Zaky invented the original idea of the study wrote and revised the main manuscript text M Al-Dossari analyzed the data and discussed the results Hendy analyzed the data and discussed the results Sallah analyzed the data and discussed the results revised the main manuscript text Aly analyzed the data and discussed the results all Authors developed the final manuscript This article does not contain any studies involving animals or human participants performed by any authors Below is the link to the electronic supplementary material Download citation DOI: https://doi.org/10.1038/s41598-025-89941-x Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research Metrics details Machine learning may enhance clinical data analysis but requires large amounts of training data While generative neural network models can create realistic synthetic data such as 3D MRI volumes and the generation of complex data remains challenging Fibre orientation distributions (FODs) represent one such complex data type modelling diffusion as spherical harmonics with stored weights as multiple three-dimensional volumes We successfully trained an α-WGAN combining a generative adversarial network and a variational autoencoder to generate synthetic FODs using the Human Connectome Project (HCP) data Our resulting synthetic FODs produce anatomically accurate fibre bundles and connectomes with properties matching those from our validation dataset Our approach extends beyond FODs and could be adapted for generating various types of complex medical imaging data particularly valuable for augmenting limited clinical datasets To achieve a clinically sufficient performance of such applications a large amount of training data is required that in turn may be augmented by synthetic data tract volumes and the fibre densities along the tracts The generated data is highly useful for the augmentation of training sets in which machine learning is applied to perform on FODs our work demonstrates that α-WGAN can be generalised to augment various 4D MRI sets which would be useful especially for pathological data where the amount of data is more limited The figure shows axial slices of single volumes representing coefficient maps from the harmonic spherical deconvolution we present synthetic data sets and in the sixth row we show one FOD from the validation set for comparison The number of the presented volume in each column refers to its position in the output volume of the MRtrix3 algorithm we present the centrum semiovale in the coronal section in which fibres from the corticospinal tract that are predominantly aligned caudocranially (blue colour encoding) cross fibres from the CC that are aligned from left to right (red colour encoding) we present the region of the forceps major in the axial section which connects the occipital parts of the hemispheres and merges with the optic radiation that is mainly aligned in anterior-posterior direction (green colour encoding) we chose the fasiculus arcuatus in the sagittal section that connects the Broca’s and Wernicke’s areas A Synthetic FOD volumes in coronal view. See caption of Fig. 2 for details. B Histograms of FODs. Distributions of FOD voxels for the entire synthetic data set corresponding to the volumes labelled above in A. Synthetic voxel data in red, validation voxel data in blue. Note the logarithmic scale. See caption of Fig. 2 for details The figure illustrates an exemplary connectome, derived from a synthetic FOD data set. A shows slice views through the whole brain tractogram in axial, coronal and sagittal views. B shows the obtained connectome (right), in relation to the whole brain tractogram, shown in a 3D view (left, top row, cf. [A], the AAL parcellation (left, middle row) and the connectome matrix plot (left, bottom row). We show four randomly selected structural connectomes based on the AAL atlas derived from generated FODs (A we present the connectome elements of the entire synthetic and validation data as histograms in (B) as well as the weighted global efficiency of normalised connectomes in (C) and the Mantel correlations in (D) Note that the ordinate in panel B is scaled logarithmically while (B The left column shows examples of the tractograms that we derived from one synthetic FOD data set using the FACT algorithm implemented in TractSeg we present histograms of dice scores that measure the volume overlap of selected tracts (arcuate fascicle [AF] Optic Radiation [OR] and Uncinate Fascicle [UF]) of all possible pairs of subjects that are both generated (red) We show results from tracts derived with the FACT and the iFOD2 algorithm and on the left and right-hand side we show the tract volumes from synthetic and validation data starting with the smallest one on the left-hand side to the largest one on the right-hand side The mean values are presented as crosses and dots the standard deviation is indicated by the coloured area we present the mean volume ratio of synthetic to validation data we show the fibre density along the indicated tracts The lines present the mean values at each streamline increment The relative differences of fibre-density mean and standard deviations (see Eq. (5)) in the AAL parcels reveal a bias in the cerebellum where they are up to 12 and 45 percent we find lower values under 5 and 20 percent that are distributed along the cortex without deducible bias towards a certain area The highly individual structure of the cerebellum may be more difficult to be learned by our model Here we have shown that GANs are capable of generating FODs as an example of complex MRI data that consist of several volumes compared to a pure VAE we tested for FODs with a lower spatial resolution (not shown) our α-WGAN generated less blurry FODs with increased spatial details While brain regions with lower variability in the training data such as the ventricles basal ganglia or CC are well represented in the synthetic data the model has demonstrated some difficulties in accurately generating high-variability regions with finely detailed structures characterised by individual configurations of sulci and gyri we find that synthetic data are slightly less correlated to each other than the validation data which might be a consequence of numerical noise incorporated by the CNNs As a test of the resulting bundle-specific tractograms we compared their dice-score distributions We found that tractograms derived from synthetic data resemble the ones from the validation data the dice scores indicate lower variation in the synthetic tractograms compared to the tractograms derived from the validation data that we already discovered comparing the pairwise mean squared values of each set that is substantially higher on average for the validation data This observation is in contrast to the mantel correlations it seems that the numerical noise has lower impact on single tracts than it has on whole-brain tractograms probably due to the restrictions by starting and ending ROIs Streamlines disturbed by the noise might be more likely to be rejected since they do not arrive at the ending ROI while they are counted in the whole-brain tractography This would also explain the lower volume of the tracts derived from synthetic data we have shown that our generated FODs capture the essential anatomical features but are still distinguishable from the validation FODs and the variation within the group of generated FODs is slightly smaller might enable us to generate high-resolution FODs indistinguishable from real data in the future It is worth to note that our approach is not limited to synthetic FODs but may be generalisable to synthesise any 4D MRI data functional MRI data or data from an entire MRI protocol containing multiple contrasts The generation of such multimodal data might be useful in the training of neural networks performing various tasks not only on single volumes but may take into account multiple modalities or entire MRI acquisition schemes may perform fibre bundle segmentations with high accuracy on healthy subjects a suitable training dataset consisting of pathology-related FOD image acquisitions supplemented by synthetic data could improve performance such as using data on tractograms of patients with brain tumours Diffusion MRI GAN may also play a role in mitigating data imbalance as it can generate additional data for the underrepresented groups which can lead to more balanced and accurate models it could also show its value for medical training and the teaching of data analysis especially if data privacy protection restricts access One may also think of synthetic MRI sessions serving as a control group for a clinical study with certain age and gender distributions to match a patient cohort or the use of synthetic data for a preliminary study in which the effect size and the required number of patients can be estimated synthetic FOD data allow for rigorous testing of algorithms and methodologies in a controlled environment thus leading to the development of more accurate and robust analysis techniques for clinical applications we consider FODs represented by spherical harmonics up to lmax = 6 corresponding to 28 three-dimensional volumes per subject We split the data in a training set containing 965 FOD and a validation set with 100 FOD image data We artificially augmented the training data pool by four slightly rotated versions of each data set We used Gaussian distributed random angles with zero mean and a standard deviation of 1° and spatially rotated the volumes around each main axes we also added a version in which we performed three subsequent rotations with independent Gaussian distributed random angles with zero mean and a standard deviation of 1° The network model used in this paper is the combination of the Wasserstein GAN (red path) and a variational autoencoder (blue path) we reduce the output channels from layer to layer The Discriminator is a classifier that aims to distinguish real FODs from the training data from generated FODs. This classification is represented by a single real number output. It consists of six 3D convolutional layers with leaky ReLU activation functions (see Table 2 for all details) After layer 2–5 a batch normalisation is performed The Encoder projects the FODs of the training set into a vector in the latent space ze = E(xtrain) that has the same dimensionality as zr. Except the output of its last layer, its implementation is same as the Discriminator (see Table 3 for details) Note that the Generator also works as the decoder in a VAE such that xdec = G(ze) ≈ xtrain) The Code Discriminator (CD) classifies data as Gaussian distributed random variables and is trained by zr. It ensures that the encoded training data ze follow a continuous Gaussian distribution. It consists of three linear layers. After layer one and two, batch normalisations are performed and leaky ReLUs are used as activation functions (see Table 4) the first loss function contains the Wasserstein distances used to train the Generator and the Encoder and the L1 distance for FODs from the training data xtrain and their corresponding decoded versions xdec = G(E(xreal)) where the angular brackets denote the averages over the batches the generator and the encoder have no effect on it we drastically increased the influence of the VAE by setting λ = 50 000 in order to avoid mode collapse of the generated data To train the Discriminator and the code discriminator respectively, where ϕD and ϕCD are the gradient penalty terms (see also ref. 48) where \(\hat{x}=\alpha {x}_{{{{{\rm{train}}}}}}+(1-\alpha ){x}_{{{{{\rm{syn}}}}}}\) and \(\tilde{x}=\beta {x}_{{{{{\rm{train}}}}}}+(1-\beta ){x}_{{{{{\rm{dec}}}}}}\) the angular brackets denote the average over the batch ϕCD is calculated similarly: \({\phi }_{CD}=\langle {(\parallel \!\!{\nabla }_{\hat{z}}CD(\hat{z}){\parallel }_{{L}_{2}}-1)}^{2}\rangle\) with \(\hat{z}=\gamma {z}_{e}+(1-\gamma ){z}_{r}\) Here γ is a uniformly distributed random number that is drawn for each batch component we first update LG,E and perform one optimiser step for E and two for G we perform four loops of updating LD and perform an optimiser step for D we update LC and perform the update step for CD we trained our network for 37,000 epochs with a computation time of seven days We restructured the streamlines of the tracts into 100 equidistant points and sampled the fibre densities along these tracts We also tested our data for a regionally specific bias by determining the mean and standard deviations of the fibre densities in AAL parcels we calculated the relative differences in mean (\(\overline{fd}\)) and standard deviations (σ) We conducted the experiments at our HPC environment of the Charité - Universitätsmedizin Berlin The model was trained on an NVIDIA DGX A100 80G system 32TB local scratch space and 8x NVIDIA A100 80G GPUs Both the frontend and the computing nodes run CentOS 8.3 We used Python 3.10 with the Pytorch deep learning library Training required around 7 days of computation time Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article Automated lesion detection on MRI scans using combined unsupervised and supervised methods Automated segmentation of tissues using ct and MRI: a systematic review A review of deep learning with special emphasis on architectures Synthetic data augmentation using gan for improved liver lesion classification IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) 289–293 (IEEE Synthetic data in machine learning for medicine and healthcare Synthetic data as an enabler for machine learning applications in medicine Generative adversarial networks: an overview Gan-based synthetic brain MR image generation IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) 734–738 (IEEE Bermudez, C. et al. Learning implicit brain MRI manifolds with deep learning. In Medical Imaging 2018: Image Processing (eds Angelini, E. D. & Landman, B. A.) Vol. 10574, 408–414. International Society for Optics and Photonics (SPIE, 2018). https://doi.org/10.1117/12.2293515 Brain imaging generation with latent diffusion models In MICCAI Workshop on Deep Generative Models 117–126 (Springer 4d semantic cardiac magnetic resonance image synthesis on xcat anatomical model In Medical Imaging with Deep Learning 6–18 (PMLR Paired conditional generative adversarial network for highly accelerated liver 4D MRI White matter tractography for neurosurgical planning: a topography-based review of the current state of the art Sporns, O., Tononi, G. & Kötter, R. The human connectome: a structural description of the human brain. PLOS Comput. Biol. 1, https://doi.org/10.1371/journal.pcbi.0010042 (2005) Clinical applications of magnetic resonance imaging based functional and structural connectivity Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution Investigating the prevalence of complex fiber configurations in white matter tissue with diffusion magnetic resonance imaging Lesion-symptom mapping of language impairments in patients suffering from left perisylvian gliomas Tractography from T1-weighted MRI: empirically exploring the clinical viability of streamline propagation without diffusion MRI Generation of 3D brain MRI using auto-encoding generative adversarial networks In International Conference on Medical Image Computing and Computer-Assisted Intervention 118–126 (Springer Variational approaches for auto-encoding generative adversarial networks Complex network measures of brain connectivity: uses and interpretations Reproducibility and intercorrelation of graph theoretical measures in structural brain connectivity networks The detection of disease clustering and a generalized regression approach flexible and open software framework for medical image processing and visualisation Robust determination of the fibre orientation distribution in diffusion MRI: non-negativity constrained super-resolved spherical deconvolution Mri to CT synthesis of the lumbar spine from a pseudo-3D cycle gan IEEE 17th International Symposium on Biomedical Imaging (ISBI) 1784–1787 (IEEE A review on medical imaging synthesis using deep learning and its clinical applications The minimal preprocessing pipelines for the human connectome project Towards deeper understanding of variational autoencoding models Variational autoencoder for reference based image super-resolution IEEE/CVF Conference on Computer Vision and Pattern Recognition 516–525 (IEEE Hierarchical amortized gan for 3d high resolution medical image synthesis A 3d wide residual network with perceptual loss for brain MRI image denoising 10th International Conference on Computing Communication and Networking Technologies (ICCCNT) 1–7 (IEEE Ai supported detection of cerebral multiple sclerosis lesions decreases radiologic reporting times Brain tumor segmentation of MRI images: a comprehensive review on the application of artificial intelligence tools The intelligent imaging revolution: artificial intelligence in MRI and MRS acquisition and reconstruction Gan-based synthetic medical image augmentation for increased CNN performance in liver lesion classification Generative adversarial networks in medical image augmentation: a review Tractseg-fast and accurate white matter tract segmentation Probabilistic diffusion tractography with multiple fibre orientations: what can we gain The wu-minn human connectome project: an overview How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging A novel iterative approach to reap the benefits of multi-tissue CSD from just single-shell (+ b= 0) diffusion mri data Advances in neural information processing systems 30 (MIT press Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain Sift2: enabling dense quantitative assessment of brain white matter connectivity using streamlines tractography Measures of the amount of ecologic association between species A method of establishing groups of equal amplitude in plant sociology based on similarity of species content and its application to analysis of vegetation on Danish commons Apparent fibre density: a novel measure for the analysis of diffusion-weighted magnetic resonance images Download references Open Access funding enabled and organized by Projekt DEAL Department of Neuroscience and Biomedical Engineering Dental Sciences and Morphological and Functional Images L.S.F.; Project administration: L.S.F.; Supervision: L.S.F.; Writing—original draft: S.V. A.C.; Writing—review & editing: L.S.F. Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work Primary Handling Editor: Benjamin Bessieres Download citation DOI: https://doi.org/10.1038/s42003-025-07936-w Metrics details Missing pixel imputation is a critical task in image processing where the presence of high percentages of missing pixels can significantly degrade the performance of downstream tasks such as image segmentation and object detection This paper introduces a novel approach for missing pixel imputation based on Generative Adversarial Networks (GANs) We propose a new GAN architecture incorporating an identity module and a sperm motility-inspired heuristic during filtration to optimize the selection of pixels used in reconstructing missing data The intelligent sperm motility heuristic navigates the image’s pixel space identifying the most influential neighboring pixels for accurate imputation Our approach includes three essential modifications: (1) integration of an identity module within the GAN architecture to mitigate the vanishing gradient problem; (2) introduction of a metaheuristic algorithm based on sperm motility to select the top 10 pixels that most effectively contribute to the generation of the missing pixel; and (3) the implementation of an adaptive interval mechanism between the discriminator’s actual value and the weighted average of the selected pixels enhancing the generator’s efficiency and ensuring the coherence of the imputed pixels with the surrounding image context We evaluate the proposed method on three distinct datasets (Energy Images demonstrating its superior performance in maintaining pixel integrity during the imputation process Our experiments also confirm the approach’s effectiveness in addressing everyday challenges in GANs such as mode collapse and vanishing gradients This problem is exacerbated in deep architectures where the information from the loss function fails to propagate effectively back through the network layers Addressing this problem is essential for improving GANs’ performance in tasks like missing pixel imputation This phenomenon is particularly detrimental in missing pixel imputations where diverse and contextually accurate pixel generation is crucial a key focus in advancing GAN-based imputation methods This paper proposes a novel GAN architecture that integrates an identity module designed to tackle the vanishing gradient problem The identity module ensures that the gradients are preserved across the network layers facilitating more effective learning and improving the overall quality of the generated pixels This architectural modification enhances the robustness of the GAN and contributes to solving the challenges associated with deep networks in pixel imputation tasks A unique aspect of our approach is incorporating a metaheuristic algorithm inspired by sperm motility which guides the selection of pixels used in the imputation process The sperm motility heuristic simulates the natural movement of sperm cells known for their ability to navigate complex environments to reach their target our algorithm identifies the most relevant and influential pixels that should be used to reconstruct missing data thereby enhancing the accuracy and efficiency of the imputation process In addition to the identity module and sperm motility heuristic we introduce a third innovation: an adaptive interval mechanism This mechanism creates a dynamic interval between the actual value of the discriminator and the weighted average of the selected pixels The adaptive interval plays a crucial role in improving the generator’s efficiency by reducing the time required for pixel generation while ensuring that the generated pixels are coherent with the surrounding image context This mechanism speeds up the imputation process and enhances the overall quality of the imputed images To validate the effectiveness of our proposed approach we conducted extensive experiments on three distinct datasets each representing different types of image data and varying levels of complexity Our results demonstrate significant improvements in pixel integrity and a marked reduction in common GAN-related issues such as mode collapse and vanishing gradients The experimental outcomes provide strong evidence of the robustness and reliability of our method in addressing the challenges of missing pixel imputation Introduces a novel sperm motility-inspired heuristic that emulates the natural movement patterns of sperm cells This innovative approach facilitates intelligent navigation through pixel space identifying the most influential neighboring pixels The strategy significantly enhances the contextual coherence of imputed pixels Its versatility across diverse image processing applications further underscores its potential impact and broad applicability The proposed methodology effectively addresses persistent challenges in Generative Adversarial Networks (GANs) such as mode collapse and the vanishing gradient problem through specific architectural modifications This results in improved diversity and accuracy in image reconstruction compared to other GAN architectures Enhances the accuracy and computational efficiency of missing pixel imputation ensuring the high integrity of imputed pixels about the overall image structure thereby facilitating reliable downstream analysis Comprehensive testing and case studies demonstrate the efficacy of each methodology component including the sperm motility heuristic and the identity block providing empirical evidence of their contributions to the proposed approach’s overall performance and efficiency The organization of this paper is structured as follows: Sect 2 presents a detailed review of related work examining existing approaches for missing pixel imputation and the application of Generative Adversarial Networks (GANs) in image processing such as the vanishing gradient problem and mode collapse Section 3 outlines the proposed methodology and the adaptive interval mechanism designed to enhance imputation accuracy and efficiency Section 4 discusses the experimental results comparing the performance of our approach with existing methods and analyzing its effectiveness in solving the identified challenges and implications and offering insights into potential future research directions and applications of the proposed approach Generative Adversarial Networks (GANs) have become a cornerstone in image processing particularly in tasks requiring high-quality image synthesis This section reviews the most relevant advancements in GAN-based missing data imputation from 2021 to 2024 highlighting key contributions and identifying gaps our work aims to address incorporated domain-specific loss functions tailored to preserve anatomical correctness MedGAN struggled with generalizability when applied to non-medical datasets particularly those with complex textures and patterns a new missing data imputation method based on Generative Adversarial Networks (GAN) The process aims to overcome challenges such as gradient vanishing and mode collapse common in existing GAN-based imputation methods The new method improves the accuracy of missing data imputation by employing a least squares loss function to address gradient vanishing and a dual discriminator to prevent mode collapse The authors stated that MGAIN outperformed cutting-edge methods lowering the root mean square error by 21.66% The issue of missing data in multi-view datasets is addressed by Zhang et al.20 with a novel technique called VIGAN which uses Generative Adversarial Networks (GANs) This technique is beneficial when a whole data view is missing for specific samples Conventional methods such as matrix completion and multiple imputations are useless VIGAN treats each view as a distinct domain and then reconstructs the missing view using a multi-modal denoising autoencoder (DAE) Benchmark datasets and a genetic study validate the approach demonstrating its usefulness in the life sciences and its ability to recover missing data The method uses seismic layer velocity as a constraint to guide CGAN in producing realistic well-log data corresponding to geological features the swarm intelligence algorithm optimizes the generation process The approach outperforms other algorithms and provides a novel solution for sequence data prediction in oil and gas development The proposed method employs a generative adversarial architecture to generate missing modalities from any combination of existing ones using a single model It introduces a Commonality- and Discrepancy-Sensitive Encoder that exploits modality-invariant and specific information resulting in anatomically consistent and realistic images a Dynamic Feature Unification Module is intended to robustly integrate a variety of available modalities while preventing information loss The method is validated on public datasets and outperforms existing approaches hinders the practical training of GANs by diminishing gradient updates during backpropagation restricts the diversity of generated outputs leading to repetitive and less varied imputation results Our proposed approach addresses these limitations through several innovative strategies we introduce an identity block within the GAN architecture to mitigate the vanishing gradient problem ensuring better gradient flow and more stable training This approach is particularly novel in its application to missing pixel imputation as it allows for more effective learning in deep networks Our method also incorporates a metaheuristic algorithm inspired by sperm motility which intelligently selects the most influential neighboring pixels for imputation This biologically inspired approach enhances the accuracy and context-awareness of the imputed pixels setting it apart from traditional heuristic methods We also introduce an adaptive interval mechanism that dynamically adjusts the interval between the discriminator’s actual value and the weighted average of selected pixels This mechanism accelerates the imputation process and ensures that the imputed pixels are coherent with the surrounding image context reducing artifacts and improving overall image quality While our method significantly improves the imputation process The integration of a sperm motility-inspired metaheuristic introduces additional computational complexity which may limit its scalability for large-scale datasets while effective in maintaining pixel coherence the adaptive interval mechanism requires careful calibration to avoid overfitting particularly in datasets with high variability The core of the proposed approach is a Generative Adversarial Network (GAN) architecture which serves as the framework for the missing pixel imputation task The GAN model consists of two key components: the Generator and the Discriminator The Generator network is responsible for producing the imputed pixel values to fill in the missing regions of the input image and a crucial innovation introduced in this work This novel architectural component addresses the vanishing gradient problem a common issue encountered in deep neural networks By integrating the Identity block within the Generator the paper aims to improve the gradient flow throughout the network ensuring more stable and effective training of the GAN model the Generator network also utilizes a metaheuristic approach inspired by sperm motility to guide the selection of the most influential neighboring pixels during the imputation process This biologically inspired technique mimics the natural movement patterns of sperm cells which are known for their ability to navigate complex environments to reach their target the algorithm identifies the most relevant and impactful pixels to be used in the reconstruction of missing data enhancing the accuracy and contextual coherence of the imputed pixels The Open Energy Images dataset is an extensive collection of over 240,000 annotated images covering various energy infrastructures and technologies The NREL Solar Images dataset provides over 4,000 categorized and labeled images focused explicitly on solar photovoltaic systems the NREL Wind Turbine Dataset contains around 5,000 images of wind turbines with annotated bounding boxes these three datasets offer a comprehensive visual resource for researchers and practitioners working on energy-related computer vision and machine learning applications enabling studying different energy generation and utilization systems through high-quality and accessibility of these datasets make them valuable tools for advancing research and development in the energy domain The sperm motility-inspired metaheuristic is a unique and innovative component of the proposed approach for missing pixel imputation This metaheuristic algorithm draws inspiration from the natural movement patterns of sperm cells which are known for their remarkable ability to navigate complex environments to reach their target destination In the context of the missing pixel imputation task the sperm motility-inspired metaheuristic is used to intelligently select the most influential neighboring pixels that should be used to reconstruct the missing pixel values as the quality and coherence of the imputed pixels directly depend on selecting the appropriate neighboring pixels The core idea behind the sperm motility-inspired metaheuristic is to mimic the way sperm cells navigate through their environment Sperm cells are guided by various chemical and physical cues which allow them to efficiently navigate towards the sperm the proposed metaheuristic algorithm leverages the inherent patterns and relationships within the image data to identify the most influential neighboring pixels for the imputation process The process begins with the Identification of Neighboring Pixels the algorithm first identifies the neighboring pixels that could potentially contribute to imputing the missing value This involves examining the pixel’s spatial vicinity and considering the pixels close The Evaluation of Pixel Influence takes place The metaheuristic evaluates the influence of each neighboring pixel based on a set of criteria that mimic the navigation strategies of sperm cells These criteria may include factors such as the spatial proximity of the adjacent pixel and its structural similarity to the surrounding pixels The algorithm assigns a weight to each neighboring pixel based on how well it aligns with these sperm-inspired evaluation criteria sperm mobility metaheuristic algorithm flow chart. Motility Attitude During Filtration approach The pseudocode’s architecture is based on a GAN framework enhanced with an adaptive interval mechanism and a biologically inspired pixel selection process. The generator and discriminator form the core components of the GAN. The generator produces imputed pixel values based on input from selected neighboring pixels. In contrast, the discriminator evaluates these generated pixels against real ones to guide the generator’s learning process. A key innovation in this architecture is the integration of the Motility Attitude During the Filtration approach, which intelligently selects neighboring pixels to influence the imputation. This selection process is critical as it ensures that the pixels chosen for imputing the missing ones are contextually relevant and contribute meaningfully to the overall image quality. to evaluate the similarity between the feature representations of the generated and real The FID considers both the mean and covariance of the feature distributions providing a holistic assessment of how closely the reconstructed images match the characteristics of the original data these complementary metrics allow us to thoroughly evaluate the efficacy of our proposed generative model for the critical task of missing pixel imputation This section of the paper presents the comparison between GSIP-GAN and Wasserstein GAN (WGAN), Spectral Normalized GAN (SNGAN), Progressive Growing of GANs (PGGAN), Cycle-Consistent GANs (CycleGAN), and Spatially Adaptive Normalization (SPADE) using three different metrices for the imputation process. Table 2 showcases the Root Mean Squared Error (RMSE) comparison which is a metric used to evaluate the accuracy of image reconstruction The results indicate that the proposed “Our” model outperforms the other GAN models consistently achieving the lowest RMSE values across all three datasets This suggests that the “Our” model is highly effective in accurately reconstructing the missing or corrupted pixels in the energy-related images RMSE Comparison of five Models and GSIP-GAN Across three Datasets In Table 3 the RSNR (Relative Signal-to-Noise Ratio) comparison reveals the performance of various GAN models across three datasets: Open Energy Images The results indicate that “Our model” achieves the highest RSNR values highlighting its superior capability in preserving signal integrity the traditional GAN model shows significantly lower RSNR values particularly with an RSNR of 50.43 for Open Energy Images indicating a substantial loss of signal quality RSNR Comparison of five Models and GSIP-GAN Across three Datasets In Table 4 the SSIM (Structural Similarity Index Measure) comparison assesses the performance of various GAN models across three datasets: Open Energy Images The results show that “Our model” achieves the highest SSIM scores This indicates a superior ability to preserve the generated outputs’ structural information and visual quality particularly with a value of 61.32 for Open Energy Images highlighting its limitations in maintaining structural fidelity with WGAN achieving 72.32 for Open Energy Images and 70.11 for NREL Solar Images These results demonstrate that while these models perform better than the traditional GAN they still fall short of the performance achieved by “Our model.” The ablation tests further elucidate the contributions of specific enhancements The GAN with the identity block matches the WGAN’s performance on Open Energy Images and also shows improved results for the other datasets SSIM Comparison of five Models and GSIP-GAN Across three Datasets This part of the results shows the comparison between GSIP-GAN and Wasserstein GAN (WGAN) and Spatially Adaptive Normalization (SPADE) based on FID and IS to measure the diversity of generation of the missing pixel In Table 5 the FID (Fréchet Inception Distance) comparison evaluates the diversity of generated images across three datasets: Open Energy Images A lower FID score indicates better quality and diversity in the generated samples “Our model” demonstrates superior performance with the lowest FID scores of 178.23 for Open Energy Images signifying its ability to generate diverse and high-quality outputs the traditional GAN model exhibits relatively high FID scores particularly 253.87 for Open Energy Images and 276.09 for NREL Solar Images indicating significant challenges in producing diverse and realistic images with WGAN achieving 232.32 for Open Energy Images and 193.43 for NREL Solar Images These results suggest that while these models perform better than the traditional GAN they still lack the diversity and quality exhibited by “Our model.“ FID Comparison of five Models and GSIP-GAN Across three Datasets IS Comparison of five Models and GSIP-GAN Across three Datasets Performance metrics bar chart before using GSIP model for image reconstruction. Performance Metrics bar chart After Using GSIP Model for Image Reconstruction. Examples of solar panels fault detection using Swin Transformer before pixel imputation. Examples of solar panels fault detection using Swin Transformer after pixel imputation This paper introduced a novel GAN-based approach for missing pixel imputation that addresses key challenges in GAN training and pixel selection The proposed method integrates three key innovations: an identity module to mitigate the vanishing gradient problem a sperm motility-inspired metaheuristic algorithm to optimize pixel selection and an adaptive interval mechanism to enhance the generator’s efficiency and coherence of the imputed pixels Extensive experiments on diverse datasets demonstrated the superior performance of the proposed method in maintaining pixel integrity and addressing common GAN issues like mode collapse The integration of biological inspiration through the sperm motility heuristic combined with the architectural enhancements enabled the system to generate high-quality contextually accurate pixel imputations even in the presence of high percentages of missing data The contributions of this work extend beyond just pixel imputation as the proposed techniques have the potential to benefit a wide range of image processing tasks where robust reconstruction of missing or corrupted data is crucial The proposed GAN-based approach for missing pixel imputation has demonstrated promising results but there are several avenues for future research and improvement and investigating alternative biological inspirations could further enhance the method’s adaptability Adapting the framework to handle missing pixel imputation in video sequences could enable its application to a broader range of multimedia processing tasks such as video restoration and anomaly detection like depth information or semantic context could potentially improve the accuracy and robustness of the pixel imputation process Investigating optimization techniques or hardware-specific implementations to improve the computational efficiency of the inference process could make the proposed method more suitable for real-time or resource-constrained applications Exploring other biologically inspired mechanisms and their integration into the GAN framework could lead to new insights and advancements in image processing The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request Shelhamer, E., Long, J. & Darrell, T. Fully Convolutional Networks for Semantic Segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 39, 640–651. https://doi.org/10.1109/TPAMI.2016.2572683 (2017) in IEEE Conference on Computer Vision and Pattern Recognition (CVPR) in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 Frangi) 234–241 (Springer International Publishing) Koide-Majima, N., Nishimoto, S. & Majima, K. Mental image reconstruction from human brain activity: Neural decoding of mental imagery via deep neural network-based bayesian estimation. Neural Netw. 170, 349–363. https://doi.org/10.1016/j.neunet.2023.11.024 (2024) Seminal Graphics Papers: Pushing the BoundariesVolume 2 (Vol Volume 2 Article 65 (Association for Computing Machinery Telea, A. An image inpainting technique based on the fast marching method. J. Graphics Tools. 9, 23–34. https://doi.org/10.1080/10867651.2004.10487596 (2004) in Proceedings of the 27th annual conference on Computer graphics and interactive techniques 417–424 (ACM Press/Addison-Wesley Publishing Co. Criminisi, A., Perez, P. & Toyama, K. Region filling and object removal by exemplar-based image inpainting. IEEE Trans. Image Process. 13, 1200–1212. https://doi.org/10.1109/TIP.2004.833105 (2004) Volume 2 Article 71 (Association for Computing Machinery Goodfellow, I. et al. Generative adversarial networks. Commun. ACM. 63, 139–144. https://doi.org/10.1145/3422622 (2020) in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Bengio, Y., Simard, P. & Frasconi, P. Learning long-term dependencies with gradient descent is difficult. IEEE Trans. Neural Networks. 5, 157–166. https://doi.org/10.1109/72.279181 (1994) in IEEE International Conference on Computer Vision (ICCV) in Proceedings of the 30th International Conference on Neural Information Processing Systems 2234–2242 (2016) Armanious, K. et al. MedGAN: Medical image translation using GANs. Comput. Med. Imaging Graph. 79, 101684. https://doi.org/10.1016/j.compmedimag.2019.101684 (2020) Qin, X., Shi, H., Dong, X., Zhang, S. & Yuan, L. Improved generative adversarial imputation networks for missing data. Appl. Intell. 54, 11068–11082. https://doi.org/10.1007/s10489-024-05814-2 (2024) IEEE International Conference on Big Data (Big Data) Qu, F. et al. A novel well log data imputation methods with CGAN and swarm intelligence optimization. Energy 293, 130694. https://doi.org/10.1016/j.energy.2024.130694 (2024) Zhang, Y. et al. Unified multi-modal image synthesis for missing modality imputation. IEEE Trans. Med. Imaging. 1–1. https://doi.org/10.1109/TMI.2024.3424785 (2024) Sajeeda, A. & Hossain, B. M. M. Exploring generative adversarial networks and adversarial training. Int. J. Cogn. Comput. Eng. 3, 78–89. https://doi.org/10.1016/j.ijcce.2022.03.002 (2022) Tufail, S., Riggs, H., Tariq, M. & Sarwat, A. I Advancements and challenges in Machine Learning: a comprehensive review of models, libraries, applications, and algorithms. Electronics 12, 1789. https://doi.org/10.3390/electronics12081789 (2023) in GANs for Data Augmentation in Healthcare (eds Arun Solanki & Mohd Naved) 21–40Springer International Publishing in 2020 IEEE 4th Conference on Information & Communication Technology (CICT) 563–574 (Springer International Publishing) Kim, A. et al. Aerial imagery dataset of lost oil wells. Sci. Data. 11, 1005. https://doi.org/10.1038/s41597-024-03820-0 (2024) Hartono, N. T. P. et al. Stability follows efficiency based on the analysis of a large perovskite solar cells ageing dataset. Nat. Commun. 14, 4869. https://doi.org/10.1038/s41467-023-40585-3 (2023) Zhou, J. et al. SDWPF: a dataset for spatial dynamic wind power forecasting over a large turbine array. Sci. Data. 11, 649. https://doi.org/10.1038/s41597-024-03427-5 (2024) in International conference on machine learning Spectral normalization for generative adversarial networks Progressive Growing of GANs for Improved Quality in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Yang, R. et al. CNN-LSTM deep learning architecture for computer vision-based modal frequency detection. Mech. Syst. Signal Process. 144, 106885. https://doi.org/10.1016/j.ymssp.2020.106885 (2020) An image is worth 16x16 words: Transformers for image recognition at scale 833–851 (Springer International Publishing) in 2021 IEEE/CVF International Conference on Computer Vision (ICCV) YOLOv6: A single-stage object detection framework for industrial applications Download references Open access funding provided by The Science Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB) Computer Engineering and Systems Department The authors contributed collaboratively to the paper by developing a novel GAN architecture for missing pixel imputation Mahmoud spearheaded the design of the identity module to mitigate the vanishing gradient problem while Wael Said focused on the sperm motility-inspired metaheuristic for pixel selection and conducted experimental evaluations Fadel assisted in formulating the methodology and assessing performance metrics and Mostafa Elbaz coordinated the research efforts and enhanced the manuscript’s clarity Their combined efforts led to a robust approach for addressing missing pixel imputation challenges Download citation DOI: https://doi.org/10.1038/s41598-024-82242-9 Metrics details The photoelectrochemical synthesis of valuable multicarbon products from carbon dioxide sunlight and water is a promising pathway for clean energy generation and carbon neutrality it is challenging to create and stabilize efficient C–C coupling sites to achieve multicarbon products with high selectivity Here we designed a low-coordinated copper-cluster catalyst interfacially coupled in situ with a GaN nanowire photocathode achieving a high ethylene Faradaic efficiency of ∼61% and a partial current density of 14.2 mA cm−2 The in situ self-optimized Ga–N–O interface was confirmed to facilitate and stabilize the interfacially oxidized copper species of copper clusters which function as efficient C–C coupling sites for ethylene production the hydrogen-feeding effect of GaN for promoting CO hydrogenation also guides the facile CHO-involved C–C coupling pathway This work sheds light on the interface design and understanding of efficient and stable (photo)electrosynthesis of highly valuable fuels from CO2 The data supporting the findings of this study are available within the article and its Supplementary Information files Light-driven synthesis of C2H6 from CO2 and H2O on a bimetallic AuIr composite supported on InGaN nanowires Molecularly engineered photocatalyst sheet for scalable solar formate production from carbon dioxide and water CO2 reduction using water as an electron donor over heterogeneous photocatalysts aiming at artificial photosynthesis An overview of solar-driven photoelectrochemical CO2 conversion to chemical fuels Strategies to improve light utilization in solar fuel synthesis Finding the way to solar fuels with dye-sensitized photoelectrosynthesis cells Binary molecular–semiconductor p–n junctions for photoelectrocatalytic CO2 reduction Bias-free solar syngas production by integrating a molecular cobalt catalyst with perovskite–BiVO4 tandems Photoelectrochemical conversion of carbon dioxide (CO2) into fuels and value-added products Sn-coupled p-Si nanowire arrays for solar formate production from CO2 Aqueous CO2 reduction on Si photocathodes functionalized by cobalt molecular catalysts/carbon nanotubes Bias-free solar-driven syngas production: a Fe2O3 photoanode featuring single-atom cobalt integrated with a silver–palladium cathode Highly efficient binary Cu–iron catalyst for photoelectrochemical carbon dioxide reduction toward methane Photoelectrochemical CO2 reduction into syngas with the metal/oxide interface Photocatalytic and photoelectrochemical carbon dioxide reductions toward value-added multicarbon products Photoelectrochemical CO2 reduction toward multicarbon products with silicon nanowire photocathodes interfaced with cu nanoparticles Codesign of an integrated metal–insulator–semiconductor photocathode for photoelectrochemical reduction of CO2 to ethylene Solar-driven liquid multi-carbon fuel production using a standalone perovskite–BiVO4 artificial leaf Investigation and mitigation of degradation mechanisms in Cu2O photoelectrodes for CO2 reduction to ethylene Design principles for efficient and stable water splitting photoelectrocatalysts Interfacial engineering of metal–insulator–semiconductor junctions for efficient and stable photoelectrochemical water oxidation Quantifying losses and assessing the photovoltage limits in metal–insulator–semiconductor water splitting systems The role of in situ generated morphological motifs and Cu(I) species in C2+ product selectivity during CO2 pulsed electroreduction Tailored catalyst microenvironments for CO2 electroreduction to multicarbon products on copper using bilayer ionomer coatings Progress and perspectives of electrochemical CO2 reduction on Cu in aqueous electrolyte Advances and challenges in understanding the electrocatalytic conversion of carbon dioxide to fuels Stability of residual oxides in oxide-derived copper catalysts for electrochemical CO2 reduction investigated with 18O labeling Towards molecular understanding of local chemical environment effects in electro- and photocatalytic CO2 reduction Regulating the oxidation state of nanomaterials for electrocatalytic CO2 reduction Dopant-induced electron localization drives CO2 reduction to C2 hydrocarbons Highly electrocatalytic ethylene production from CO2 on nanodefective Cu nanosheets Experimental evidence of distinct sites for CO2-to-CO and CO conversion on Cu in the electrochemical CO2 reduction reaction Highly active and stable stepped Cu surface for enhanced electrochemical CO2 reduction to C2H4 Efficient electrocatalytic conversion of carbon monoxide to propanol using fragmented Cu Operando studies reveal active Cu nanograins for CO2 electroreduction Robust and synthesizable photocatalysts for CO2 reduction: a data-driven materials discovery On factors limiting the performance of photoelectrochemical CO2 reduction Recent advances in solar-driven carbon dioxide conversion: expectations versus reality Development of a photoelectrochemically self-improving Si/GaN photocathode for efficient and durable H2 production Oxynitrides enabled photoelectrochemical water splitting with over 3,000 hrs stable operation in practical two-electrode configuration Structure of the catalytically active Cu–ceria interfacial perimeter Electrochemical CO2 reduction to ethylene by ultrathin CuO nanoplate arrays X-ray photoelectron spectroscopy and X-ray diffraction study of the thermal oxide on gallium nitride Tailoring *H intermediate coverage on the CuAl2O4/CuO catalyst for enhanced electrocatalytic CO2 reduction to ethanol Single atom Bi decorated copper alloy enables C–C coupling for electrocatalytic reduction of CO2 into C2+ products A fundamental viewpoint on the hydrogen spillover phenomenon of electrocatalytic hydrogen evolution Electrocatalytic reduction of CO2 to ethylene and ethanol through hydrogen-assisted C–C coupling over fluorine-modified Cu Stabilizing Cu2+ ions by solid solutions to promote CO2 electroreduction to methane Time-resolved observation of C–C coupling intermediates on Cu electrodes for selective electrochemical CO2 reduction Electrocatalytic reduction of CO2 to ethanol at close to theoretical potential via engineering abundant electron-donating Cuδ+ species Copper/alkaline earth metal oxide interfaces for electrochemical CO2-to-alcohol conversion by selective hydrogenation Download references The authors gratefully acknowledge support from the United States Army Research Office Award (grant no The authors also acknowledge technical support from the Lurie Nanofabrication Facility and the Michigan Center for Materials Characterization These authors contributed equally: Bingxing Zhang designed the catalyst system and carried out (photo)electrochemical experiments participated in discussion of the theoretical calculations prepared the n+–p silicon solar cell and performed MBE growth of n+-GaN nanowires on silicon participated in data analysis and discussion All the authors contributed to the overall scientific discussions and edited the paper Some intellectual property related to the synthesis of GaN nanowires was licensed to NS Nanotech have a financial interest in these companies The other authors declare no competing interests Nature Synthesis thanks Longhua Li and the other reviewer(s) for their contribution to the peer review of this work in collaboration with the Nature Synthesis team Download citation DOI: https://doi.org/10.1038/s44160-024-00648-9 Transformers are a foundational technology underpinning many advances in large language models (LLMs), such as generative pre-trained transformers (GPTs). They're now expanding into multimodal AI applications such as large vision models capable of correlating content as diverse as text audio and robot instructions across numerous media types more efficiently than techniques like GANs can GANs and transformers can also be combined in various ways to generate content from a prompt target adjustments to existing content or interpret content Let's explore the beginnings of each technique their use cases and how researchers are combining the two techniques into various transformer-GAN combinations GANs were introduced in 2014 by Ian Goodfellow and associates to generate realistic-looking numbers and faces They combine the following two neural networks: while the other one vetoes them if they don't come close to the authentic images from a given data set researchers are exploring ways to use other neural network models Transformers were introduced by a team of Google researchers in 2017 who were looking to build a more efficient translator. In a paper entitled "Attention Is All You Need," the researchers laid out a new technique to discern the meaning of words based on how they characterized other words in phrases Earlier tools to interpret text frequently used one neural network to translate words into vectors using a previously constructed dictionary alongside another neural network to process a sequence of text, such as a recurrent neural network (RNN) transformers essentially learn to interpret the meaning of words directly from processing large bodies of unlabeled text The same approach can also be used to identify patterns in other kinds of data This lets researchers scale the LLMs driving recent advances -- and publicity -- in the field Transformers can also find relationships between words that are far apart Transformers also play an essential role in the development of multimodal AI that combines multiple modalities of data the attention mechanism can find relationships and connections across multiple modalities of data hackers constantly introduce new inputs to fool fraud detection algorithms GANs tend to be better at adapting to and protecting against these kinds of techniques where a set of potential responses must be evaluated with respect to the conditional sequence of inputs One limitation of transformers is that they're not as computationally, memory or data efficient as GANs. Compared to GANs, transformer models require considerably more IT resources to train and run; this is known as inference GANs are a better option for generating synthetic data based on existing data sets There's also active research into combining GANs and transformers into so-called GANsformers The idea is to use a transformer to provide an attentional reference so the generator can increase the use of context to enhance content "The intuition behind GANsformers is that human attention is based on the specific local features of an object of interest in addition to the latent global characteristics," Searle explained The resulting improved representations are more likely to simulate both the global and local features a human might perceive in an authentic sample such as a realistic face or computer-generated audio consistent with a human voice's tone and rhythm GANsformers could also help map linguistic concepts to specific areas of adjustments or types of adjustments to photos, videos or 3D models. In addition, they could help fill in or improve the rendering of details in generated images. For example, they could make sure humans have five fingers as opposed to the different numbers of fingers often found in many deepfakes They could also fill in local features to create better deepfakes such as the subtle color changes in skin tones associated with increased blood flow that aren't noticed by humans but are used to detect fake video content Transformers are growing in awareness thanks to their role in popular tools, such as ChatGPT But transformers won't necessarily replace GANs for all applications the inclusion of idiosyncratic errors or stylistic traits could mask the true origin of an AI-derived output Conversely, improved realism might be problematic with deepfakes used to launch cyberattacks, damage brands or spread fake news. In these cases, GANsformers could provide better filters to detect deepfakes "The use of adversarial training and contextual evaluation could produce AI systems able to provide enhanced security improved content filtering and defense against misinformation attacks using generative botnets," Searle said But Zidaritz believes transformers can potentially edge out GANs in many use cases since they can be applied to text and images more easily but their applications will be more limited than that of GPTs," he said "It is also likely that we will see more GAN-like transformers and transformer-like GANs in both of which the transformer with its self-attention or its self-similarity mechanism will be central." It's important to note that there has been considerably more research and development on multimodal LLMs which can sometimes achieve content-generation results like those of GANs They also make it easier to translate human intent into results While they might not be as computationally efficient as GANs considerably higher levels of investment in LLM research is closing the gap Other promising machine learning techniques for generative models are starting to mature. Both variational autoencoders (VAEs) and diffusion methods can be applied to many existing GAN use cases both VAEs and diffusion figure prominently in Nvidia's new Cosmos platform for physical AI simulation Editor's note: This article was updated in January 2025 to improve the reader experience GANs vs. VAEs: What is the best generative AI approach? Generative models: VAEs, GANs, diffusion, transformers, NeRFs History of generative AI innovations spans 9 decades CNN vs. GAN: How are they different? Metrics details Group-III nitride semiconductors such as GaN have various important applications based on their three-dimensional form Previous work has demonstrated the realization of buckled two-dimensional GaN which can be used in GaN-based nanodevices the understanding of buckled two-dimensional GaN remains limited due to the difficulties in experimental characterization we have experimentally determined the phonon dispersion of buckled two-dimensional GaN by using monochromatic electron energy loss spectroscopy in conjunction with scanning transmission electron microscopy A phonon band gap of ~40 meV between the acoustic and optical phonon branches is identified for buckled two-dimensional GaN This phonon band gap is significantly larger than that of ~20 meV for the tetrahedral-coordinated three-dimensional GaN Our theoretical calculations confirm this larger phonon band gap Our findings provide critical insights into the phonon behavior of buckled two-dimensional GaN which can be used to guide high-performance thermal management in GaN-based high-power devices Buckled 2D GaN is embedded in the sample with a thickness of only a few atomic layers making experimental characterization difficult the phonon dispersion of buckled 2D GaN has never been experimentally reported due to the limitations of traditional spectral measurement methods in terms of spatial resolution and spectral range Our experimental results of phonon dispersion agree well with the numerical calculations These findings on phonon behaviors and phonon band gap demonstrate that buckled 2D GaN can play an important role in the thermal management of GaN-based devices Our methods for detecting phonon dispersion enrich the understanding of buckled 2D GaN and the thermal engineering of GaN-based devices which are calculated based on our established atomic model The distance between atomic columns is closer at the \([10\bar{1}0]\) zone axis making it more difficult to distinguish the positions of light atoms the \([11\bar{2}0]\) zone axis is preferable for displaying the atomic structure of buckled 2D GaN Through our elemental and structural analysis we classified the GaN layers in our sample as buckled 2D GaN It can be found that the EELS peak at 70 meV for buckled 2D GaN at on-axis geometry vanishes at off-axis geometry peaks at ~80 meV for bulk GaN are classified as optical phonons The blueshift of the optical phonons is the major difference in phonon excitation between buckled 2D GaN and bulk GaN buckled 2D GaN can aid in the fundamental understanding of the phonon transport of 2D materials providing insights for the design of nanoelectronic devices with considerations for thermal management we fabricated a two-dimensional buckled GaN layer on a 4H-SiC substrate Using an electron microscope with high spatial and energy resolution we investigated the phonon behaviors of buckled 2D GaN By performing on-axis and off-axis EELS measurements we extracted the phonon signals of buckled 2D GaN in the energy range of 20–120 meV the optical phonons of buckled 2D GaN are blueshifted from ~80 to 110 meV The 4D-EELS measurements revealed the phonon dispersion of buckled 2D GaN along a specific momentum path The phonon band gap between the acoustic and optical phonon branches of buckled 2D GaN is ~40 meV This value is larger than the ~20 meV phonon band gap of the bulk GaN The experimental results of the phonon dispersion agree well with our numerical calculations Our work experimentally demonstrates the phonon behaviors of buckled 2D GaN The techniques used to extract phonon signals and dispersion via EELS can also be applied to study other spatially confined structures Our investigation of the phonon behaviors of buckled 2D GaN significantly contributes to high-performance thermal management in nano- and optoelectronics The GaN films were obtained following an AlN buffer layer on graphene/SiC substrates by metal-organic chemical vapor deposition (MOCVD). The AlN buffer layer grown on graphene was polycrystalline. As shown in Fig. S1 a columnar structure was found in the AlN layer The Ga and N atoms migrated through the boundaries of AlN columns and the damaged regions of graphene during the growth of GaN and then formed a buckled 2D GaN layer on the 4H-SiC substrate Atomic resolution structural analysis was conducted by an aberration-corrected transmission electron microscope operated at 300 kV (Thermo Fisher Scientific Titan Cubed Themis G2) The convergence semiangle was 30 mrad for the iDPC and HAADF-STEM imaging and the collection semiangles were 4–21 mrad for the iDPC-STEM imaging and 39–200 mrad for the HAADF-STEM imaging we selected acquisition times of 400–800 ms per pixel and ~30 min for each dataset in total Four-dimensional EELS datasets were acquired with a 3 mrad convergence semiangle. A slot aperture with an aspect ratio of 16:1 was placed along the \(\Gamma -{{\rm M}}-\Gamma\) line. To avoid the strong central diffraction spot and enhance the signal-to-background ratio, the slot aperture was displaced by a reciprocal lattice vector (Fig. 4a) The acquisition time was 15 seconds per pixel and ~40 min for each dataset in total the signals were first normalized by the ZLP intensity We fitted the spectra to a Pearson function in two energy windows This approach allowed us to remove the ZLP the signals showed several energy loss peaks Lucy–Richardson deconvolution was employed to reduce the broadening effect caused by finite energy resolution while taking the elastic peak as the point spread function For the 4D-EELS datasets, a correction for the statistical factor was performed following the literature53 This process suppressed the low-energy peaks because they had a higher occupation number the ZLP was automatically extremely small and could be neglected The data that support the findings of this study are available from the corresponding author upon request The codes that support the findings of this study are available from the corresponding author upon request Current status of GaN-based solid-state lighting Recent development of vertical GaN power devices visible-blind GaN photodiodes fabricated by molecular beam epitaxy low noise ultraviolet photodetectors based on GaN p-i-n and AlGaN(p)-GaN(i)-GaN(n) structures Fabrication of room temperature continuous-wave operation GaN-based ultraviolet laser diodes Gallium nitride devices for power electronic applications GaN-on-Si power technology: devices and applications Gallium-nitride semiconductor technology and its practical design challenges in power electronics applications: an overview Graphitic nanofilms as precursors to wurtzite films: theory Computational synthesis of single-layer GaN on refractory materials Ab initio synthesis of single-layer III-V materials Computational discovery of single-layer III-V materials Two-dimensional gallium nitride realized via graphene encapsulation Two-dimensional wide band-gap nitride semiconductor GaN and AlN materials: properties Nanophononics: phonon engineering in nanostructures and nanodevices Phonon-isotope scattering and thermal conductivity in materials with a large isotope effect: a first-principles study Phonon-engineered extreme thermal conductivity materials Anomalously temperature-dependent thermal conductivity of monolayer GaN with large deviations from the traditional 1/T law Orbitally driven low thermal conductivity of monolayer gallium nitride (GaN) with planar honeycomb structure: a comparative study Two-dimensional buckling structure induces the ultra-low thermal conductivity: a comparative study of the group GaX (X = N Two-dimensional mapping of chemical information at atomic resolution Recent progress of vibrational electron energy-loss spectroscopy in scanning transmission electron microscope Vibrational spectroscopy in the electron microscope Mapping vibrational surface and bulk modes in a single nanocube Advances in ultra-high energy resolution STEM-EELS Spectroscopic imaging of single atoms within a bulk solid In situ STEM-EELS observation of nanoscale interfacial phenomena in all-solid-state batteries Towards atomically precise manipulation of 2D nanostructures in the electron microscope Atomic imaging and spectroscopy of low-dimensional materials with interrupted periodicities Measurement of vibrational spectrum of liquid using monochromated scanning transmission electron microscopy-electron energy loss spectroscopy Visualizing quantum phenomena at complex oxide interfaces: An atomic view from scanning transmission electron microscopy Four-dimensional vibrational spectroscopy for nanoscale mapping of phonon dispersion in BN nanotubes Measuring phonon dispersion at an interface Four-dimensional electron energy-loss spectroscopy Phase contrast STEM for thin samples: Integrated differential phase contrast Phonon dispersion and Raman scattering in hexagonal GaN and AlN Effects of localized interface phonons on heat conductivity in ingredient heterogeneous solids Systematic absences of optical phonon modes in phonon dispersion measured by electron microscopy Raman scattering microscopy imaging of basal plane stacking faults and associated partial dislocations in 4H-SiC crystals Vibrational spectroscopy at atomic resolution with electron impact scattering Advanced capabilities for materials modelling with Quantum ESPRESSO Quantum ESPRESSO: a modular and open-source software project for quantum simulations of materials A review of focused ion beam milling techniques for TEM specimen preparation A review of focused ion beam applications in microsystem technology Image denoising by sparse 3-D transform-domain collaborative filtering Observing crystal nucleation in four dimensions using atomic electron tomography Interpretation of meV resolution phonon EELS data Download references The work was supported by the National Key R&D Program of China (No the National Natural Science Foundation of China (No Beijing Outstanding Young Scientist Program (No the Beijing Natural Science Foundation (No and the Guangdong Major Project of Basic and Applied Basic Research (No The work was supported by the High-performance Computing Platform of Peking University We acknowledge Electron Microscopy Laboratory of Peking University for the use of electron microscopes State Key Laboratory for Mesoscopic Physics and Frontiers Science Center for Nano-optoelectronics State Key Laboratory of Integrated Optoelectronics College of Electronic Science and Engineering Collaborative Innovation Center of Quantum Matter International Center for Quantum Materials Peking University Yangtze Delta Institute of Optoelectronics All authors contributed to this work through useful discussion and/or comments on the manuscript Wandong Xing and the other anonymous reviewer(s) for their contribution to the peer review of this work Reprints and permissions Download citation DOI: https://doi.org/10.1038/s41467-024-54921-8 Foresight’s Ventures team continues to support Cambridge GaN Devices with £1.1m investment a leading innovator in gallium nitride (GaN) power devices has successfully closed a $32 million Series C funding round The investment was led by a strategic investor and supported by existing investors Foresight Group IQ Capital and with participation from British Patient Capital CGD is a University of Cambridge spin-out founded by Dr Florin Udrea in 2016 to commercialise a new generation of GaN power devices Power devices are used to control and convert electricity in all modern power electronics systems including power supplies CGD has developed a new architecture that makes its GaN chips a like-for-like replacement for silicon chip devices but with greater efficiency The Company continues to lead in the development of GaN-based power devices offering high efficiency solutions that can enable up to 10% savings in energy bills and potentially save millions of tons of CO2 emissions in applications such as data centre power supplies The global GaN power device market is projected to reach $5.15 billion by 2032 growing at a CAGR of 36% from 2024 to 2032 This substantial market opportunity underscores the potential for CGD's innovative technologies This significant investment will fuel CGD's growth strategy and boost its ability to meet the growing demand for efficient GaN products from the high-power industrial The funding will also enable the Company to expand its operations in Cambridge and deliver its unique value proposition to its growing customer base Foresight's Ventures team makes equity investments into fast-growing companies that are building some of the most exciting deep tech and hard tech solutions and enterprise software Foresight works with the founders and management teams to build resilient prosperous companies to deliver strong economic returns and to have a positive impact on the societies in which they operate Foresight’s Ventures strategy has invested £80 million and supported 39 deep tech investments commented on the investment: "This funding round marks a pivotal moment for CGD It validates our technology and vision to revolutionize the power electronics industry with our efficient GaN solutions and make sustainable power electronics possible We're now poised to accelerate our growth and make a significant impact in reducing energy consumption across multiple sectors We look forward to collaborating with our strategic investor to penetrate the automotive market" added: “We are delighted to continue backing the ambitious CGD team in bringing groundbreaking GaN technologies to market CGD’s products can deliver higher performance while enabling up to 50% energy savings across multiple growing sectors such as automotive We look forwarding to seeing the impact that CGD can make in enabling greener power electronics” Advisers to the Company: Mishcon de RayaAdvisers to Foresight: Ashfords For more information contact:Chris Barry at Influential on 07733 103693 / foresight@thisisinfluential.comForesight Group on +44 (0)20 3667 8100 / marketing@foresightgroup.eu About Foresight’s Ventures StrategyForesight's Ventures team makes equity investments into fast-growing companies that are building some of the most exciting deep tech hard tech solutions and enterprise software It is part of Foresight Group’s award-winning Private Equity division Its investment team is based across the UK and Ireland enhance nature recovery and realise the economic potential and positive impact of ambitious companies Foresight Group Holdings; investing to build a sustainable future and grow thriving economies Metrics details Phonon engineering is crucial for thermal management in GaN-based power devices where phonon-defect interactions limit performance detecting nanoscale phonon transport constrained by III-nitride defects is challenging due to limited spatial resolution we used advanced scanning transmission electron microscopy and electron energy loss spectroscopy to examine vibrational modes in a prismatic stacking fault in GaN By comparing experimental results with ab initio calculations we identified three types of defect-derived modes: localized defect modes the PSF exhibits a smaller phonon energy gap and lower acoustic sound speeds than defect-free GaN Our study elucidates the vibrational behavior of a GaN defect via advanced characterization methods and highlights properties that may affect thermal behavior rigorous understanding of phonon behaviors and transport in the presence of defects is critical for designing enhanced high-power-density III-Nitride based devices correlating phonon transport with defects is challenging particularly for obtaining mode-resolved phonon information result in lattice strain and broken periodicity which make correlation of specific defect atom vibrations with particular phonon modes difficult poor periodicity also leads to oversized computational models and thus exaggerated numerical challenges localized defect vibrations can only be detected within a few nanometers of the defects beyond the resolution of traditional measurement methods such as Raman scattering spectroscopy whose spatial resolution is constrained by laser wavelength it is important to study the vibrations of PSF since it is correlated with other planar defects as well Our study builds a bridge between defect-phonon interactions and thermal conductivity in materials with high defect densities through direct observation of defect phonon modes a Schematic of the experimental setup for phonon measurements The electron beam is focused on the GaN sample with a PSF The energy loss of the electron beam is analysed using an EELS spectrometer to determine the vibrational energy of the sample b Schematic of the EELS aperture placement in reciprocal space The green circle marks the on-axis geometry of the aperture with a 35 mrad convergence semiangle and the yellow circle marks the position of the aperture with a 25 mrad collection semiangle The yellow rectangle illustrates the position of the slot aperture used in 4D-EELS d HAADF-STEM image of the GaN PSF viewed along the [0001] zone axis e GPA map of the PSF corresponding to the same area with (d) which demonstrates that the strain is located at the center of the PSF with a range of approximately 1 nm diminishing to become negligibly small in the bulk GaN beyond this and (c) IDB-I viewed along the [0001] zone axis and the \([10\bar{1}0]\) zone axis The calculated DOS were extracted from defect-free GaN (blue) and at the PSF (red) with atomic configurations of (d) DPB-I The phonon DOS of bulk GaN exhibits four peaks at around 18–32 meV 69–79 meV (attributed mostly to transverse optic (TO) modes and 83–88 meV (attributed mostly to longitudinal optic (LO) modes) These peaks are also found in all three models demonstrating that the bulk like modes are captured in these supercells additional peaks occur in DPB-I around 45–50 meV These additional vibrational peaks are attributed to defect modes originating from the PSF The phonon energies around these calculated defect modes are shown in the red (LDM1 and LDM2) blue (CBM) and yellow (FEM) shaded regions of (b) The figures below are the corresponding experimental phonon intensity maps with the energy ranges shown in the red (LDM1 and LDM2) blue (CBM) and yellow (FEM) shaded regions of (a) and the HAADF-STEM images within the same areas further supporting the conclusion that surface-phonon polaritons are not playing a significant role here Experimental phonon dispersion line profiles of (a) bulk GaN and (b) the GaN PSF defect, acquired with the slot aperture placed along the Γ-M- Γ line, which is shown in the yellow rectangle in Fig. 1c The white dotted curves are eye guides for the approximate boundaries of the acoustic and optical phonon signals to show the energy gap The white dots mark the points with the highest intensity at that momentum and are fitted to a polynomial curve whose slope gives a semi-quantitative representation of the average sound speed as the acoustic branches have the only appreciable dispersion in this frequency range near the zone center The corresponding simulation results of (c) bulk GaN and (d) the PSF defect based on DFT calculations and scattering cross section in contrast to interfaces between two different materials Further exploration is necessary to detect symmetry breaking defects which often exhibit strong strain fields and are difficult to visualize our study gives mode-resolved microscopic details regarding phonon behaviors near extended defects providing valuable insights into defect engineering for targeted phonon-derived applications such as thermal management Wurtzite quasi-van der Waals epitaxial (vdWe) GaN films were grown on a graphene/sapphire substrate using plasma-assisted molecular beam epitaxy (MBE). The in situ RHEED image of GaN films is shown in Supplementary Fig. 8 The streaky RHEED patterns indicate a flat surface where no nitrogen reconstruction can be detected nucleation islands coalesce and eventually form a GaN film consisting of a prismatic stacking fault (PSF) defect A transmission electron microscopy (TEM) lamella was cut by a Thermo Fisher Scientific Helios G4 UX focused ion beam (FIB) system To determine the atomic arrangement of the GaN planar defect scanning transmission electron microscopy (STEM) measurements were performed using a spherical aberration-corrected FEI Titan Cubed Themis G2 300 STEM-EELS measurements were performed using a Nion HERMES 200 microscope operating at 60 kV at room temperature equipped with both a monochromator and an aberration corrector 3D-EELS datasets were obtained with a convergence semiangle of 35 mrad and a collection semiangle of 25 mrad The STEM-EELS on-axis dataset was acquired in a 100 × 30 pixels mapping area covering a region of 3 × 0.9 nm that included the PSF The dwell time for each pixel was set to 200 ms resulting in a total acquisition time of approximately 10 minutes for each dataset The sample drift during the acquisition was typically less than 1 nm The energy resolution achieved under these conditions was approximately 10 meV while the spatial resolution was approximately 0.2 nm For the acquisition of the 4D-EELS datasets a convergence semiangle of 3 mrad was used and a slot aperture with an aspect ratio of 16:1 was placed along the Γ-M-Γ direction To increase the signal-to-background ratio and avoid strong central diffraction spots the slot aperture was displaced away from the central Γ point The typical dwell time per pixel was set to 15 s resulting in a total acquisition time of approximately 32 minutes for each dataset The energy resolution achieved under these conditions was approximately 15 meV Model interfacial PSF structures (see Fig. 2) were built using the bulk lattice parameters in 56-atom orthorhombic unit cells that consisted of the PSF structure (16 atoms) and two bulk hexagonal ring structures on either side of the interface (20 atoms each) Variations between the different model arrangements were described above The use of periodic boundary conditions (as in typical DFT calculations) with this unit cell gives periodic PSF interfaces with an 11.32 Å centre-to-centre distance between them This provides enough space to nearly recover the bulk vibrational behaviour for atoms sandwiched between the two interfaces 3 × 1 × 2 supercells (336 atoms) were built (in-plane parallel to the interface × in-plane perpendicular to the interface × cross-plane parallel to the interface) to avoid spurious periodic forces in the DFT simulations Note that long-range Coulomb corrections were not explicitly included for the defect dynamical matrices the Ga and N vibrations away from the interface exhibited similar behaviour as that in the bulk calculations with polar corrections The vibrational density of states (\({{\rm{DOS}}}(\omega )\)) and atom-projected \({{\rm{DOS}}}(\omega,k)\) were obtained via: where the wavevector sum is over the first Brillouin zone and the energy delta function is approximated by a Lorentzian of smearing width 0.15 meV where \(n_{{{\bf{q}}}j}\) is the phonon occupancy number The Dirac delta function is represented by δ(x) to make it more comparable to the real phonon DOS and these dots are then fitted with a polynomial curve for visualization We obtain the sound speed ratio between PSF and GaN by averaging the slope ratios for the two systems on 6 different momentum ranges from \(\Gamma \sim 0.3{{\rm{M}}}\) to \(\Gamma \sim 0.5{{\rm{M}}}\) Polarization-induced hole doping in wide–band-gap uniaxial semiconductor heterostructures GaN/NbN epitaxial semiconductor/superconductor heterostructures A polarization-induced 2D hole gas in undoped gallium nitride quantum wells Complete composition tunability of InGaN nanowires using a combinatorial approach High-luminosity blue and blue-green gallium nitride light-emitting diodes Room temperature lasing at blue wavelengths in gallium nitride microcavities Thermal management of GaN-on-Si high electron mobility transistor by copper filled micro-trench structure Deep‐ultraviolet micro‐LEDs exhibiting high output power and high modulation bandwidth simultaneously High‐output‐power ultraviolet light source from quasi‐2D GaN quantum structure Drive high power UVC‐LED wafer into low‐cost 4‐inch era: effect of strain modulation Effects of point defects and dislocations on spectral phonon transport properties of wurtzite GaN Impact of screw and edge dislocations on the thermal conductivity of individual nanowires and bulk GaN: a molecular dynamics study Effect of nitrogen and vacancy defects on the thermal conductivity of diamond: an ab initio Green’s function approach Phonon scattering effects from point and extended defects on thermal conductivity studied via ion irradiation of crystals with self-impurities GaN thermal transport limited by the interplay of dislocations and size effects Dislocation-induced thermal transport anisotropy in single-crystal group-III nitride films Effect of dislocations on thermal conductivity of GaN layers Intersecting faults on basal and prismatic planes in aluminium nitride Domain boundaries on {\(11\bar{2}0\)} planes in GaN: a theoretical study Observation of two atomic configurations for the {\(1\bar{2}10\)} stacking fault in wurtzite (Ga Vibrational and valence aloof beam EELS: a potential tool for nondestructive characterization of nanoparticle surfaces The influence of surfaces and interfaces on high spatial resolution vibrational EELS from SiO2 Electron Energy Loss Spectroscopy and Surface Vibrations (Academic Press Structural TEM study of nonpolar a-plane gallium nitride grown on (\(11\bar{2}0\)) 4H-SiC by organometallic vapor phase epitaxy Atomic structure of prismatic stacking faults in nonpolar a-plane GaN epitaxial layers Structural defects in GaN crystals grown by HVPE on needle-shaped GaN seeds obtained under high N2 pressure Prismatic defects in GaN grown on 6H-SiC by molecular beam epitaxy and emission characterization of trench defects in InGaN/GaN quantum well structures Structural defects and luminescence features in heteroepitaxial GaN grown on on-axis and misoriented substrates The heterogeneous nucleation of threading dislocations on partial dislocations in III‑nitride epilayers Formation mechanism and relative stability of the {\(11\bar{2}0\)} stacking fault atomic configurations in wurtzite (Al,Ga,In) nitrides Probing hyperbolic shear polaritons in β-Ga2O3 nanostructures using STEM-EELS Probing low-energy hyperbolic polaritons in van der Waals crystals with an electron microscope Temperature measurement by a nanoscale electron probe using energy gain and loss spectroscopy Direct observation of highly confined phonon polaritons in suspended monolayer hexagonal boron nitride Revealing nanoscale confinement effects on hyperbolic phonon polaritons with an electron beam Three-dimensional vectorial imaging of surface phonon polaritons Position and momentum mapping of vibrations in graphene nanostructures Manipulation of surface phonon polaritons in SiC nanorods Vibrational spectroscopy of water with high spatial resolution Thermometry with subnanometer resolution in the electron microscope using the principle of detailed balancing Single-defect phonons imaged by electron microscopy Imaging of isotope diffusion using atomic-scale vibrational spectroscopy Single-atom vibrational spectroscopy in the scanning transmission electron microscope Single-atom vibrational spectroscopy with chemical-bonding sensitivity Atomic-resolution mapping of localized phonon modes at grain boundaries Direct visualization of localized vibrations at complex grain boundaries Nanoscale Localized Phonons at Al2O3 Grain Boundaries Nanoscale imaging of phonon dynamics by electron microscopy Emergent interface vibrational structure of oxide superlattices Atomic-scale probing of heterointerface phonon bridges in nitride semiconductor Experimental observation of localized interfacial phonon modes Thermal conductivity and large isotope effect in GaN from first principles Three-phonon phase space and lattice thermal conductivity in semiconductors First-principles determination of ultrahigh thermal conductivity of boron arsenide: a competitor for diamond Nonperturbative quantum nature of the dislocation-phonon interaction Thermal transport in defective and disordered materials Phonon dispersion curves in wurtzite-structure GaN determined by inelastic x-ray scattering High-order phonon anharmonicity and thermal conductivity in GaN infrared and terahertz nanophotonics using surface phonon polaritons Phonon vortices at heavy impurities in two-dimensional materials Thermal transport by first-principles anharmonic lattice dynamics In: Handbook of Materials Modeling 735-765 (Springer Electrons and Phonons (Oxford University Press Theory of momentum-resolved phonon spectroscopy in the electron microscope Theoretical X-ray absorption Debye- Waller factors New analytical scattering-factor functions for free atoms and ions Download references This work was supported by the National Key R&D Program of China (No the Guangdong Major Project of Basic and Applied Basic Research (2023B0303000012) (W.X.) and the National Natural Science Foundation of China (No This work was supported by the High-performance Computing Platform of Peking University We acknowledge the Electron Microscopy Laboratory of Peking University for the use of electron microscopes Calculations and manuscript development (L.L.) were supported by the U.S Material Sciences and Engineering Division The calculations used resources of the Compute and Data Environment for Science (CADES) at the Oak Ridge National Laboratory which is supported by the Office of Science of the U.S and resources of the National Energy Research Scientific Computing Center Wang conceived and supervised the project; H.L Jiang performed the STEM-EELS experiment and data analysis assisted by Z.Y Shi designed the toolbox for data processing; F Liu grew the samples with the guidance of X.Q All the authors contributed to this work through useful discussion and/or comments on the manuscript Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work Download citation DOI: https://doi.org/10.1038/s41467-024-53394-z APEC takes place at Atlanta's Georgia World Congress Center from March 16th to 20th The company's “Planet Navitas” stand (Booth #1107) will showcase the company’s mission to ‘Electrify our World™’ by advancing the transition from legacy silicon to next-generation clean energy GaN and SiC power semiconductors These technologies are designed for high-growth markets that demand the highest efficiency and power density The shift from silicon to GaN and SiC technologies has the potential to save over 6,000 megatons of CO2 per year by 2050 Recent Navitas breakthroughs that will be highlighted on the stand include: Navitas will participate and present in the Industry Session which showcases the latest work in all areas of power electronics Technical Presentations:Wednesday 19th March Stephen Oliver, VP Corporate Marketing & Investor Relationsir@navitassemi.com Ethan Spigland is a filmmaker, visual artist, writer and curator based in Brooklyn. He is a Professor in the Graduate Program of Media Studies at Pratt Institute. Home Designers Lulu Tan-Gan and Ria Bolivar are proof that it doesn’t matter whether your journey as a fashion designer started over 40 years ago or just over the pandemic it’s the love of fashion and the craft that keeps you going and creating This is an excerpt from the MEGA May 2025 Designer Profile LULU TAN-GAN: Almost half a century, 48 years ago. But my brand is 40 years RIA BOLIVAR: I started during pandemic. I was thinking that I know how to crochet so I’m going to make this for my kid. Like cute bags for my son. And after that cute bags for my friends can I buy that from you?” and “Can you make it like this?” So I started from there RELATED: Treasures in Thrifting with Fed Pua and Jodinand Aguillon You kind of think of what design you want to make which sometimes there are years where knitwear is not so much in trend we were following trends… My process is really having to adjust to the limitations of loom crafting You have to make sure everything is in measure with crochet and knitting It’s been my thing ever since I wanted to be a fashion designer in my senior year in high school—because I was deciding what course to take for college The second reason is I think I enjoy the challenge of creating something new or innovative But I think what I did differently was I was just doing stitching and colors So I became known as “Queen of Knitwear” which I can’t shake off That was like 20 years… Then I started thinking I want to do something as a Filipino designer I was thinking of how I could put knitwear and piña together And so enjoying that challenge makes me continue with my innovation I have a lot of designer peers and models becoming designers When I stepped into Manila and started modelling here LTG: I feel very lucky to be born as a creative person and being surrounded by creative people because I think we have the least problems we just laugh about it and we find a solution Read more of Lulu Tan-Gan and Ria Bolivar’s conversation in MEGA’s May 2025 issue Production and Art Direction JONES PALTENG © 2025 MEGA is a published and presented by One Mega Group, Inc., a subsidiary of AGC Power Holdings Corp. All rights reserved. Sylvio Martins Highland Park CoffeeDonuts Koreatown Memorylook is a sleek coffee shop in Koreatown that serves cream-layered coffee drinks and fancy donuts CoffeeBakery/Cafe Cafe Upper is a modern coffee shop in Koreatown that's great for getting work done and drinking a busy coffee drink and they've got "croffles," too Sylvio moved to LA over a decade ago and still misses his exit on the 10 He came to us as a freelancer and wrote so many guides that we gave him a job Metrics details Understanding the thermal influence on gallium nitride (GaN) single crystal substrates is critical for the advancement of GaN-based optoelectronic devices we comprehensively characterized the thermal effects on the optical properties of silicon-doped GaN substrates using spectroscopic ellipsometry over a broad wavelength range from 250 nm to 1600 nm The dielectric function of GaN was determined at temperatures ranging from 298 K to 873 K demonstrating consistent temperature-dependent behavior The exciton transitions were precisely characterized and modeled using the empirical Varshni expression the thermo-optic coefficients across the wide spectrum This work significantly expand the GaN optical properties database beyond thin films and provide essential insights for the design and optimization of next-generation GaN-based optoelectronic devices To optimize the design of optoelectronic devices it is essential to understand the optical properties of GaN materials at their application temperatures it becomes imperative to isolate and mitigate the impact of extrinsic uncertainties associated with GaN film characteristics by investigating GaN single crystal substrates for their intrinsic optical properties the optical properties of temperature dependence of GaN substrate across a wide spectral and application temperature range is rarely reported and its thermo-optic response over a broad bandwidth UV-Vis-NIR is not fully explored the complete dielectric function and the derived thermo-optic coefficients which are crucial for precise optoelectronic device simulations The parameters \(\it \Psi\) and \(\it \Delta\) are related to the complex ratio of the reflection coefficients for light polarized parallel (\(p\)) and perpendicular (\(s\)) to the plane of incidence the optical properties of N-type GaN substrates are systematically investigated by the spectroscopic ellipsometry over the wavelength range from 250 nm to 1600 nm at elevated temperatures from room temperature to 873K The dielectric function of the GaN substrates is firstly determined and the exciton transitions are further extracted through critical points analysis the first-order thermo-optic coefficient is provided for the first time over a wide spectral range These findings on GaN substrates are analyzed and compared with previous studies on thin films Commercially available single-crystal N-type Si-doped GaN substrates (doping concentration of \(10^{18} \ grown by Hydride Vapor Phase Epitaxy (HVPE) and supplied by Hefei Jinko Material Technology Co. The GaN substrates possess a hexagonal wurtzite structure (\(\alpha\)-GaN uniaxial crystal) with a crystal orientation of \(\langle 0001 \rangle \pm 30'\) indicating that the optical axis is perpendicular to the substrate surface The Ga-face of the substrate was processed by chemical mechanical polishing achieving a surface roughness of \(\le\) 0.5 nm an isotropic model can be directly utilized for analysis of ellipsometric data from single surface reflection in the subsequent analysis of the ellipsometric parameters it is essential to consider the presence of this extra ultra-thin overlayer on the GaN substrate Pseudo dielectric function \(\varepsilon _{pseudo}\) of Si doped GaN substrate measured by the spectroscopic ellipsometer at 298 K The left arrow and right arrow point the real part and imaginary part of the pseudo dielectric function The inset shows the zoomed-in view at the wavelength range from 250 nm to 450 nm The inset shows the zoomed-in view in the spectral range of 3-4 eV and the exciton and inter transition is marked by arrow Real part (a) and imaginary part (b) of dielectric function \(\varepsilon\) of silicon-doped GaN at elevated temperatures ranging from 298 K to 873 K The inset provides a zoomed-in view of the spectral range highlighted by the black frame with the black arrow indicating the trend of changes at elevated temperatures To further clarify the movement of the transitions the second derivatives \(d^2\varepsilon /dE^2\) of these spectra is calculated and the standard analytic critical points (CPs) expression is employed to fit these data Exciton transition of Silicon doped GaN at elevated temperature from 298 K to 873 K. (a) the second derivatives \(d^2\varepsilon /dE^2\) and fitted by Eq.(2). (b) Exciton transition fitted by Eq.(3) the thermo-optic coefficient is calculated for values from 400 nm to 1600 nm As shown in Fig. 5(a) five sets of refractive index \(n\) values below the bandgap are selected for presentation It is observed that \(n\) varies almost linearly with temperature from room temperature to 873 K implying that \(n\) can be simply fitted by a first-order expression Here, \(\alpha\) with units of \(\hbox {K}^{-1}\) represents the thermo-optic coefficient on the wavelength, \(T_0\) is the reference temperature, and \(n_0\) is the corresponding refractive index. The black spheres in Fig. 5(b) indicate the \(\alpha\) fitted wavelength by wavelength over the range from 400 nm to 1600 nm \(\alpha\) remains flat slopes at around 0.0005 but sharply increases near the bandgap \(\alpha\) is further parameterized using a Sellmeier model \(R^{2}\)=0.999 indicates a good fit with the paramters of \(A=1.396\times 10^{-5}\pm 0.004\times 10^{-5}\), \(B=2.895\times 10^{-5}\pm 0.003\times 10^{-5}\) and \(C=1.447\times 10^{5}\pm 0.002\times 10^{5} nm^{2}\). By virtue of Eq.(4) the refractive index can be inferred at any wavelength within the measurement range the dielectric function of Silicon doped GaN substrates from room temperature to 873 K over the wavelength from 250 nm to 1600 nm substrates exhibit a regular and consistent temperature dependence Temperature dependent exciton transitions fit well with empirical Varshni expression Additionally,the thermo-optic coefficient of GaN substrate is parameterize by a Sellmeier model over the wavelength range from 400 nm to 1600 nm It is found that the temperature dependence of both the exciton transition and the refractive index of the substrate is lower than that reported in thin films which can support the design and optimization of GaN-based optoelectronic devices All data generated or analysed during this study are included in this published article and are available from the corresponding author on reasonable request GaN-on-Si power technology: Devices and applications Application of GaN-based ultraviolet-c light emitting diodes-UV LEDs-for water disinfection GaN technology for power electronic applications: A review Structural and optical properties of Si-doped GaN Dislocation core structures in Si-doped GaN The doping process and dopant characteristics of GaN an ideal candidate for high-temperature applications: A review Modelling and optical response of a compressive-strained AlGaN/GaN quantum well laser diode Design of GaN-based pcsel with temperature-insensitive lasing wavelength Temperature dependent dielectric function and the E critical points of hexagonal GaN from 30 to 690 k Spectroscopic ellipsometry of wurtzite ZnO and GaN: Examination of a special case Ordinary optical dielectric functions of anisotropic hexagonal GaN film determined by variable angle spectroscopic ellipsometry Ellipsometric and first-principles study on temperature-dependent UV-Vis dielectric functions of GaN Effects of thermal annealing on optical properties of Be-Implanted GaN thin films by spectroscopic ellipsometry A review of characterization of perovskite film in solar cells by spectroscopic ellipsometry Ellipsometry: dielectric functions of anisotropic crystals and symmetry Comprehensive study on initial thermal oxidation of GaN (0001) surface and subsequent oxide growth in dry oxygen ambient Birefringence and refractive indices of wurtzite GaN in the transparency range Pseudodielectric functions of uniaxial materials in certain symmetry directions Mueller matrix ellipsometer based on discrete-angle rotating Fresnel rhomb compensators Double-sided and single-sided polished 6H-SiC wafers with subsurface damage layer studied by mueller matrix ellipsometry Excitons and exciton-phonon coupling in the optical response of GaN 373–426 (Springer International Publishing Characterization of undoped and Si-doped bulk GaN fabricated by hydride vapor phase epitaxy Electron capture behaviors of deep level traps in unintentionally doped and intentionally doped n-type GaN Temperature dependence of optical properties of h-GaN films studied by reflectivity and ellipsometry Data analysis for spectroscopic ellipsometry Temperature dependence of the energy gap in semiconductors Thermal tuning of light-emitting diode wavelength as an implication of the Varshni equation Physics of group iv elements and iii–v compounds Handbook of optical constants of solids: Handbook of thermo-optic coefficients of optical materials with applications (Academic Press Dependence of the refractive index of Al x Ga 1–x N on temperature and composition at elevated temperatures The temperature dependence of the refractive indices of GaN and AlN from room temperature up to \(515^{\circ }\) c Experimental characterization of the thermo-optic coefficient vs temperature for 4h-SiC and GaN semiconductors at the wavelength of 632 nm Electronic structure and optical properties of semiconductors Download references discloses support for the research of this work from National Natural Science Foundation of China (Grant52275531) discloses support for the research of this work from Ministerio de Ciencia Innovación y Universidades TED2021-129639B-I00 (MCIU/AEI/FEDER College of Metrology Measurement and Instrument X.C and SB.B analysed the results and wrote the original manuscript Download citation DOI: https://doi.org/10.1038/s41598-025-87243-w Please enable JS and disable any ad blocker Gan Golan’s work combines grassroots community organizing with public spectacles that shift popular narratives and mobilize communities In 2015 he was awarded a Rauschenberg Foundation “Artist As Activist” Fellowship as a member of People’s Climate Arts He is a co-collaborator of the Climate Clock that counts down the critical time window to reach zero emissions Throughout the fall and spring semesters, the Artists Now! Guest Lecture Series welcomes a diverse group of nationally and internationally recognized artists working across traditional Join us to explore and expand boundaries of creative visual practices alongside others with an interest in contemporary visual art Visit Gan Golan’s website Sign up to receive emails by choosing which artistic fields resonate best with you We’ll only send you information you’re interested in – and don’t worry Copyright © University of Wisconsin-Milwaukee and University of Wisconsin Board of Regents. Camp Gan Yisroel New York is proud to announce that Rabbi Moshe Aaron Geisinsky of Pittsburgh Zal, will be joining the camp this summer as Mashpia. Full Story Camp Gan Yisroel New York is proud to announce that Rabbi Moshe Aaron Geisinsky of Pittsburgh Zal will be joining the camp this summer as Mashpia Rabbi Geisinsky will focus on strengthening the Chassidishe atmosphere among both staff and campers he’ll help ensure that even during the summer break bochurim stay connected to learning and continue growing in their Yidishkeit and Chassidishkeit “Camp is not just a break from yeshiva—it’s a time to grow,” said Rabbi Geisinsky “Our goal is to create a space where every bochur feels the warmth of Chassidus and grows from it all while enjoying all the excitement that camp offers.” the camp’s legendary Learning Director of nearly 40 years who remains a pillar of chinuch in Gan Yisroel—guiding both campers and staff with care and a deep understanding of how to reach each child “This couldn’t come at a better time,” said Rabbi Yossi Futerfas “As we mark 70 years since the Rebbe founded Gan Yisroel New York Rabbi Geisinsky’s addition is a powerful step in ensuring this summer is filled with true Chassidishe energy—for both campers and staff.” Sign up for the COLlive Daily News Roundup and never miss a story Δdocument.getElementById("ak_js_1").setAttribute("value",(new Date()).getTime()) Δdocument.getElementById("ak_js_2").setAttribute("value",(new Date()).getTime()) Gan Yisroel NY gets the best of the best 🔥🔥🔥 The nicest and most chashuve mashpia in lubavitch today Metrics details Recently in diagnosis of Aortic dissection (AD) the synthesis of contrast enhanced CT (CE-CT) images from non-contrast CT (NC-CT) images is an important topic Existing methods have achieved some results but are unable to synthesize a continuous and clear intimal flap on NC-CT images we propose a multi-stage cascade generative adversarial network (MCGAN) to explicitly capture the features of the intimal flap for a better synthesis of aortic dissection images For the intimal flap with variable shapes and more detailed features we extract features in two ways: dense residual attention blocks (DRAB) are integrated to extract shallow features and UNet is employed to extract deep features; then deep features and shallow features are cascaded and fused we use spatial attention and channel attention to extract key features and locations multi-scale fusion is used to ensure the continuity of the intimal flap We perform the experiment on a set of 124 patients (62 with AD and 62 without AD) The evaluation results show that the synthesized images have the same characteristics as the real images and achieves better results than the popular methods the accessibility of CE-CT is limited during specific times and in regions where medical resources are scarce there is a risk of delay from preliminary diagnosis to subsequent examination finding a diagnostic method that combines the high clarity of CE-CT with the low risk of NC-CT is important to improve the diagnostic efficiency and patient safety of AD using NC-CT but the model is large and has big memory cost There is no proven CE-CT image synthesis method for aortic dissection The challenge of synthesizing CE-CT from NC-CT due to the intricate nature of its feature distribution. as shown in Fig. 1 The first row of the figure is the NC-CT scans and the second row is the corresponding CE-CT scans The figure illustrates the following scenarios: (a) the indistinguishable image intensities between the true and false lumen in aortas with complete intimal flaps; (b) the presence of incomplete intimal flaps in some cases; (c) the stark contrast in image intensities within the true and false lumen in the presence of thrombus; and (d) non-aortic dissection image It can be seen that an intimal flap dividing the aortic lumen into two distinct regions is crucial to the successful synthesis of CE-CT images in patients with AD the intimal flap is thinness and indistinctness coupled with the great variability in its morphological characteristics across different pathological cases renders its identification and synthesis arduous The contributions of this work are the following: In order to learn the shape-varying intimal flap features we designed a new GAN model to synthesize aortic CE-CT images from NC-CT images which extracts the deep and shallow features of the image separately and then cascades them to ensure the adequate preservation of the feature In order to synthesize the location information of the true and false lumen we propose an image mapping and fusion module for feature decoding that combines spatial and channel attention as well as multi-scale fusion to ensure the clarity and continuity of the intimal flap in the synthetic image their study demonstrated the effectiveness of deep convolutional neural networks for aortic dissection image recognition The above methods have made substantial progress in the study of automatic segmentation and detection of aortic dissection and they have laid foundation for AD image synthesis these methods are associated with inconsistent data variable degrees of disease-related regions Highly nonlinear relationships between different imaging modalities makes it challenging to learn the mapping from one modality to another the synthesized medical images often exhibit discrepancies in fine-scale structure when compared to the real ones DenseNet is composed of iterative dense residual attention blocks (DRAB) with different output dimensions the information loss in the feedforward of CNNs can be reduced ensuring maximum utilisation of information the problem of vanishing gradient can be alleviated dense connections have a regularizing effect which reduces the risk of over-fitting on tasks with smaller training sets in order to adequately extract the aortic intima flap features in the NC-CT images the network retains the shallow features during the DRAB while the deep features are obtained through a global generator The global generator is a UNet with two convolution downsampling layers and two deconvolution layers The extracted features from top to bottom are of size 256 × 256 × 32 The output of the global generator has the same dimension as the input Figure 3a exhibits the structure of DRAB three concatenation layers and an identity connection Dense connections are used to ensemble the features from different layers for strengthening feature propagation The 1 × 1 convolution is adopted to decrease network width while retaining salient features and thus improve computational efficiency Dense connections have significant advantages they can reduce information transfer loss in feedforward CNNs alleviate the problem of vanishing gradient dense connections exhibit a regularizing effect which reduces the risk of overfitting on tasks with relatively small training sets Figure 3b illustrates the structure of residual attention block (RAB) one BN-ReLU- conv(1 × 1) and one BN-ReLU-conv(3 × 3) are stacked followed by a full-spectrum channel attention module (FCA) The channel attention is introduced into DenseUNet so that the network can focus more on critcal information when extracting features The details of the CA are introduced in the following sub-sections After the feature extraction, the shallow features at different scales and the deep global features are transferred to the local feature mapping part. In this part, the shallow features and deep features are cascaded four times, as shown in Fig. 2b In order to extract the key features of the images this part contains two attention modules: residual attention block (RAB) and spatial attention module (SA) Complementary information extracted by two attention models is employed to improve feature mapping The inputs of the attention module are two vectors xd and xs with same dimension xs respect a shallow feature map from DenseNet feature extraction and xd respect the deepest feature map from global generator They are compressed to the same output channels of C by a 1 × 1 convolution The outcomes are summed and subsequently processed through a ReLU activation function The ReLU-generated feature map is subsequently inputted into another 1 × 1 convolutional layer with a single output channel which is then passed through a sigmoid function to yield a pixel-by-pixel attention coefficient α xs undergoes calibration by being multiplied with α the spatial attention maps for both pathways are represented by α and αʹ where \(conv_{{1 \times 1}}^{C}( \cdot )\)is the 1 × 1 convolution with output channel C 64 and 32 from bottom to top in feature mapping part in the deepest layer of the feature mapping xs is the input of the global generator and xd is the output of the global generator the two features cascaded first time into the SA xs and xd cascaded three times into the SA and xd is the deconvolution of the deep features the intraluminal membrane of the aorta at the diseased site is ruptured and the visual characteristics of the boundary between the true and false cavities are not obvious on the CT image They are mostly manifested in the form of dissection tear lines because the sharp changes in the image gray value at the edge of the aortic tear belong to high-frequency information in order to improve the feature expression ability the frequency channel attention module is used to further filter the features of different channels in the feature map to improve the feature expression ability of key channels and further guide the network to focus attention to areas containing key information we concatenate the frequency components of the parts we performed a multi-layer perceptron to obtain the cross-channel interaction information and a convolution to adjust the channel number we computed the attention values α using a Sigmoid function the normalized weights are multiplied channel-by-channel with the original input feature map to generate the weighted feature map The attention mechanism is introduced into the feature mapping and the feature of the positioning is more accurate and the details are richer This approach leverages the multi-scale information to enrich the feature maps allowing our network to capture and synthesize finer image details In order to be able to better form aortic dissection images multifaceted constraints are introduced in the loss function These constraints can help the generator to better learn the structure and content of the aortic dissection images The main reason for using feature matching loss and perceptual loss instead of MSE is the excessive smoothness and lacking in details of the generated images by MSE loss The feature matching loss and the perceptual loss ensure that the content is consistent and the details are learned by the adversarial loss n is the number of layers and Ni is the number of elements per layer Minimizing this loss ensures that each layer extracts the same features from pairs of images where, \({F_i}(x)\) and \({F_i}(y)\) denote their feature representations at layer i in the pre-trained neural network, respectively, and N is the number of feature layers. The pre-trained neural network in the experiment is VGG19 network, which utilizes perceptual loss to extract features from paired images. The total loss function is defined in formulas (6) The dataset used in this study consisted of 124 different patients including 62 each of aortic dissection patients and non-aortic dissection patients The dataset was collected and manually annotated by specialized imaging physicians The study was approved by the institutional review board Each subject consists of a NC-CT scan and a corresponding CE-CT scan The paired scans were collected at the end of respiration and share same spatial resolution that range from 0.549 × 0.549 × 0.625 to 0.977 × 0.977 × 1.250mm3 The tube voltage of the collection device is 100–120KVp the personal information of each subject was desensitized Before feeding the data into the model, some preprocessing operations are required. The cross-sectional images obtained from CE-CT and NC-CT images are sharper than sagittal and coronal images, with higher density resolution. Thus, the following pre-processing procedure is performed on the paired cross-section as shown in Fig. 4 NC-CT (the first row) and corresponding CE-CT (the second row) Example of the dataset before and after segmentation on axial All experiments were implemented with Pytorch We trained the model on two NVIDIA GeForce RTX 2080 Ti with 11GB memory each We set the intimal learning rate 2 × 10−4 and use Adam optimizer with beta1 0.5 \(\lambda\) for feature matching loss is set to 10 This comparative analysis was conducted on both qualitative and quantitative fronts with a focus on the PSNR and SSIM metrics to evaluate the synthesized images’ quality relative to ground truth Comparative visualization of synthesized CE-CT: axial CycleGAN is completely unable to synthesize the intimal flap in the yellow box in all three views of the aorta Pix2pix is a bit better than CycleGAN in terms of brightness but it only synthesizes some noise in the lesion area and does not synthesize the intimal flap Pix2pixHD generates images with higher resolution and brightness but also lacks grey scale detail of the flap We observe that the CDLF generates intimal flaps with a degree of blurriness the results achieved by CDLF are superior to those produced by the 3DMTGA This may be attributed to the insufficiency of a small dataset to train a robust 3DMTGA model clear and correct synthesis results for all three views MCGAN is able to synthesize the intimal flap of aortic dissection close to the ground truth in all three views the features of the curved and irregular intimal flap are also well preserved thus MCGAN captures subtal image details and improves the overall performance indicating that MCGAN achieves better visual effect and less distortion than other methods shallow features and deep features are extracted by different methods To verify the benefit of the multi-stage cascades two-stage and three-stage cascade networks by choosing the same parameters Ablation visualization of cascade times: axial Figure 7 shows views of axial sagittal and coronal synthesized CE-CT images for a randomly selected subject with AD the synthesized image as a whole looks noisy and not smooth enough the aortic edges on the axial view are not clear enough and the synthesis of the intimal flap is incomplete the synthesized intimal flap is not continuous the synthesized aortic valve is distorted on the axial view and blurred on the coronal view intimal flap features were extracted in all three views the features were discontinuous in the axial view the intimal flap was unclear in the coronal view and no intimal flap is synthesized in the sagittal view there is inconsistency in the intensity values of the true and false lumen caused by the thrombus noise and artefacts in the synthesizes image increase interfering with the extraction of intima features the intimal flap in the lesion area is very thin and difficult to distinguish from the surrounding tissue and noise may interfere with intimal recognition and synthesis our method with cascade four times synthesizes images closer to the true value in three views It is evident that features are extracted both shallowly and deeply and then integrated in multi-stage cascades This approach ensures the retention of both global and detailed features This indicates the necessity and superiority of cascade Our method uses the DRAB and UNet to obtain four different scale features and then implement the SA and CA to perform the synthesis these attentions play an important role in our method To verify the necessity of the spatial attention and channel attention modules in local feature mapping we separately synthesized CE-CT images from the same set of NC-CT images without attention with spatial attention only and channel attention only Ablation visualization of attention combinations: axial Figure 8 shows axial sagittal and coronal views of synthesized CE-CT images endocardial slices were blurred or even absent the generated image endomembrane slices are incomplete and unclear the generated image inner membrane slice is not continuous enough The whole image generated by using double attention is more uniform and the continuity of the length of the inner membrane slice is also better spatial attention only and channel attention only MCGAN with double attention makes great contributions to performance improvement of image synthesis This indicates the necessity and superiority of double attention The four scale features obtained from the cascade of shallow feature and deep feature of MCGAN generator are multi-scale fusion To evaluate the superiority of feature fusion we perform the GAN with and without feature fusion separately Ablation visualization of multiscale fusion: axial Figure 9 shows three views of synthesized CE-CT images the synthesized image has a grid effect and overall lack of clarity And the synthesized image after performing multi-scale fusion is richer in detail information Multi-scale feature fusion allows for better understanding and reconstruction of shape and texture information of the intimal flap which improves the accuracy of the synthesis Feature maps and attention maps in different scales In Table 5 we delineate the floating-point operations (FLOPs) and average inference times for the models under comparison The MCGAN exhibits a computational complexity of 122.96 GFLOPS and inference time is extended to 0.0092 s substantially higher than its counterparts Despite the increased computational expense the MCGAN’s architectural sophistication has been empirically demonstrated to enhance the quality of synthesized images significantly capturing and rendering image key features with greater finesse we argue that the increased FLOPs and inference time are a justifiable trade-off for the performance gains achieved by MCGAN Given the ongoing advancements in hardware capabilities especially in the realms of parallel processing and specialized hardware acceleration the computational demands of MCGAN are manageable and are projected to diminish over time The synthesis of CE-CT images from NC-CT images is an important technique in AD diagnosis The NC-CT images exhibit numerous artefacts and noise resulting in small density variations that hinder the precise discrimination of intricate vascular structures the dynamic alterations in the shape and position of the true and false lumens pose a significant challenge for accurate imaging capture MCGAN is proposed in this paper to overcome these difficulties the aortic regions of interest in paired NC-CT and CE-CT were segmented using the nnU-Net network and post processing the generator of MCGAN takes segmented aortic NC-CT slices as inputs to produce synthetic CE-CT images and the discriminator takes real CE-CT slices and synthetic CE-CT images together as inputs to guide the generator to synthesize more realistic CE-CT slices The generator of this network consists of three parts Feature extraction is performed in two ways DRAB for shallow features and UNet for deep features This approach not only captures local features of the image but also provides more global structural information The feature maps cascade and fuse shallow and deep features and exploit channel attention and spatial attention By complementing spatial and frequency domain information the network focuses on the morphology and changes of the intimal flaps improving the synthesis accuracy of true and false lumen and intimal flaps The feature fusion achieves multi-scale fusion of the four scales obtained by cascading shallow and deep features This approach better captures the richness and diversity of intimal flap features helping to improve the quality and authenticity of synthetic images making the synthesis of true and false luminal intimal flaps clearer and more accurate The PSNR and SSIM values of the synthesized image and the original image were calculated through the comparison experiment The results demonstrated that the proposed framework generally outperforms baseline methods both quantitatively and qualitatively attention modules and multi-scale fusion is also proved by the convincing results To assess the proposed method comprehensively we will research on the follow-up quantitative analysis (e.g. image segmentation and detection) of generated CE-CT images in future The validation datasets generated and/or analysed during the current study are not publicly available due to the ethical and patient privacy regulations but are available from the first author (Juanjuan Yin) on reasonable request Harris, R. J. et al. Classification of aortic dissection and rupture on post-contrast CT images using a convolutional neural network. J. Digit. Imaging 32, 939–946. https://doi.org/10.1007/s10278-019-00281-5 (2019) Hata, A. et al. Deep learning algorithm for detection of aortic dissection on non-contrast-enhanced CT. Eur. Radiol. 31, 1151–1159. https://doi.org/10.1007/s00330-020-07213-w (2021) Multi-stage learning for segmentation of aortic dissections using a prior aortic anatomy simplification D2polyp-Net: A cross-modal space-guided network for real-time colorectal polyp detection and diagnosis Usability of synthesized image using generative adversarial network for prediction model of recurrence after radiotherapy in locally advanced cervical cancer Li, Z. Y. et al. Lumen segmentation of aortic dissection with cascaded convolutional network. Int. Workshop Stat. Atlases Comput. Models Hearthttps://doi.org/10.1007/978-3-030-12029-0_14 (2019) Goodfellow, I. et al. Generative adversarial networks. Commun. ACM 63, 139–144. https://doi.org/10.1145/3422622 (2020) LR-cGAN: Latent representation based conditional generative adversarial network for multi-modality MRI synthesis Prior-guided generative adversarial network for mammogram synthesis Xiong, X. et al. A cascaded multi-task generative framework for detecting aortic dissection on 3-D non-contrast-enhanced computed tomography. IEEE J. Biomed. Health Inform. 26, 5177–5188. https://doi.org/10.1109/JBHI.2022.3190293 (2022) Ju, J. et al. Incorporating multi-stage spatial visual cues and active localization offset for pancreas segmentation. Pattern Recognit. Lett. 170, 85–92. https://doi.org/10.1016/j.patrec.2023.05.004 (2023) Ju J, Guan Z, Lei H, et al. A general global and local pre-training framework for 3D medical image segmentation. IEEE J Biomed Health PP. https://doi.org/10.1109/JBHI.2023.3339176 (2023) Gayhart, M., Arisawa, H. Automated detection of healthy and diseased aortae from images obtained by contrast-enhanced CT scan. Comput Math Method M 2013:107871. https://doi.org/10.1155/2013/107871 Hata A, Yanagawa M, Yamagata K, et al. Deep learning algorithm for detection of aortic dissection on non-contrast-enhanced CT. Eur Radiol 31, 1151–1159. https://doi.org/10.1007/s00330-020-07213-w (2021) Chen, J.L., Tian, S.W., Yu, L. et al. A deep learning algorithm using contrast-enhanced computed tomography (CT) images for segmentation and rapid automatic detection of aortic dissection. Biomed Signal Proces 62, 102145. https://doi.org/10.1016/j.bspc.2020.102145 (2020) Singh V, Gorniak M D, Flanders M D, et al. Automated assessment of acute aortic dissection on thoracic CT using deep learning. Scientific Traditional Posters S261-S262. https://doi.org/10.1016/j.jvir.2018.12.711 (2019) Kazeminia S, Baur C, Kuijper A, et al. Gans for medical image analysis. Artif intell med 109, 101938. https://doi.org/10.1016/j.artmed.2020.101938 (2020) Yi X, Walia E, Babyn P. Generative adversarial network in medical imaging: A review. Med image anal 58, 101552. https://doi.org/10.1016/j.media.2019.101552 (2019) Santini, G., Zumbo, L.M., Martini, N., et al.  Synthetic contrast enhancement in cardiac CT with deep learning. Computer Vision and Pattern Recognition: https://doi.org/10.48550/arXiv.1807.01779 Gong, E. et al. Deep learning enables reduced gadolinium dose for contrast-enhanced brain MRI. J. Magn. Reson. Imaging 48, 330–340. https://doi.org/10.1002/jmri.25970 (2018) Chen C, Raymond C, Speier B, et al. Synthesizing MR image contrast enhancement using 3D high-resolution convNets. arXiv e-prints. https://doi.org/10.48550/arXiv.2104.01592 Olut, S. et al. Generative adversarial training for MRA image synthesis using multi-contrast MRI. Artif. Intell. Med. 11121, 147–154. https://doi.org/10.1007/978-3-030-00320-3_18 (2018) Chandrashekar A, Shivakumar N, Lapolla P, et al. A deep learning approach to generate contrast-enhanced computerised tomography angiograms without the use of intravenous contrast agents. Eur heart J 41: 156. https://doi.org/10.48550/arXiv.2003.01223 (2020) Oktay O, Schlemper J, Folgoc LL, et al. Attention U-Net: Learning where to look for the pancreas. Proc MIDL:1–10. https://doi.org/10.48550/arXiv.1804.03999 Gu R, Wang G, Song T, et al. CA-Net: Comprehensive attention convolutional neural networks for explainable medical image segmentation. IEEE Trans Med Imaging 40 699–711. https://doi.org/10.1109/TMI.2020.3035253 (2020) Wang Q, Wu B, Zhu P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks. IEEE/CVF Conference on Computer Vision and Pattern Recognition: 11531–11539. https://doi.org/10.1109/CVPR42600.2020.01155 (2020) Qin Z, Zhang P, Wu F, et al. FcaNet: Frequency Channel Attention Networks. arXiv preprint arXiv:2012.11879. https://doi.org/10.48550/arXiv.2012.11879 (2020) Li, Z., Zhang, C.X., Zhang, Y.Q., et al. CAN: Context-assisted full Attention Network for brain tissue segmentation. Med image Anal 85: 102710. https://doi.org/10.1016/j.media.2022.102710 (2023) Hu, J., Shen, L., Sun, G. Squeeze-and-excitation networks. Proceedings of the IEEE conference on computer vision and pattern recognition. 7132–7141. https://doi.org/10.1109/CVPR.2018.00745 Isensee F, Petersen J, Klein A, et al. nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation. arXiv (Cornell University) https://doi.org/10.48550/arXiv.1809.10486 Isola, P., Zhu, J.Y., Zhou, T. et al. Image-to-image translation with conditional adversarial networks. Proceedings of the IEEE conference on computer vision and pattern recognition. 1125–1134. https://doi.org/10.48550/arXiv.1611.07004 (2017) Zhu, J.Y., Park, T., Isola, P., et al. Unpaired image-to-image translation using cycle-consistent adversarial networks. IEEE International Conference on Computer Vision (ICCV). 2242–2251. https://doi.org/10.1109/ICCV.2017.244 (2017) Wang, T.C., Liu, M.Y, Zhu, J.Y. et al. High-resolution image synthesis and semantic manipulation with conditional GANs. IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8798–8807. https://doi.org/10.48550/arXiv.1711.11585 (2018) Download references The authors acknowledge Guangzhou Medical University Fujian Provincial Hospital and and Capital Medical University for the clinical data This work was partially supported by the Kunshan City Traditional Chinese Medicine (TCM) Science and Technology Development special fund (KZYY202302) and the Suzhou Medical Association “Imaging Medical Star” general project (2023YX-M04) School of Information Science and Technology College of Chemistry and Chemical Engineering Juanjuan YIN and Huijuan Tu designed the study Juanjuan Yin and Jinye Peng collected the clinical data and designed the experiment Juanjuan YIN performed the experiments and wrote the first draft of the manuscript Jun Wang and Huijuan Tu revised the manuscript This study was approved by institutional review boards of Beijing Anzhen Hospital Informed consent was obtained from all participants The study was conducted in accordance with relevant guidelines and regulations Download citation DOI: https://doi.org/10.1038/s41598-024-73515-4 Metrics details Missing pixel imputation presents a critical challenge in image processing and computer vision particularly in applications such as image restoration and inpainting The primary objective of this paper is to accurately estimate and reconstruct missing pixel values to restore complete visual information This paper introduces a novel model called the Enhanced Connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) which is designed to address two fundamental issues inherent in existing GAN architectures for missing pixel generation: (1) mode collapse which leads to a lack of diversity in generated pixels and (2) the preservation of pixel integrity within the reconstructed images ECP-IGANN incorporates two key innovations to improve missing pixel imputation an identity block is integrated into the generation process to facilitate the retention of existing pixel values and ensure consistency the model calculates the values of the 8-connected neighbouring pixels around each missing pixel thereby enhancing the coherence and integrity of the imputed pixels The efficacy of ECP-IGANN was rigorously evaluated through extensive experimentation across five diverse datasets: BigGAN-ImageNet the 2024 Medical Imaging Challenge Dataset These experiments assessed the model’s performance in terms of diversity with results demonstrating significant improvements in the Inception Score (IS) and Fréchet Inception Distance (FID) ECP-IGANN markedly enhanced image segmentation performance in the validation phase across all datasets were improved substantially for various segmentation models in the 2024 Medical Imaging Challenge Dataset the Residual Attention U-Net’s Dice Score increased from 0.84 to 0.90 while accuracy improved from 0.88 to 0.93 following the application of ECP-IGANN Similar performance enhancements were observed with the other datasets highlighting the model’s robust generalizability across diverse imaging domains the imputed pixels may require more fidelity and coherence with the surrounding content resulting in visually unconvincing results these methods need help to handle challenging scenarios such as occlusions Researchers have investigated advanced techniques including machine learning-based approaches and deep neural networks to overcome these limitations and achieve more accurate and realistic missing pixel imputation This paper proposes a novel approach for missing pixel imputation called 8-connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) The proposed approach builds upon the foundation of GANs while introducing two essential modifications to enhance the imputation process we incorporate an identity block within the pixel generation process preserving the existing pixel values and ensuring consistency with the original image we introduce a novel approach to evaluate the generated pixel values by comparing them with the weighted average of the 8-connected pixels surrounding the missing pixel and the actual value of the discriminator This evaluation allows us to accept or reject the generated pixels based on the degree of error effectively enhancing the coherence and accuracy of the imputed pixels The main contributions of this paper are summarized as follows we address the challenges of missing pixel imputation by proposing a novel 8-connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) The proposed method leverages the power of Generative Adversarial Networks (GANs) while introducing two essential modifications to enhance the imputation process Incorporating an identity block into the ECP-IGANN framework mitigates the mode collapse phenomenon enabling the generator to produce a diverse range of pixel values rather than repetitive outputs This enhancement is crucial for ensuring the variability in the generated pixels which is essential for high-quality image synthesis ECP-IGANN introduces a novel loss function called neutrosophic 8-connected pixel loss designed to uphold the spatial integrity among pixels in the generated images This loss function leverages the principles of neutrosophic logic allowing for a more nuanced evaluation of pixel relationships and improving the coherence of the imputed regions The architecture of ECP-IGANN permits the generation of pixel values without a discriminator thereby simplifying the computational process This design choice facilitates the integration of generated pixels with their surrounding context using fewer computational resources and a reduced number of iterations making the model particularly advantageous for real-time and low-latency applications where computational efficiency is paramount Comprehensive experiments conducted on different new datasets validated the efficacy of the proposed ECP-IGANN method in terms of diversity and missing pixel imputation The results reveal that ECP-IGANN not only generates diverse and accurate imputed pixels but also enhances the segmentation results for different architectures of U-Nets Section 2 provides a comprehensive review of related work on missing pixel imputation including different recent versions of GANs and their advantages and disadvantages 8-connected Pixel Identity GAN with neutrosophic (ECP-IGANN) and details the modifications we introduce for enhancing the imputation process including the incorporation of an identity block and the evaluation of generated pixel values using the weighted average of surrounding pixels Section 4 describes the experimental setup including the datasets used and evaluation metrics used to assess the performance of ECP-IGANN summarizing the contributions of the proposed method and discussing potential avenues for future research in the field of missing pixel imputation Missing pixel imputation refers to estimating and reconstructing missing values in an image to restore the image as accurately as possible It is a specialized subset of missing data imputation where the challenge lies in predicting the absent pixel values by leveraging available information from the surrounding image regions The critical challenge is to ensure that the imputed pixels are not only visually plausible but also consistent with the underlying structure and content of the image This process is pivotal in image restoration and analysis across various domains as it enables the replacement of missing values with statistically or algorithmically estimated substitutes various GAN-based approaches for missing pixel imputation have been proposed to improve the accuracy and quality of the imputed pixels Each GAN variant has distinct advantages and limitations; these limitations can lead to inaccurately generated pixels or require numerous iterations to achieve satisfactory results GAN-based imputation methods are trained on datasets containing images with missing pixels enabling the model to learn the data’s underlying patterns and structural characteristics The generator network synthesizes new pixel values which are subsequently evaluated against the original image by the discriminator network This adversarial training process iterates until the generated pixels closely approximate the original missing values resulting in a reconstructed image with minimal errors Yoon et al.26 introduced the Generative Adversarial Imputation Networks (GAIN) framework in 2018 leveraging the Generative Adversarial Network (GAN) paradigm to address the challenge of missing data imputation The core concept of GAIN is to generate imputed data that closely match the observed portions of an image using an element-wise discriminator This approach ensures that the imputed values are indistinguishable from the original data enhancing the restored image’s overall fidelity GAIN was empirically validated on five open-source datasets—Breast and News—sourced from the UCI Machine Learning Repository where it outperformed existing techniques in terms of imputation accuracy for data missing completely at random (MCAR) One notable disadvantage lies in the integrity of the pixel relationships in the generated images While GAIN attempts to preserve the local structure the global coherence between pixels may be compromised leading to inconsistencies or artifacts that detract from the image’s overall quality the stability of the GAIN training process is a critical concern GANs are inherently prone to instability during training often requiring careful tuning of hyperparameters and network architectures Another significant drawback is the risk of mode collapse a common issue in GANs where the generator produces a limited variety of outputs failing to capture the full diversity of the data distribution The generator network in this model employs convolutional layers with varying dilation rates and incorporates skip connections to enhance the reproduction of fine details in the output This architecture allows the model to serve as a universal imputation tool capable of handling diverse missingness scenarios with high quality The model was experimentally validated for three distinct missingness scenarios: randomly missing pixels and a single large square centrally located in the image The proposed model demonstrated superior inpainting results in all scenarios as evaluated using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) on two real-world benchmark datasets The performance was compared favourably against biharmonic imputation and several state-of-the-art image-inpainting methods a prevalent issue in GAN frameworks where the generator may produce a limited range of outputs The mode collapse can reduce the diversity in the imputed data potentially limiting the model’s effectiveness in handling a broad array of missingness patterns Another disadvantage lies in the integrity of the pixel relationships in the generated images Although the model is designed to reproduce fine details there can be inconsistencies in the global coherence of pixel arrangements model stability during training presents a significant challenge are known for their susceptibility to instability often requiring precise tuning of hyperparameters and network structures to achieve convergence Their method addressed the inherent complexity of natural images which has challenged traditional imputation techniques effectively captured the intricate relationships between pixels and improved the quality of imputed images They emphasized the importance of maintaining spatial coherence and contextual integrity and successfully mitigating common issues such as mode collapse and training instability the proposed method still faces limitations including difficulties preserving pixel integrity which sometimes results in artifacts or unnatural transitions in the imputed areas where the generator could produce a limited variety of outputs reducing the diversity of the imputed images Reliance on extensive training data also hindered performance in scenarios with limited available data impacting the overall effectiveness of the imputation process The proposed method first treated each view as a separate domain identifying domain-to-domain mappings through a GAN using randomly sampled data from each view a multimodal denoising autoencoder (DAE) was employed to reconstruct the missing view from the GAN outputs based on paired data across the views By jointly optimizing the GAN and DAE models the models facilitate knowledge integration for domain mappings and view correspondences Empirical results on benchmark datasets validated the proposed VIGAN approach demonstrating its superiority over state-of-the-art methods the evaluation of VIGAN in a genetic study of substance use disorders underscored its effectiveness and usability in life sciences the VIGAN approach has several limitations One significant drawback was the potential for mode collapse where the GAN could produce a restricted range of outputs limiting the reconstructed views’ diversity the reliance on paired data across views could be a constraint in scenarios where such data is sparse or unavailable the complexity associated with optimizing both the GAN and DAE introduced challenges in training stability which could affect the overall performance of the imputation process particularly in environments with communication failures or cyberattacks Their approach aims to generate long-term time-series data from widely existing spatio-temporal modal data and impute missing values by replacing them with generated data The proposed CM-GAN demonstrated superior performance on a photovoltaic (PV) power output dataset outperforming baseline models and achieving state-of-the-art results They conducted extensive ablation studies to validate the contribution of the cross-modal data fusion technique and the reasonableness of the parameter settings The model exhibited challenges related to mode collapse where the generator produced less diverse outputs thereby compromising the stability of the training process the integrity between imputed pixels was sometimes compromised leading to inconsistencies in the generated data that could affect downstream tasks specifically those with over an 80% missing rate Their approach generates multiple candidate imputations and integrates a confidence prediction mechanism significantly outperforming traditional methods in handling high-dimensional and complex image data The proposed GAN-based architecture effectively offered robust statistical inference and managed varying degrees of missing data introduced two GAN-based methods for imputing missing numerical datasets where the imputed values are determined using Euclidean distance These methods outperformed conventional imputation techniques and maintained robustness across missing data scenarios both approaches face challenges related to mode collapse where the diversity of generated imputations is compromised issues related to the stability of the training process and the integrity between imputed pixels were noted potentially impacting the consistency and accuracy of the imputed data Drawing inspiration from the success of Generative Adversarial Networks (GANs) in image generation they proposed a method to learn the overall distribution of a multivariate time-series dataset using GANs which could then be utilized to generate missing values for each sample time series data are often incomplete due to the nature of the data recording process The authors employed a modified Gated Recurrent Unit (GRU) within the GAN framework to model the temporal irregularity of incomplete time series Their experiments on two multivariate time series datasets demonstrated that the proposed model outperformed baseline methods regarding imputation accuracy the experimental results revealed that a simple model applied to the imputed data achieved state-of-the-art performance on prediction tasks highlighting the effectiveness of their approach in downstream applications the method encountered challenges with mode collapse where the diversity of the generated imputations was limited reducing the variety and reliability of the generated data training stability and the integrity between imputed values were observed potentially affecting the consistency and overall quality of the imputation results The related work methods often struggle with issues such as mode collapse and poor spatial coherence between the imputed and surrounding pixels These approaches tend to create results that do not integrate well with their context leading to visually unconvincing outputs and increased computational costs due to numerous iterations and adjustments This section presents the methodology of ECP-IGANN The model solves the problems mentioned in related works ECP-IGANN has three main parts: a generator to generate the missing part of the image a discriminator to take the actual value as input and an 8-connected neutrosophic part to calculate the neutrosophic weighted average for the 8-connected pixels We integrate the identity block with the generator to guarantee the diversity of the generated pixels The ECP-IGANN (8-connected Pixel Identity GAN with Neutrosophic) model addresses the challenges of missing pixel imputation particularly issues related to mode collapse and maintaining the spatial coherence of generated images The architecture of ECP-IGANN consists of three main components: a generator The generator creates missing pixel values by leveraging an identity block that preserves the essential features of the input image ensuring that the generated pixels are consistent with the original image The discriminator evaluates the generated images against authentic images distinguishing between real and fake pixels to guide the generator in producing more accurate and coherent outputs A key innovation of the ECP-IGANN model is integrating an identity block into the generator This identity block allows the model to maintain critical spatial information from the input image while generating new pixel values By incorporating both 1 × 1 and 3 × 3 convolutional layers the identity block captures the image’s local and global features The 1 × 1 convolutions help maintain the spatial dimensions ensuring that critical relationships between neighbouring pixels are preserved while the 3 × 3 convolutions focus on extracting finer details This dual approach allows the generator to produce more diverse and accurate pixel values effectively mitigating the mode collapse problem common in traditional GANs Another significant component of the ECP-IGANN model is the neutrosophic-based 8-connected pixel loss function The loss function is designed to ensure the spatial coherence of the generated pixels by considering the relationships between each pixel and its eight connected neighbours The neutrosophic approach evaluates the generated pixels’ truth providing a more nuanced assessment of the generated images By comparing these values with those from the actual images the loss function ensures that the generated pixels are accurate and contextually consistent with their surrounding pixels thereby maintaining the overall integrity of the image The training process of the ECP-IGANN model involves a feedback loop in which the loss calculated by the discriminator is used to update the generator’s parameters The goal is to minimize loss and guide the generator to produce images that are indistinguishable from real ones The model employs a stopping criterion based on the convergence of the generated and actual pixel values which is determined by the neutrosophic loss function This approach allows the model to focus on areas where the generated image significantly deviates from the actual image leading to more efficient and targeted training An identity block stabilizes the training process by ensuring adequate gradient flow which is crucial for robust learning dynamics There is a Feedback Loop where the loss calculated from the discriminator’s output updates the generator’s parameters. The generator is trained to minimize this loss and strives to produce images that the discriminator cannot distinguish from real ones. As training progresses, the generator becomes increasingly adept at filling in the missing pixels while the discriminator continually refines its ability to distinguish real and fake images. Block diagram of the ECP-IGANN framework This section presents a method for calculating the loss in a Generative Adversarial Network (GAN) This section presents a method to calculate the loss in a Generative Adversarial Network (GAN) using a novel approach based on neutrosophic sets By associating each pixel value with corresponding membership values for truth (T) we can enhance the representation of pixel intensity thereby allowing for more nuanced comparisons between generated and real images The proposed process incorporates an 8-connected pixel neighbourhood comparison which enables the GAN to consider spatial relationships and improve the quality of generated images and (3) to convert each pixel value into a neutrosophic set Algorithm (2) presents a method to calculate the loss in a generative adversarial network (GAN) by comparing the 8-connected pixel neighbourhoods in both generated and real images The process begins by iterating over each pixel in the generated image and identifying its 8-connected neighbourhood and the corresponding neighbourhood in the real image Each pixel in these neighbourhoods is converted into a neutrosophic set consisting of three components: True (T) These components represent different levels of certainty regarding the pixel intensity the algorithm calculates the average values of these components across the entire neighbourhood for both generated and real images The critical aspect of this method is the comparison of these neutrosophic averages Suppose the difference between the averages of the generated and real neighbourhoods is smaller than a specified threshold across all components (T the algorithm stops further loss calculation for that pixel considering it sufficiently close to the real pixel the algorithm calculates the loss based on the squared differences between the neutrosophic components of the generated and real neighbourhoods This loss is then added to the total loss of the image This approach ensures that GAN considers individual pixel accuracy and spatial relationships within an image thereby encouraging the generation of visually and structurally coherent images Using neutrosophic sets allows for a more flexible representation of pixel values thus accounting for uncertainty and indeterminacy The stopping criterion based on neutrosophic averages helps optimize the learning process focusing on areas where the generated image deviates significantly from the real image Identity block in GANsIn the 8-connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) the identity block plays a central role in enhancing the model’s ability to perform missing pixel imputation The identity block is strategically positioned within the generator network in the 8-connected Pixel Identity GAN architecture with Neutrosophic (ECP-IGANN) it follows the initial convolutional layers responsible for feature extraction from the input image By placing an identity block at this juncture we ensure the generator can effectively integrate the raw pixel values and the learned features extracted from the image This placement enriches the generated output while maintaining the spatial integrity of the original image The output from the identity block is then fed into subsequent layers of the generator facilitating coherent pixel predictions that closely align with the surrounding context This careful integration enhances the model’s ability to produce diverse and high-quality reconstructions making it a pivotal element in the overall architecture of ECP-IGANN the identity block employs a skip connection that adds the original pixel values to the processed output This integration ensures that critical information is not lost and preserves image integrity during generation The enriched features are then forwarded to the generator which leverages this information to predict and fill in missing pixels accurately the identity block contributes to the overall stability of training by facilitating adequate gradient flow This mitigates the vanishing gradient problem frequently encountered in GANs allowing for more robust learning dynamics By enhancing pixel diversity and coherence the identity block significantly improves the generator’s capability to produce high-quality and realistic images thereby addressing the challenges of missing pixel imputation A vital feature of an identity block is its skip connection which allows the original input to bypass specific layers and be added directly to the output This mechanism ensures that critical pixel values are preserved and maintains visual coherence in the generated images the input image flows through an identity block where the convolutional layers extract relevant features while enriching the data with learned information The combination of features is then fed into the generator which uses this enriched data to predict and fill in missing pixels The identity block offers several benefits including preserving critical information from the input Allowing gradients to flow more effectively mitigates the risk of mode collapse a common issue in traditional GAN architectures The result is a generator that can produce various outputs ultimately improving the quality of imputed images the identity block is pivotal in addressing missing pixel imputation challenges and significantly contributes to the effectiveness of ECP-IGANN in generating high-quality image reconstructions The RMSE ensures accurate reconstruction of missing pixels and IS and FID provide insights into the generated samples’ diversity which contains fashion images with instances of missing pixels This dataset was particularly useful for exploring image completion tasks because the annotations allowed us to evaluate our model’s segmentation performance across various fashion items To comprehensively assess the performance of the proposed model we tested our primary model against seven other architectures focusing on their diversity and ability to handle missing pixel imputation we employed various U-Net architectures to compare segmentation performance before and after applying missing pixel imputation techniques This approach provided valuable insights into the strengths and weaknesses of different models in addressing the complexities associated with missing pixels in medical and remote sensing fields Table 2 presents the pixel imputation results for ECP-IGANN and seven recent GAN models for missing pixel imputation for five different datasets in terms of RMSE The results show the efficiency of ECP-IGANN and WGANs compared with other versions of GANs for missing data imputation regarding RMSE They also demonstrate the outperformance of ECP-IGANN compared with different models The table compares RMSE values for eight models across five datasets evaluating their effectiveness in missing data imputation ECP-IGANN consistently outperforms the other models highlighting its superior accuracy in imputing missing pixels significantly lower than the closest competitor These results demonstrate ECP-IGANN’s exceptional handling of complex and diverse image data In the 2024 Medical Imaging Challenge Dataset ECP-IGANN again leads with an RMSE of 0.132 which have higher RMSEs of 0.1839 and 0.184 underscoring its capability to maintain critical image details in clinical applications ECP-IGANN also excels in the Autonomous Vehicles dataset with an RMSE of 0.0697 highlighting its effectiveness in accurately imputing missing data in dynamic and safety-critical environments demonstrating its strength in managing spatial coherence and large-scale imputation tasks ECP-IGANN records the lowest RMSE of 0.0491 showing its proficiency in restoring missing details in datasets where fine textures and patterns are critical Compared to other models, such as WGAIN and CM-GAN, which generally perform well, ECP-IGANN’s consistently lower RMSE across all datasets highlights its enhanced capabilities, making it a highly effective tool for missing pixel imputation in diverse applications. Figure 2 shows the outperformance of the ECP-IGANN for the seven different models. RMSE comparison across five different datasets PSNR comparison across five different datasets FID comparison across five different datasets these results position ECP-IGANN as the leading model providing the most effective balance of image quality and accuracy across various challenging datasets IS comparison across four different datasets The experimental results highlight the effectiveness of ECP-IGANN in improving segmentation performance across different datasets and models. Initially, the segmentation results before applying ECP-IGANN, as shown in Tables 7 and 8 indicate that models such as Spatial Attention U-Net and Residual Attention U-Net achieved respectable but varied performance in terms of Dice Score and Recall on both the 2024 Medical Imaging Challenge Dataset and the 2024 Satellite Imagery Dataset Residual Attention U-Net demonstrated the highest Dice Score and Accuracy across both datasets emphasizing its ability to handle complex segmentation tasks However, after integrating ECP-IGANN for missing pixel imputation, there was a noticeable enhancement in the segmentation metrics, as presented in Tables 9 and 10 and Recall values for all models improved significantly with the Residual Attention U-Net reaching a Dice Score of 0.90 and an Accuracy of 0.93 on the Medical Imaging dataset and similar improvements were observed in the Satellite Imagery dataset These results validate the hypothesis that the superior pixel imputation provided by ECP-IGANN leads to more coherent and accurate image reconstruction thereby improving downstream tasks like segmentation This comprehensive validation across multiple models and datasets demonstrates the robust applicability of ECP-IGANN Consistent improvements across all metrics—Dice Score and Recall—underscore the model’s potential to enhance performance in various challenging imaging domains making it a valuable tool in medical and remote sensing applications we have proposed an 8-connected Pixel Identity GAN with Neutrosophic (ECP-IGANN) to address the challenges in missing pixel imputation The proposed method enhances the traditional GAN architecture by introducing two key innovations: integrating an identity block into the generator and applying a neutrosophic-based loss function that evaluates the spatial coherence of the generated pixels The identity block plays a crucial role in preserving the essential features of the original image during the generation process By allowing a parallel flow of information from input to output the identity block ensures that the generated pixels retain critical spatial relationships and do not deviate from the original content This addressing helps mitigate the mode collapse problem where the generator typically produces limited and repetitive outputs The neutrosophic-based 8-connected pixel loss function is another significant contribution of this work This loss function goes beyond simple pixel-wise comparison by evaluating the coherence of each pixel within its 8-connected neighbourhood the loss function provides a more nuanced and accurate assessment of the pixels in images generated by GANs with identity block This integration ensures that the imputed pixels match the surrounding context and maintain the overall spatial integrity of the generated image Experiments were conducted using new datasets and the results proved the model’s efficiency in diversity The results also demonstrate the efficiency of using the model as a preprocessing step for the different architectures of U-Nets in enhancing the segmentation accuracy Although ECP-IGANN demonstrates superior performance in missing pixel imputation Including the identity block and neutrosophic loss function increases the computational complexity of the model which may result in longer training times and higher resource consumption Future research could focus on optimizing the computational efficiency of ECP-IGANN perhaps by simplifying the architecture or exploring alternative loss functions that maintain performance while reducing complexity Another avenue for future work could be the adaptation of ECP-IGANN for different types of data where the spatial coherence of generated pixels is equally crucial The datasets and code generated during the current study are available from the corresponding author upon reasonable request EdgeConnect: generative image inpainting with adversarial edge learning Missing value imputation methods for electronic health records A review of image inpainting methods based on deep learning Missing traffic data imputation for artificial intelligence in intelligent transportation systems: review of methods Image inpainting via generative multi-column convolutional neural networks Proc of the 32nd International Conference on Neural Information Processing Systems (NIPS’18) Cross-to-merge training with class balance strategy for learning with noisy labels Deep contextual attention for image inpainting In Advances in Neural Information Processing Systems (NeurIPS) 2672–2680 (2024) Globally and locally consistent image completion Structural-GAN: structured generative adversarial networks for missing pixel recovery Multimodal generative adversarial network for robust image inpainting ContextNet-GAN: contextual generative adversarial network for image completion Image inpainting using contextual attention-GAN PatchGAN: adversarial inpainting of irregular holes in images Free-form image inpainting with gated convolution Generative adversarial networks (GANs): introduction Generative Adversarial Networks Assist Missing Data Imputation: A Comprehensive Survey and Evaluation A style-based generator architecture for generative adversarial networks Conditional generative learning for medical image imputation A comprehensive framework for improving remote sensing image classification: combining augmentation and missing pixel imputation In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Restoration of artwork using deep neural networks Reviewing autoencoders for missing data imputation: technical trends A survey of missing data imputation using generative adversarial networks GAIN: missing data imputation using generative adversarial nets Wasserstein generative adversarial imputation networks (WGAIN) for image inpainting A GAN-based framework for image imputation: preserving spatial coherence in missing pixel recovery In Advances in Neural Information Processing Systems (NeurIPS) 2136–2145 (2020) View imputation via generative adversarial networks for multi-view data CM-GAN: a cross-modal generative adversarial network for imputing completely missing data in digital industry GAMIN: generative adversarial multiple imputation network for highly missing data Multivariate time series imputation with generative adversarial networks Novel framework for generating criminals images based on textual data using identity GANs Stabilizing and improving training of generative adversarial networks through identity blocks and modified loss function Image quality assessment: from error visibility to structural similarity Advances in Neural Information Processing Systems (NIPS) Fast and robust algorithm for super-resolution using an adaptive edge-preserving method GANs trained by a two time-scale update rule converge to a local Nash equilibrium Large scale GAN training for high fidelity natural image synthesis Medical imaging segmentation challenge: dataset and benchmark Creating synthetic data for autonomous vehicle training: challenges and solutions The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients Fashion image recognition using deep learning: a comprehensive dataset Research on MRI brain tumor image segmentation based on the spatial attention U-Net H-DenseUNet: hybrid densely connected UNet for liver and tumor segmentation from CT volumes UNet++: a nested U-Net architecture for medical image segmentation Attention U-Net: learning where to look for the pancreas Residual dense U-Net for medical image segmentation RANet: an end-to-end recurrent attention network for skin lesion segmentation in dermoscopy images Download references This work was funded by the Researchers Supporting Project Number (RSP2024R509) College of Computer and Information Sciences The original draft preparation involved contributions from M All authors have read and reviewed the manuscript Download citation DOI: https://doi.org/10.1038/s41598-024-73976-7 Metrics details As network intrusion behaviors become increasingly complex traditional intrusion detection systems face limitations we introduce the Nash equilibrium concept from game theory into classifier ensemble optimization enhancing robustness in multi-class classification tasks we propose a network intrusion detection system based on a Conditional Generative Adversarial Network with Conditional Aggregation Encoder-Decoder Structure (CE-GAN) with a conditional aggregation encoder-decoder structure to mitigate data imbalance and improve classifier performance The model incorporates a composite loss function to maintain both the authenticity and diversity of generated samples Experiments on the NSL-KDD and UNSW-NB15 datasets show that CE-GAN effectively augments rare data samples significantly improving classification metrics for imbalanced datasets thus providing a superior solution to this challenge in network intrusion detection which addresses class imbalance by incorporating class signals into the GAN architecture While this enhances classification performance for minority classes maintaining recognition capability for emerging attack behaviors requires continual optimization aiming to address data imbalance and improve detection performance By using an autoencoder as a feature extractor the data dimensionality is effectively reduced while retaining essential feature information overlapping feature characteristics among certain attack categories might prevent the generative model from fully distinguishing between different categories impacting the detection accuracy of minority classes These studies demonstrate that using GANs for data augmentation shows great potential for addressing data imbalance in NIDS continual optimization and innovation in GAN models are necessary to enhance their ability to detect new types of attacks Existing data augmentation algorithms mainly rely on Generative Adversarial Networks (GANs) these algorithms lack effective conditional constraints during the training process which leads to imbalanced generation results further exacerbating the issue of data imbalance during augmentation Most existing algorithms use simple loss strategies to guide model training for complex Generative Adversarial Networks (GANs) more sophisticated loss strategies are often needed to constrain the generator ensuring the quality of the generated results while simultaneously accelerating the training process Current Generative Adversarial Network (GAN) algorithms typically use relatively simple network structures to train the generator this approach struggles to capture deep features The model requires a more concise data representation method to facilitate feature learning and model training To address the limitations of single classification algorithms in handling multi-class classification tasks with poor performance metrics the concept of Nash equilibrium from game theory is introduced This enables the optimal strategy combination of classifiers collectively improving overall performance By optimizing the classification algorithm’s ability to detect minority class attack samples this approach effectively mitigates the data imbalance issue and resolves the problem of artificially inflated classification metrics To address the issue of data imbalance that persists in existing data augmentation algorithms a Conditional Generative Adversarial Network (CGAN) is introduced this approach ensures both the balance and diversity of the generated results effectively solving the problem of inflated classification metrics caused by data imbalance To address the limitations of existing simple loss strategies a composite loss strategy is introduced to guide the training of Generative Adversarial Networks (GANs) This strategy constrains the generator’s training from multiple dimensions ensuring the quality of the generated results while simultaneously speeding up the model’s training process To address the difficulty Generative Adversarial Networks (GANs) face in learning superficial features we propose a conditional constraint-based encoding representation method This approach facilitates the training of imbalanced data by ensuring the model learns deep features while addressing typical issues with conventional encoding methods the dataset contains a total of 41 network traffic features along with corresponding attack types and their significance This dataset covers the key characteristics of network traffic facilitating the analysis of the relationship between feature importance and attack types The features are mainly divided into four categories: basic features Analysis of NSL-KDD dataset categories Analysis of UNSW-NB15 dataset categories The results show that all three algorithms perform well in overall metrics effectively distinguishing between normal and abnormal samples when it comes to multi-class classification the performance of all three classifiers is less satisfactory the performance of these algorithms varies significantly Ensemble learning classifier model under game theory strategy At the start of training,three classifiers are initialized: the Random Forest Classifier The initial weights for each classifier are as follows: \(w_i\) satisfy \(w_1+w_2+w_3=1\).For each classifier Assume the error rates for each classifier are as follows: \(error_k\) \(x_i\) represents the feature vector of the i-th sample \(y_i\) represents the true label of the i -th sample \(\textrm{II}\) is an indicator function,When the classifier’s prediction \(M_1\) for sample \(x_i\) differs from the true label \(y_i\) ,the value of \(\textrm{II}\) is 1,otherwise The payoff for each classifier is calculated based on its error rate where a lower error rate results in a higher payoff The weights of the three classifiers are defined as follows \(w_1,w_2,w_3\) .The weights are adjusted based on the payoff of each classifier \(w _ { i } ^ { ( t ) }\)represents the weight of the i-th classifier at the t-th iteration \(p a y o f f _ { k } ^ { ( t ) }\) represents the payoff of the i-th classifier at the t-th iteration Repeat the steps of calculating error rate and updating the strategy until the change in weights is smaller than a small threshold \(\epsilon\) ,indicating that Nash equilibrium has been reached: the final prediction result is the weighted average of the predictions from each classifier: \(w _ {rf} ^ { * },w _ {et} ^ { * },w _ {gb} ^ { * }\) represents the optimal weight of the i -th classifier in the Nash equilibrium state To better demonstrate the performance advantages of our proposed classifier, we compared its performance on the KDDTrain+ and KDDTest+ datasets. As shown in Table 3 our optimization strategy outperforms traditional single classifiers in classification tasks particularly in the recognition of rare samples where it exhibits significant performance improvement This further proves the effectiveness and practicality of optimizing classifier combinations using game theory strategies Furthermore, we have listed the various metrics for the multi-class tasks in ensemble learning, as shown in Table 4 The overall accuracy of the model reaches 99.62% which is excellent compared to similar fields when specifically distinguishing each subclass the classification performance for smaller categories is significantly lower due to the limited representation in the dataset the failure to detect any type of attack can potentially lead to serious consequences due to the impact of the extremely small representation of certain categories the performance metrics for these classes fluctuate significantly Take loadmodule as an example: since there is only one instance recorded in the dataset the test accuracy for this category can only be 0 Therefore when evaluating classification models using datasets the overall evaluation metrics are not suitable for describing rare samples data augmentation algorithms are needed to expand the number of rare samples allowing the model’s performance to be evaluated across a larger Generative Adversarial Networks (GANs) become the best tool for data augmentation Through the adversarial training between the generator and discriminator networks the generator is able to create realistic data samples effectively compensating for the lack of rare class samples in the dataset This approach not only increases the number of rare samples but also preserves the diversity of the samples making the classification model’s performance on minority classes more stable and reliable we can significantly expand the number of rare samples without altering the structure of the original dataset resulting in more representative performance metrics when evaluating classification models GAN-generated samples retain the key features of the original data avoiding the risk of model overfitting that can occur with simple duplication or excessive augmentation the application of GANs not only overcomes the limitations of traditional data augmentation methods but also lays a solid foundation for further improving the generalization ability of classification models Data augmentation flowchart of generative adversarial network CE-GAN consists of four key components: an encoder While the encoder-decoder pair handles dimensional transformation the generator creates samples based on low-dimensional features and the discriminator evaluates both sample authenticity and conditional constraints The system is trained through multiple loss functions and incorporates modules such as normalization and classifier evaluation to achieve conditional generation In data augmentation tasks, enhancing the diversity of the augmented data should be the primary goal. Therefore, we conducted a comprehensive comparative analysis between the improvement in classification performance and the similarity between the generated and original samples. The pseudocode of the CE-GAN model is shown in Table 5 and the improvement in classification model metrics serves as evidence of the effectiveness of the data augmentation algorithm we employed three evaluation metrics for data augmentation: Precision-Recall Distance (PRD) and Mean Absolute Error (MAE) to validate and quantify the data augmentation effect of CE-GAN These generation metrics must remain within a reasonable range-neither too high This ensures that the generated samples strike the optimal balance between diversity and authenticity we introduced multiple loss functions to constrain the training outcomes Each loss function plays a crucial role in different parts of the model contributing to the overall effectiveness of the training process c represents the conditional information,G represents the Generator Reconstruction loss ensures effective data dimensionality reduction and restoration between the Encoder and Decoder. By combining data with conditional information, it enlarges the distribution gap between normal and anomalous samples during the dimensionality reduction process. The reconstruction loss is described in Eq. (7): Diversity loss is used to ensure the diversity of the generated samples, preventing overfitting and thus maintaining the significance of the Generative Adversarial Network (GAN). It also helps to make the generated samples as realistic as possible. The form of the diversity loss is described in Eq. (8): \(\sqrt{\sum _{k=1}^{d} \left( X_{i,k} - X_{j,k} \right) ^2}\) represents the Euclidean distance between samples \(X_i\)and \(X_j\) represents the minimum distance between each sample \(X_i\) and other samples \(X_j\) \(\mathcal {L}_{\text {cst}}\) represented in the form of a small negative mean in the overall training process it prevents the generated samples from losing authenticity in the pursuit of model diversity Temporal loss is used to ensure that the generated samples in the low-dimensional space are overall similar to the original data, enhancing the reliability of the generated data. The form of this loss is described in Eq. (9): By combining the four loss functions, the model’s training process is constrained from multiple aspects, accelerating the convergence speed and ensuring both the authenticity and diversity of the generated samples. This balance between authenticity and diversity enhances the quality of the samples generated by the CE-GAN. Detailed structure of the CE-GAN Model As shown in Fig. 6 we have outlined the model architectures used by each network in CE-GAN Given the high complexity and diversity inherent in network attacks the multi-head attention mechanism of the Transformer allows for flexible extraction of rich feature representations from the input data To ensure that both the Generator and Discriminator in the CE-GAN model have equal capabilities we utilized the Transformer architecture for both enabling them to handle complex data with consistent feature extraction and representation abilities Through the multi-head attention mechanism the model can focus on different parts of the data simultaneously identifying hidden attack features within a broader feature space This capability is crucial for recognizing complex network attack behaviors MAE typically represents the absolute error between the generated samples and the real samples, providing a straightforward reflection of the quality of the generated samples. The formula is shown in Eq. (12) the higher the quality of the generated samples MAE is equally sensitive to all error values making it a relatively stable and straightforward measure of error In this section, we first validate the effectiveness of the model through K-fold cross-validation experiments. Using PRD, RMSE, and MAE as the three evaluation metrics, the experiments focus on augmenting rare attack samples in individual classes. As shown in Table 6 we combined buffer_overflow and normal samples to create new datasets with \(K-1\) parts used as the training set and 1 part as the test set thereby cross-validating the model’s effectiveness and generalization the CE-GAN model generated network attack samples with an average PRD of 64.3868 These generation metrics confirm the effectiveness of the CE-GAN model In the experiment, we conducted multiple tests on the composition of the Generator’s loss values. The Generator’s loss function is formulated as shown in Eq. (13) where \(\alpha\) and \(\beta\) are used to control the loss coefficients,typically set to \(\alpha =0.1,\beta =0.01\) the balance between the two components remains stable and does not significantly impact the overall training process the degree of diversity should be controlled within a smaller range to minimize its impact on the training process mainstream approaches predominantly rely on Generative Adversarial Networks (GANs) typically either inputting the raw dataset directly into the model or applying conditional constraints to the GAN these models commonly face a significant issue: the imbalance in the input data directly affects the learning process of the GAN leading to the generated distribution being similarly affected by the imbalance problem this paper aims to improve this issue by addressing it from the model input stage enhancing the learning distribution of the Generative Adversarial Network (GAN) By applying conditional encoding to the input data the boundaries between different categories can be distinguished after dimensionality reduction preventing them from being mixed together during the encoding process the five main attack categories and their five subcategories in the NSL-KDD dataset are used to verify the changes in data distribution before and after applying conditional encoding T-SNE visualization of multi-class changes in NSL-KDD with conditional encoding T-SNE visualization of single-class changes in NSL-KDD dataset with conditional encoding makes the distribution boundaries between different categories clearer effectively mitigating the impact of data imbalance on model performance Similarly, we performed single-class data augmentation on the UNSW-NB15 dataset using the same approach. We selected six types of rare attack categories from the dataset: Worms, Shellcode, Backdoor, Analysis, Reconnaissance, and Dos. The results are shown in Table 8 the phenomenon observed is similar to that in the NSL-KDD dataset where the augmentation ratio reflects the degree of data scarcity After completing the augmentation of the two datasets, we summarized the distributions of the augmented datasets in Table 9 The NSL-KDD augmented dataset showed significant changes in sample data with categories such as Spy expanding from single digits in the original dataset to hundreds This increase in recorded data was achieved without overshadowing other categories and the overall proportion of anomalies in the dataset remained around 50% the UNSW-NB15 dataset saw a notable increase in sample data effectively addressing the classification model’s performance issues Proportion of different categories in the NSL-KDD dataset Proportion of different categories in the UNSW-NB15 dataset. T-SNE visualization of the NSL-KDD dataset before and after augmentation Performance comparison of classification metrics before and after CE-GAN data augmentation Transformer is used as the primary component in all four parts: Encoder To validate the effectiveness of the proposed network architecture we used the NSL-KDD dataset as an example and replaced the Transformer with more commonly used modules such as GRU we tested the removal of the Encoder and Decoder we verified the effectiveness of CE-GAN while demonstrating the advantage of using Transformer in all four components of the model To comprehensively evaluate the performance of CE-GAN, we conducted systematic comparisons with representative variants from the GAN family. The selected benchmark models include DCGAN, WGAN, CGAN, and InfoGAN. We employed three key metrics - PRD, RMSE, and MAE - for evaluation. As shown in Table 12 the experimental results demonstrate that CE-GAN achieves optimal performance with PRD and MAE values of 64.3868 and 0.1264 Although WGAN shows a slight advantage in RMSE (0.1932) CE-GAN exhibits more balanced overall performance These results validate the effectiveness of our conditional encoding and composite loss function while highlighting CE-GAN’s unique advantages in addressing data imbalance issues in network intrusion detection artificial intelligence has made significant strides in the field of network attacks with machine learning algorithms achieving effective results in tasks such as detecting and classifying network attacks This progress largely depends on the quality of network attack datasets we first analyzed the imbalance in the distributions of the NSL-KDD and UNSW-NB15 datasets and proposed a game theory-based classification algorithm to address the imbalance improving classification performance to some extent the model struggled to recognize rare samples we introduced a Conditional Encoding-based Generative Adversarial Network (CE-GAN) which combines an encoder-decoder structure with conditional GAN Unlike traditional encoder-decoder structures we incorporated a conditional aggregation method to prioritize the separation of different categories during the learning process effectively addressing the imbalance issue The model was trained using a composite loss function to ensure both the authenticity and diversity of the generated samples while Transformer architecture was used in all components to enhance feature extraction and MAE metrics and compared classification results before and after data augmentation The results demonstrated the effectiveness of CE-GAN ablation experiments further confirmed the value of the conditional aggregation encoder-decoder method providing an effective approach for improving network attack datasets.In future work we plan to evaluate the performance of the CE-GAN model on more representative network attack datasets to further validate its generalization capabilities These datasets encompass diverse attack characteristics across various network environments which will help comprehensively assess the model’s adaptability and effectiveness in different network attack scenarios we aim to explore the potential application of the model in real-time network traffic analysis providing more reliable technical support for cybersecurity defense The limitations of this study are primarily reflected in the following aspects although the CE-GAN model effectively addresses the issue of data imbalance the diversity and authenticity of the generated samples remain somewhat constrained particularly when dealing with extremely rare samples where the model’s performance may not be ideal The experiments were conducted only on the NSL-KDD and UNSW-NB15 datasets These datasets may not fully represent attack characteristics in other complex network environments thus limiting the model’s generalizability the complexity of the model is relatively high especially in practical applications where computational resources and training time could pose challenges its effectiveness in resource-constrained environments may be limited This study utilizes two publicly available datasets: NSL-KDD and UNSW-NB15. The NSL-KDD dataset is provided by the NSL-KDD project and can be accessed via the following link: NSL-KDD. The UNSW-NB15 dataset is provided by the University of New South Wales’ Cybersecurity Research Center, with detailed information and download options available at: UNSW-NB15 Adversarial machine learning for network intrusion detection systems: A comprehensive survey Adversarial machine learning in network intrusion detection systems Improving the reliability of network intrusion detection systems through dataset integration A comprehensive survey of generative adversarial networks (gans) in cybersecurity intrusion detection A review of generative adversarial networks and its application in cybersecurity A wasserstein generative adversarial network-gradient penalty-based model with imbalanced data enhancement for network intrusion detection Abnormal traffic detection: Traffic feature extraction and dae-gan with efficient data augmentation Mcgan: modified conditional generative adversarial network (mcgan) for class imbalance problems in network intrusion detection system Aec_gan: unbalanced data processing decision-making in network attacks based on acgan and machine learning Gru-gbm: A combined intrusion detection model using lightgbm and gated recurrent unit Apelid: Enhancing real-time intrusion detection with augmented wgan and parallel ensemble learning Nads-ra: network anomaly detection scheme based on feature representation and data augmentation An enhanced ai-based network intrusion detection system using generative adversarial networks Boosting methods for multi-class imbalanced data classification: an experimental review Learning equilibria in symmetric auction games using artificial neural networks Bat: Deep learning methods on network intrusion detection using nsl-kdd dataset Intrusion detection system for nsl-kdd dataset based on deep learning and recursive feature elimination Performance analysis of intrusion detection systems using a feature selection method on the unsw-nb15 dataset Download references This research was supported by the Science and Technology Research Projects of the Jilin Provincial Department of Education under Grant Numbers JJKH20241705KJ and JJKH20241706KJ We sincerely thank the Jilin Provincial Department of Education for their generous support Changchun University of Science and Technology Jilin Province Advanced Control Technology and Intelligent Automation Equipment R&D Engineering Laboratory Beijing Aerospace Era Laser Navigation Technology Co. Yang Yang led the overall research project contributed to the conceptualization and design of the study including data analysis and model development Xiaoyan Liu played a key role in the theoretical framework and methodology providing significant input in the model construction and optimization process Dianli Wang was responsible for performing the experiments and evaluating the model’s performance on various datasets Qingru Sui contributed to the literature review and provided crucial insights into the application of game theory and Nash equilibrium in classification algorithms Chao Yang was involved in data preprocessing and preparing the NSL-KDD and UNSW-NB15 datasets for experimentation Hengxu Li assisted in the implementation of the CE-GAN model and supported the coding and debugging phases Yifeng Li contributed to the statistical analysis and provided technical support for the experiment setup Tianyun Luan helped with manuscript editing and formatting ensuring the overall coherence and quality of the final draft Download citation DOI: https://doi.org/10.1038/s41598-025-90815-5 Growth Next-Generation Agriculture (GAN) is a climate resilience debt fund accelerating Brazil’s transition to regenerative agriculture It works through financing the purchase of biological inputs by farmers from local small and medium-sized enterprises (SMEs) Brazil’s agriculture relies heavily on chemical inputs which can degrade soil health and contribute to greenhouse gas (GHG) emissions Transitioning to biological inputs is crucial for restoring environmental sustainability and meeting Brazil’s climate goals large market participants selling chemical inputs are inhibiting this transition These companies offer comprehensive services which makes it difficult for smaller biological input providers to compete This system limits farmers’ access to alternative biological inputs and slows the adoption of regenerative agriculture practices Growth Next-Generation Agriculture (GAN) revolutionizes agricultural finance by redefining risk assessment through AI-driven credit analysis breaking barriers of traditional methodologies At the heart of this innovation lies a robust tech stack tailored specifically for agriculture seamlessly supporting the credit analysis process and dynamically assessing eligibility criteria This dynamic approach is further fortified by a dedicated credit and impact committee ensuring allocation of funds while propelling the transition from conventional guarantees towards portfolio diversification By facilitating greater access to capital and expediting sales GAN catalyzes the acceleration of sustainable progress within tropical agriculture “We believe effective finance strategies for climate change require collaborative models Applying for the Lab offers us the chance to connect the agricultural value chain with our technology aligning with world-leading institutions in climate policy development and collaboratively build a sustainable financial product for new tropical agriculture stakeholders.” GAN will invest BRL 250 million (USD 44.4 million) in approximately ten securitizations issued by established Brazilian biological input SMEs for its demonstration fund The demonstration fund will prove GAN’s commercial viability paving the way for larger-scale investments and potential expansion to other Latin American markets With Brazil’s total market for chemical fertilizers reaching BRL 200 billion (USD 35 billion) GAN has a significant opportunity to catalyze a shift towards regenerative agriculture GAN is a debt fund that finances farmers’ purchase of biological inputs from SMEs through asset-backed securities (ABSs) The fund structure leverages a common Brazilian receivables instrument to invest in securitized receivables issued by bio-input SMEs offer credit to farmers for purchasing their products driving market growth and accelerating the transition to regenerative agriculture GAN employs a blended finance structure with a senior tranche financed by commercial and impact investors and a junior tranche supported by concessional capital providers This structure mitigates investment risk while maximizing market-rate returns for the senior tranche Traive will originate securitizations in partnership with Folio which provides technical assistance to accelerate the adoption of regenerative practices By addressing the critical barrier of access to affordable working capital for biological input SMEs GAN aims to catalyze Brazil’s transition to regenerative agriculture combined with the deep expertise of Traive and Folio position it as a powerful tool to drive climate resilience in Brazil Metrics details Coronary artery disease (CAD) is one of the most common cardiovascular disorders affecting millions of individuals globally It is the leading cause of mortality in both the wealthy and impoverished nations CAD patients exhibit a wide range of symptoms some of which are not evident until a major incident occurs The development of techniques for early detection and precise diagnosis is heavily dependent on research The proposed system introduces a novel approach Generative Adversarial Networks Augmented Naïve Bayes (GAN-ANB) to classify high-risk CAD patients using Coronary Computed Tomography Angiography (CCTA) imaging data The database included images from Coronary Computed Tomography Angiography (CCTA) records of 5,000 individuals The developed GAN framework consists of a generator to generate synthetic patient profiles and a discriminator to distinguish between genuine and synthetic profiles to improve the identification of high-risk CAD patients Adding synthetic data to the training process allowed the discriminator to be utilized further to improve predictive modeling The performance of the GAN-enhanced prediction model was assessed using accuracy and area under the Receiver Operating Characteristic curve (ROC) The model exhibited an outstanding Dice Similarity Coefficient (0.91) and precision (0.98) in differentiating between high-risk and low-risk individuals The identification of high-risk patients with CAD is greatly enhanced by the integration of GANs with clinical and imaging data ROC of 0.99 was achieved by the GAN-ANB model which outperformed conventional machine learning models and some CCTA-derived imaging characteristics including plaque load and luminal stenosis This method offers a powerful tool for early diagnosis and intervention potentially leading to improved patient outcomes and lower healthcare expenditure The identification and management of these risk factors are critical A world with 50% fewer age-standardized cardiovascular deaths and a much lower age-standardized incidence of cardiovascular disease by 2050 a society in which everyone has an equal opportunity to achieve good cardiovascular health and premature cardiovascular death can be prevented globally we envision a world in which everyone has a personal right to and responsibility for maintaining good cardiovascular health As coronary angiography makes coronary artery blockages directly visible and measurable it continues to be the gold standard for detecting coronary artery disease (CAD) (Patel et al. The following are the novel findings of the proposed research which uses Coronary Computed Tomography Angiography (CCTA) imaging data to identify high-risk patients for CAD using Generative Adversarial Networks (GANs) and Naïve Bayes The proposed study addresses the problem of sparsely annotated imaging data using GANs to produce realistic synthetic CCTA images the generalization and resilience of the classification model are strengthened Adversarial training with GANs increases the CCTA image quality and resolution which enhances the visibility of anatomical features for CAD diagnosis GANs can help strengthen the resilience of the classification model by generating various samples to reduce overfitting and enhance generalization to fresh and unseen CCTA images the proposed model performs better in correctly identifying individuals at a high risk for CAD 4 presents the acquired results and discussion the identification and diagnosis of diseases have greatly benefited from the use of Machine Learning (ML) techniques CNNs have the potential to inherit biases from the training set which may result in skewed predictions that are not fair to all patient groups One major problem is maintaining balance in CNN-based predictions16 This study did not compare the performance of the machine learning models with conventional clinical risk assessment tools which could have revealed important information about the added value of using ML in this situation even though the gradient-boosted tree algorithm demonstrated promising results in predicting CAD outcomes lipid indicators may not fully represent the complexities of cardiovascular risk or in those with metabolic syndrome or diabetes These conditions can alter lipid metabolism and the risk profile for cardiovascular events potentially reducing the reliability of lipid markers as standalone predictors which is particularly problematic when working with high-dimensional datasets or when the model is excessively complex with too many trees This leads to the model capturing noise instead of underlying patterns which reduces its generalizability to new unseen data SVMs may have a bias in favor of the majority class which makes it difficult to identify occurrences of the minority class such as infrequent but important positive examples RNNs experience vanishing and growing gradient challenges during training particularly with extensive medical histories limiting their ability to learn the long-term dependencies required for good CVD prediction Training RNNs is computationally expensive and time-consuming because of their sequential structure This restriction makes it difficult to apply RNNs to large-scale healthcare datasets that are routinely utilized in CVD research The requirement for vast amounts of labeled data for training RNNs is a considerable barrier to disease prediction KNN performance is largely dependent on and distance metric A small value can make the model overly sensitive to noise and outliers inpatient data while a large value can over smooth the model and miss important variations in patient conditions that are clinically relevant The requirement for substantial parameter adjustment increases complexity and necessitates significant subject expertise and experimentation Estimating parameters entails computing conditional probabilities from data which may be resource-intensive and time consuming Learning the network structure requires the investigation of a potentially large space for possible interactions among variables LSTMs are intricate and computationally taxing consume large amounts of memory and computing power and take longer to train than straightforward models it requires a large amount of sequential data that are properly labeled which can be problematic in clinical situations The amount and quality of the input data they receive have a significant impact on their performance and training them may be computationally costly considerable processing power and a wide variety of training data are required It is not always easy to assess the quality of the generated data; in many cases manual examinations or specific metrics are required which is a simple technique but has limitations owing to its performance on complicated high-dimensional datasets and its reliance on distance measurements which can be less successful in capturing nuanced patterns Starting at the nadirs of all three aortic valve cusps the PAA corresponded to the plane closest to the origin of the brachiocephalic artery The DA began distal to the origin of the left subclavian artery and extended to the lowest axial disc The vena cavae are venous veins that emerge from the right middle mediastinum directly to the right of the trachea and PAA The PA included the left and right main pulmonary arteries The CS exits the great cardiac vein at the left circumflex coronary artery Drawing the endocardial and ventricular walls allows one to calculate the RVW of the cardiac tissue As shown in Table 2 the baseline characteristics of the study population (n = 5,000) indicated a mean age of 58.4 years half of the patients had a history of hypertension and 22% had been diagnosed with diabetes The laboratory results show an average total cholesterol level of 200.5 mg/dL with LDL and HDL cholesterol levels averaging 130.2 mg/dL and 48.6 mg/dL Imaging results showed that 36% of the patients had significant coronary artery stenosis Table 3 outlines the baseline characteristics of the male and female patients with CAD Males showed higher rates of smoking history and myocardial infarction than females which may contribute to higher overall risk levels females have a higher prevalence of post-menopausal status while high LDL cholesterol levels are more prevalent in males The p-values indicate the statistical significance of these differences with smoking history and history of myocardial infarction showing significant sex disparities N represents the number of patients in each group and the p-value indicates statistical significance Block diagram of the proposed model to identify CAD The discriminator distinguishes between actual and synthetic pictures to enhance the output of the generator which produces synthetic CCTA images or enriches the original images to emphasize the CAD-relevant aspects the intermediate layers of the discriminator were used to extract rich and informative features from the real CCTA images These features serve as input to the Naïve Bayes classifier which models the probability distribution of the features for CAD and non-CAD cases to identify high-risk patients The process of converting cardiovascular images from the spatial domain to the frequency domain for GAN-based segmentation preprocessing involves the use of Fourier Transform visual data may be altered in terms of their frequency components which makes it possible to reduce noise or accentuate pertinent information Certain features of cardiovascular architecture can be efficiently accentuated or suppressed using filters the images were transformed back into the spatial domain using inverse Fourier Transform The treated images exhibited decreased noise artifacts and improved structural features These preprocessed images were used to train models for the precise segmentation of cardiovascular anatomy using Generative Adversarial Networks (GANs) as optimal inputs During acquisition, all types of noise, including electrical and photon noise, might impact CCTA images. The pre-processing steps are illustrated in Fig. 2. As shown in Fig. 2 the input CCTA images were first captured and normalized to standardize intensity values This is where the block diagram for preparing CCTA images using the Fourier Transform starts a 2D Fourier Transform was applied to these images shifting their spatial domains to their frequency domains Different frequency-domain filtering techniques are used in the frequency domain to amplify or suppress particular frequency components A 2D Inverse Fourier Transform was then used to return the filtered frequency-domain images to the spatial domain the final images undergo post-processing procedures including noise reduction and image enhancement The accuracy of diagnostic evaluations can be affected by noise artifacts which can hide crucial anatomical information and be reduced with the use of preprocessing procedures The Fourier Transform preprocessing of the cardiovascular images for GAN segmentation is described as follows: To transform an image from the spatial domain to the frequency domain By using the sine and cosine components to dissect an image the Fourier Transform can be used to describe the CCTA image in terms of frequencies where \(\:f(x,y)\) is the pixel value at position \(\:(x,y)\); M and N are the dimensions of the image u and v are the frequency-domain coordinates The FFT shift moved the zero-frequency component to the center of the frequency-domain image for better visualization and processing The filters in the frequency domain were applied to enhance the features and reduce noise The shifted Fourier Transform is multiplied using the chosen filter to enhance or suppress specific frequencies The inverse 2D Fourier Transform is applied to convert the processed image back to the spatial domain Diagnostic accuracy and image quality can be compromised by CCTA image artifacts; such as motion artifacts resulting from patient movement or beam-hardening abnormalities caused by variations in tissue density To ensure that the pixel values fall within a standard range (a) to (c): Input CCTA images; (b)–(f) Preprocessed CCTA images by Fourier Transform. Preprocessed CCTA- CS images (CS) through the step-by-step process of Fourier Transform GAN for segmentation of CCTA imaging data As shown in Fig. 5 the generator network is accomplished through the following process: The input of generator G is random noise (z) Generator G consists of multiple layers of neural network units These layers are used to transform the input noise z into a higher-dimensional representation The generator G’s\(\:{\theta\:}_{G}\) parameters of the generator G are tuned during the training phase to obtain a mapping between the output image space and the input noise space The gradients from discriminator D are utilized to update \(\:{\theta\:}_{G}\) throughout this backpropagation learning phase within the GAN framework The final output of the generator network is a synthetic CCTA image \(\:\widehat{X}\) This image is generated in such a way that it should resemble real CCTA scans of high-risk patients in terms of key features and patterns Discriminator D’s task is to reduce the degree to which the produced images \(\:\widehat{X}\:\) can be distinguished from the actual CCTA images X by the generator To deceive D into thinking that the pictures are genuine G wants to create \(\:\widehat{X}\:\) that is sufficiently realistic The generator G is updated during training using the gradients obtained from the feedback of discriminator D this adversarial process enhances G’s capacity of G to produce increasingly accurate and lifelike synthetic CCTA images The following elements are part of the GAN framework The generator network G learns to generate synthetic CCTA images \(\:\widehat{X}\) from random noise z where \(\:{\theta\:}_{G}\) represents the parameters of the generator network The following is a representation of the overall framework: where \(\:{f}_{L}\) denotes the function represented by the lth layer in the network and L is the number of layers Discriminator network D aims to distinguish between real CCTA images X from high-risk patients and synthetic images \(\:\widehat{X}\) where \(\:{\theta\:}_{D}\) represents the parameters of the discriminator network and \(\:{g}_{K}\) denotes the function represented by the kth layer in the network The training process involves minimizing the following objective function which balances the generator’s goal of fooling the discriminator and the discriminator’s goal of correctly distinguishing real images from synthetic images where \(\:{P}_{data\left(X\right)}\) is the distribution of real CCTA images from high-risk patients and \(\:{p}_{z\left(Z\right)}\) is the prior distribution of the input noise generator G can produce synthetic CCTA images \(\:\widehat{X\:}\)that resemble CCTA images of high-risk patients Clinicians can detect the characteristics or trends that point to CAD risk factors Different loss functions are used for the training of the discriminator and the generator Training involves updating the parameters G and D using gradient descent The update rule for the discriminator is as follows CAD risk assessment based on CCTA scans may be more accurate using GANs for this work because of their capacity to recognize intricate patterns and variability in medical imaging data This approach aims to help physicians with early identification and individualized treatment planning for high-risk CAD patients This is an inventive use of deep learning in the healthcare industry The GAN training algorithm involves iterating over the following steps: Step 1: Sample a mini-batch of real CCTA images \(\:{\left\{{X}^{\left(i\right)}\right\}}_{i=1}^{m}\) from the real data distribution \(\:{p}_{data}\left(X\right).\) Step 2: Sample a mini-batch of noise vectors \(\:{\left\{{z}^{\left(i\right)}\right\}}_{i=1}^{m}\) ​ from prior distribution \(\:{p}_{z}\left(z\right).\) Step 3: Generate synthetic images \(\:{\left\{{\widehat{X}}^{\left(i\right)}\right\}}_{i=1}^{m}\) using the generator G, as shown in Eq. (1) Step 3: Compute the discriminator loss \(\:{L}_{D}\)​ using the real images and generated images, as shown in Eq. (5) Step 4: Update the discriminator parameter \(\:{\theta\:}_{D}\) using gradient descent, as shown in Eq. (6) Step 5: Compute the generator loss \(\:{L}_{G}\)​ using the generated images, as shown in Eq. (7) Step 6: Update the generator parameters \(\:{\theta\:}_{G}\) using gradient descent, as shown in Eq. (8) generator G gains the ability to produce artificial CCTA images by converting random noise z into images X using a neural network trained with the adversarial feedback of discriminator D optimization entails reducing the generator’s loss and optimizing the discriminator’s capacity to discern between genuine and artificial images the generator can understand the intricate patterns and characteristics found in actual CCTA images from high-risk CAD patients The discriminator D consists of multiple layers and \(\:{\upvarphi\:}\left(x\right)\:\)denotes the output of the intermediate layers Let \(\:{X}_{real}\:\)be the actual CCTA-segmented image from the generator These images pass through the discriminator to obtain intermediate features \(\:D\left(x\right)\:\)is composed of L layers Where \(\:{D}_{i}\)​ denotes the ith layer of the discriminator The feature representation \(\:{\upvarphi\:}\left(x\right)\:can\:be\) obtained from any layer i: To obtain the feature representations the following equation is used where \(\:{\text{F}}_{real}\) is the matrix of feature vectors for the real images The procedure for extracting features from CCTA images using a GAN discriminator is described in the following pseudocode: # Train discriminator D with real and fake images fake_images = G(generate_random_noise(batch_size)) # Train generator G to fool discriminator D # Use the trained discriminator D to extract features for each preprocessed_image in the dataset: features = D.extract_features(preprocessed_image) High-quality features from the GAN discriminator help the GAN Augmented Naïve Bayes (GAN-ANB) Bayes to perform better as these features are likely to be less correlated \(\:{\text{F}}_{real}\) is used by the Gaussian Naive Bayes classifier to classify these features Naïve Bayes provides a simple yet efficient method that helps with illness detection and clinical decision-making by assuming the conditional independence of features given the class label To determine a class’s posterior probability \(\:{C}_{k}\) given the features \(\:{x}_{1},{x}_{2},\cdot\cdot\cdot,\:{x}_{n}\:\)are as follows: \(\:P\left(\left.{C}_{k}\right|{x}_{1},{x}_{2},\dots\:,{x}_{n}\right)\)is the posterior probability of class ​ \(\:{C}_{k}\:\)given the features \(\:{x}_{1},{x}_{2},\:\dots\:,\:{x}_{n}.\) \(\:P\left({C}_{k}\right)\)is the prior probability of class \(\:{C}_{k}\)​ \(\:P\left({{x}_{i}\left|C\right.}_{k}\right)\:\)is the conditional probability of feature \(\:{x}_{i}\) given class \(\:{C}_{k}.\) \(\:{P(x}_{1},{x}_{2},\dots\:,{x}_{n})\) is the evidence probability GAN-ANB is suitable for discrete features and is typically used with counts or frequencies where \(\:count({x}_{i},\:{C}_{k})\:\)is the count of features \(\:{x}_{i}\)​ in class \(\:{C}_{k}\)​ and \(\:\alpha\:\:\)is a smoothing parameter used to handle unseen features Calculate the posterior probability of each class \(\:{C}_{k\:}\)based on the training set \(\:\:\:\:\:\:P\left(\left.{x}_{i}\right|{C}_{k}\right)\:es\)timate the conditional probabilities for each feature \(\:{x}_{i}\) ​ given each class \(\:{C}_{k}\)​ based on the training data For a new CCTA image with features \(\:{x}_{1},{x}_{2},\dots\:,{x}_{n}\) calculate the posterior probability is calculated as follows: Normalize using the evidence probability for all classes to obtain \(\:P(\left.{C}_{k}\right|{x}_{1},{x}_{2},\cdot\cdot\cdot,\:{x}_{n}).\) Assign the class label \(\:\widehat{y}\:\)to \(\:arg\underset{k}{\text{max}}P(\left.{C}_{k}\right|{x}_{1},{x}_{2},\cdot\cdot\cdot,\:{x}_{n})\:\) the class with the largest posterior probability or plaques in the coronary arteries is a part of the treatment of CAD or aberrant heart wall movement are considered when classifying heart failure The pseudocode outlines the process of classifying CCTA images using GAN-ANB incorporating feature extraction from the GAN discriminator features_array = convert feature_list to an array # Divide data into sets for testing and training test_labels = split dataset into train and test sets naive_bayes = initialize GAN-ANB classifier predicted_labels = GAN_ANB.predict(test_features) To generate personalized risk predictions for CAD using the GAN-ANB model including demographic information and CAD indicators were extracted from the clinical and imaging data The GAN component is then used to augment the dataset by generating synthetic data that simulate various CAD manifestations which helps balance and enhance the dataset The classifier calculates the probabilities of high- and low-risk for CAD based on the patient’s features Risk score S was computed using the following formula: where \(\:(Y=1|X)\) denotes the probability of being high risk given features X and \(\:P(Y=0|X)\) is the probability of being low risk it is represented as \(\:S=f(X;\theta\:)\) where f is the function of the GAN-ANB model This process allows the GAN-ANB model to generate nuanced sex-specific risk scores and recommendations thereby improving personalized risk assessment and management for CAD by incorporating both real and synthetic data to capture a comprehensive risk profile for each patient The proposed study employed a thorough strategy to identify high-risk categories defined as a coronary artery narrowing of ≥ 70% Also taken into account of patients whose CAD affects two or more coronary arteries high-risk individuals are identified by the important features of ACS-prone plaques wherein the plaques induce the artery to grow outward patchy calcification that increases plaque instability High-risk individuals also have noticeable clinical symptoms or a history of significant CAD events such as myocardial infarction or recurrent angina This comprehensive description enabled the proposed method to concentrate on patients with the most severe and unstable CAD characteristics The proposed model leverages a combination of Generative Adversarial Networks (GAN) and GAN-ANB the dataset underwent extensive pre-processing The training and testing sets of the dataset were separated to enable a thorough assessment of model performance The output of the GAN is used to augment the training data which enhances the variability and richness of the dataset This augmented dataset is then fed into a Naïve Bayes classifier which is selected for probabilistic classification jobs owing to its efficiency and simplicity CVD classification results of the proposed model The DSC is a statistical metric that compares the similarity between two sets of data It is very useful in image segmentation to determine the overlap between the expected segmentation and ground truth whereas a DSC of zero indicates no overlap mIoU is another method for assessing image segmentation performance It computes the average IoU for all classes in multiclass segmentation Higher mIoU values indicate an improved segmentation performance An IoU of one indicates complete segmentation also known as the sensitivity or true positive rate is the proportion of genuine positives that the model accurately detects with 1 indicating that all affirmative cases were properly recognized which is often referred to as the Positive Predictive Value is the proportion of correctly predicted positives Comparison of obtained segmentation results with competitive methods Comparison of CVD classification results with competitive methods The proposed method outperformed all the other methods delivering greater performance across measures it may increase the implementation complexity and require careful calibration of both GAN and Naïve Bayes components GAN-ANB had the highest DSC value of 0.91, indicating a good balance in segmenting CCTA images. CNN (0.76), U-Net (0.75), and KNN (0.74) showed strong but slightly lower performance. ResUNet scored the lowest at 0.72, as shown in Fig. 9. GAN-ANB leads with an mIoU of 0.90, indicating its effectiveness in identifying the actual positives. KNN, CNN, and U-Net followed with mIoU values of 0.81, 0.75, and 0.74, respectively. ResUNet had the lowest recall of 0.71, as shown in Fig. 10. GAN-ANB showed the best recall with a value of 0.96. This means that they had a high proportion of true positives among the predicted positives. CNN, U-Net, and KNN had similar recall values of 0.78, 0.77, and 0.76, respectively. ResUNet again scored the lowest, with a recall of 0.74, as shown in Fig. 11. GAN-ANB achieved the highest precision of 0.98, indicating a high proportion of correctly predicted pixels. CNN (0.86), U-Net (0.85), and KNN (0.84) followed closely, showing their robustness. ResUNet has the lowest precision at 0.83, indicating that it is slightly less reliable in terms of pixel-wise accuracy, as shown in Fig. 12. ROC of the proposed technique for classification of CVD A potential method for detecting Coronary Computed Tomography Angiography (CCTA) imaging data-derived high-risk patients for Coronary Artery Disease (CAD) is to integrate Generative Adversarial Networks (GANs) with a generative adversarial network-augmented naïve Bayes (GAN-ANB) classifier Through the creation of artificial CCTA images GANs enrich the dataset and may improve the classification accuracy by capturing subtle patterns and characteristics that are essential for CAD risk assessment In addition to improving the model’s generalization and robustness the interaction between GANs and GAN-ANB classification provides the possibility of using cutting-edge image processing methods for CVD diagnostics Through more precise and timely detection of high-risk CAD patients based on thorough imaging data analysis this research path holds great promise for the development of individualized therapy Future research should focus on validating the model across diverse populations and incorporating it into clinical decision support systems for real-time risk assessment and management The datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request Biology of Cardiovascular and Metabolic Diseases (Academic and infection in atherothrombosis: JACC review topic of the week Third universal definition of myocardial infarction and clinical risk factors of recurrent coronary artery disease events: a population-based cohort study Risk Factors of Coronary Artery Disease: A Hospital-Based Study Association of Non-modifiable Risk factors with coronary artery disease (CAD) in a Tertiary Care Hospital in Peshawar History of global burden of disease assessment at the World Health Organization estimating the future burden of cardiovascular disease and the value of lipid and blood pressure control therapies in China and future of global health financing: a review of development assistance and other private spending on health for 195 countries Defining and setting national goals for cardiovascular health promotion and disease reduction: the American Heart Association’s strategic impact goal through 2020 and beyond Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): case-control study Dissecting racial bias in an algorithm used to manage the health of populations Use of machine learning to identify risk factors for coronary artery disease Predictive Modeling of Cardiovascular Disease using Machine Learning Techniques In 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC) Age-biomarkers-clinical risk factors for prediction of cardiovascular events in patients with coronary artery disease The Emerging Risk Factors Collaboration Lipid-related markers and Cardiovascular Disease Prediction Improving Heart Disease Prediction using Random Forest and AdaBoost algorithms Machine Learning with R: Expert Techniques for Predictive Modeling (Packt publishing ltd The Elements of Statistical Learning: Data Mining SMOTE: synthetic minority over-sampling technique Yap drives the development of cardiovascular disease in patients with rheumatoid arthritis Baskaran, L. Data from: Automatic segmentation of multiple cardiovascular structures from cardiac computed tomography angiography images using deep learning, Dryad, Dataset, (2024). https://doi.org/10.5061/dryad.9s4mw6mc9 A few useful things to know about machine learning External validation of prognostic models for critically ill patients required substantial sample sizes Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning Arrhythmia Detection Using Deep Convolutional Neural Network with long Deep patient: an unsupervised representation to predict the future of patients from the electronic health records Deep belief network for lung nodules diagnosed in CT imaging Color classification of extrasolar giant planets: prospects and cautions Identification of cardiovascular disease risk factors among diabetes patients using ontological data mining techniques & U-Net Convolutional Networks for Biomedical Image Segmentation International Conference on Medical Image Computing and Computer-Assisted Intervention Alom, M. Z., Hasan, M., Yakopcic, C., Taha, T. M. & Asari, V. K. Recurrent residual convolutional neural network based on u-net (r2u-net) for medical image segmentation, Arxiv, arXiv preprint arXiv:1802.06955 (2018) Evaluating traditional machine learning methods for Cardiovascular Risk Prediction Cardiac Disease classification with deep convolutional neural networks Enhancing Cardiovascular Disease classification with GAN-Augmented Naïve Bayes Download references This research is supported by the Yangtze River Delta Science and Technology Innovation Community Joint Research Project (2022CSJGG1000/2023ZY1068) Open access funding provided by Manipal Academy of Higher Education School of Information and Electronic Engineering Zhejiang University of Science and Technology Key Laboratory of Biomedical Intelligent Computing Technology of Zhejiang Province Department of Mechanical and Industrial Engineering and Anandakumar Haldorai (A.H.); Methodology: Lei Zhang (L.Z.) and Nithesh Naik (N.N.); Writing — Original Draft Preparation: Lei Zhang (L.Z.) and Anandakumar Haldorai (A.H.); Writing — Review and Editing: Nithesh Naik (N.N.); Supervision: Anandakumar Haldorai (A.H.) All authors have read and agreed to the published version of the manuscript Download citation DOI: https://doi.org/10.1038/s41598-024-73176-3 (Nasdaq: TGAN)—a pioneer in and a global supplier of high reliability high performance gallium nitride (GaN) power conversion products—announced today the latest reliability ratings for its GaN power FETs Reliability is measured by Failures in Time (FIT) an analysis that considers the number of devices reported by customers to have failed in the field when used in applications the company's total product portfolio has achieved an average < 0.1 FIT rate based on more than 85 billion hours of field operation This rating stands as one of the industry's best and only reported broad power spectrum reliability rating of any GaN power solution available today Transphorm built its GaN platform with reliability in mind understanding its importance when the wide bandgap technology first hit the market: even though GaN boasted higher performance than silicon-based transistors customers wouldn't opt to switch to the then-new technology if the devices failed in real-world use Transphorm was the first GaN manufacturer to publish a complete validation data set backing its reliability claims the company regularly shares its GaN reliability achievements to help potential customers make informed decisions when choosing semiconductor suppliers Transphorm last reported its FIT rate to be < 0.3 in Q1 2022 Transphorm takes another step toward changing how customers assess GaN FET options The company has taken its reliability data and segmented it into two categories: When looking at device performance by power level type Transphorm's GaN FETs yield the following reliability ratings that are notably similar to those of Silicon-based power devices: "Our high voltage GaN devices are designed into the broadest range of applications covering the widest power spectrum from 45 W to 4 kW today with the potential to reach 10+ kW as GaN is adopted into new markets This shows the immense versatility of our technology," said Philip Zuk Senior Vice President of Business Development and Marketing we realized that reporting just a singular reliability rating that lumps all application types together may not be as useful to customers We felt it necessary to help them access more nuanced data that would apply to their specific design requirements the breakdown between low and high power." Transphorm's device quality + reliability continues to position the company as a high voltage GaN leader The company ships into a variety of end markets To learn more about the SuperGaN technology difference designs and manufactures high performance and high reliability GaN semiconductors for high voltage power conversion applications Having one of the largest Power GaN IP portfolios of more than 1,000 owned or licensed patents Transphorm produces the industry's first JEDEC and AEC-Q101 qualified high voltage GaN semiconductor devices The Company's vertically integrated device business model allows for innovation at every development stage: design Transphorm's innovations are moving power electronics beyond the limitations of silicon to achieve over 99% efficiency 40% more power density and 20% lower system cost California and has manufacturing operations in Goleta and Aizu Follow us on Twitter @transphormusa and WeChat @ Transphorm_GaN is based on the information as of the date indicated on the document but may be subject to change without prior notice and solutions delivered straight to your inbox Developing peptide-based tools to study streptococcal quorum sensing Yftah Tal-Gan, Ph.D. Chemistry has established a chemical biology research program with the overarching goal of developing and utilizing peptide-based probes to study bacterial communication pathways and their role in bacterial pathogenesis and inter-species competition This research program is multidisciplinary and spans from organic synthesis and analytical characterization structural determination of biomacromolecules Current projects in the lab are funded by one National Institutes of Health (NIH) and one National Science Foundation (NSF) grant and have resulted in more than 40 papers from the lab The Tal-Gan lab group explores Streptococcal quorum sensing through a multifaceted lens Quorum sensing (QS) is a cell-cell signaling mechanism that enables bacteria to assess their cell density in a given environment; and at a high population density synchronize the transcription of genes associated with group-behavior phenotypes Streptococcal QS circuits are centered around the production and detection of peptide signals (termed autoinducers) The peptide signal concentration is directly proportionate to the population density As the bacteria reach a critical concentration indicative of a high population density quorum sensing peptides activate a membrane-bound receptor leading to the transcription of group behavior genes Because many pathogens utilize QS to initiate and synchronize their attack on the host QS interference can be used to reduce or eliminate bacterial pathogenicity without leading to resistance Our interdisciplinary work resides at the intersection of organic chemistry Our research efforts are focused on the development and analysis of peptide-based probes to study QS in streptococci we have developed QS peptide analogs of native streptococcal QS peptides that effectively silence QS signaling These molecules are potential candidates for therapeutic alternatives to antibiotics using structural determination and molecular biology approaches we analyze and manipulate QS peptide molecules and circuitry to understand binding dynamics that mediate downstream bacterial virulence and survival processes we investigate how these systems influence the behaviors of host immune cells the PREP undergraduate researcher will synthesize and purify peptide-based signal analogs of the streptococcal ComABCDE QS circuitry termed competence stimulating peptide (CSP) The PREP undergraduate researcher will then assist in evaluating the biological activity of the different CSP analogs they produced to delineate structure-activity relationships and define the molecular interactions that govern signal-receptor binding This information would then be used to design novel CSP-based QS modulators with desired activity profiles.  Pack Research Experience Program information and application Metrics details Braille serves as an efficient means for visually impaired individuals to access textual information and engage in communication the process of reading Braille can often be cumbersome and time-intensive particularly in bidirectional human-machine interaction a compact optical device for contactless detection of Braille is fabricated and characterized serves as the core for both light emission and photodetection significantly reducing its overall footprint The incorporation of the semi-ellipsoid epoxy lens with optimized dimensions ensures consistent and accurate detection The sensing device demonstrates high stability and fast response through its line-scanning capabilities on Braille codes The captured signals are analyzed using a microcontroller and the Braille recognition results are wirelessly transmitted to a portable mobile device enabling the conversion into audio and visual formats This innovative design not only facilitates Braille reading but also holds the potential to advance human-machine interaction The widely used optical method is the FBG optical fiber waveguide pursuing highly integrated devices offering stable detection and contactless capabilities is paramount for widespread application a notable strategy involves minimizing the device footprint through the monolithic integration of optical components detectors and waveguides on a GaN-based material platform a compact optical device is introduced for contactless detection of Braille patterns which serves as the core component for light emission and photodetection and is integrated with an epoxy lens In addition to the characteristics of the GaN-based chip the structural optimization of the epoxy lens is performed to ensure consistent and reliable Braille detection The electronic modules are further implemented for signal analysis Schematic diagrams showing the configuration of (a) the sensing device and (b) the detection system for converting Braille reading signal into slide-to-audio and mobile display a Bird’s eye view image of the fabricated GaN chip The inset shows the chip packaged on the PCB b Optical images of the GaN chip integrated with epoxy lenses of different heights c Lateral-view image of the experimental setup The inset shows the top-view image of the Braille pad for testing d Schematic diagram showing the working principle of the sensing device Figure 2d illustrates the working principle of the sensing device The monolithic integration design of the LED and PD on a GaN-on-sapphire chip allows the device to operate in reflection mode the multi-quantum wells (MQW) act as active regions to generate light through radiative recombination The emitted light is partially extracted through transparent sapphire and the epoxy lens The amount of reflected light is governed directly by the Braille patterns the reflected light is partially captured by the MQW of the PD and the resulting photocurrent can be used to indicate the Braille patterns effectively translating tactile information into electrical signals a Emission spectrum of the LED and absorption spectrum of the PD The inset shows the relationship between the light output power of LED and the driving current c I-V characteristic of the PD measured at varying LED driving currents d Plot of the relationship between PD photocurrent and LED current e Cross-sectional images showing three sliding tests using the bare chip at different distances f Plots of measured relative photocurrent responses (ΔI/I0) of the bare chip at different distances The shaded region represents the time intervals where the chip spatially overlaps the dot After studying the properties of the device, its sensing characteristics are examined. Figure 3e-f depicts the photocurrent responses of the bare chip as it slides across the Braille pad at varying distances (D) of 0.1 mm represents the ratio of the variation in photocurrent upon encountering a Braille dot compared to the baseline photocurrent value of the device ΔI/I0 increases from the average baseline of 4.41% to the average peak of 6.23% as the device moves over a Braille dot as the distance increases to 1 mm and 3 mm an inverse trend is observed in the ΔI/I0 ratio wherein the presence of a Braille dot leads to a decrease from the average baseline levels of 1.30% and 0.42% to respective minima of 0.66% and 0.22% The inconsistent fluctuations and weak sensing responses as distance increases make the bare chip unsuitable for reliable detection of Braille Simulation results of (a) a bare chip and (b) a chip integrated with a 1.8-mm-height lens under D = 0.1 mm and 1 mm The w/ dot and w/o dot refer to cases with dot and without dot c Simulation results of the relative photocurrent changes in chips with the epoxy lens of different heights versus varying distances to the Braille dot To address the inconsistent behavior observed due to the divergence nature of the LED emission, integrating a semi-ellipsoid epoxy lens structure on the chip can be a practical and efficient solution. To ensure the complete coverage of the GaN chip, the bottom radius of the semi-ellipsoid lens is set to 1 mm. From the simulation result shown in Fig. 4b the inclusion of a 1.8 mm-high epoxy lens not only reduces the critical angle at the sapphire boundary due to the diminished refractive index contrast but also shapes the light beam extracted from the LED forms into a more collimated profile at both close and far measurement distances the PD receives more light from the non-dotted surface than the dotted one Positive and negative deviations are indicated by red and blue colors the light flux changes exhibit inconsistent trends as the distance increases When using the lens with H greater than 0.8 mm the device can provide a monotonous response trend the lens of H = 1.8 mm enables the device to have the largest sensitivity and ΔФ/Ф0 achieves −20.14% at D = 0.1 mm Since the luminous flux received in the PD region directly correlates with the photocurrent generated the simulated results provide guidance for experimental device design Relative photocurrent responses of the device to the Braille pad with epoxy lens of (a) H = 0.2 mm a Photocurrent of the devices with and without epoxy lens measured at varying distances b Cyclic measurement of the sensing device over 5000 cycles The inset shows an enlarged view of the repeated cycles c Transient response of the sensing device which indicates the swift and consistent performance of the sensing device a Optical images of the overall slide-to-audio and display conversion system b Schematic diagram showing the arrangement of Braille dots for reading c Plot of the photocurrent responses when setting the device at varying distances above the Braille patterns d Tolerance test to measure the photocurrent responses when the device is offset from the Braille patterns a Schematic diagram showing the flowchart of the detection system b Plot of the relative voltage responses for Braille recognition Table 1 presents a comparative performance analysis of the previously reported Braille sensor Most of these sensors rely on contact mechanisms and their sensing performances are highly related to material characteristics While some reported sensors can achieve outstanding response time typically managing cycles below 1000 times sensors with excellent repeatability are influenced by material limitations and cannot achieve highly rapid response time Only a few studies have managed to balance these properties contactless sensing devices have emerged as viable alternatives potentially exhibiting less performance degeneration the GaN device demonstrated in this work maintains relatively excellent repeatability with 5000 cycles and achieves fast response/recovery times of 24.14/24.15 ms by incorporating a chip-scale integration design that eliminates the need for external optical components the investigated device significantly reduces its footprint to the millimeter scale a chip-scale optical Braille sensing unit is introduced comprising a monolithic GaN device integrated with an epoxy lens the device demonstrates a consistently reliable photocurrent response when sliding over Braille patterns The developed device exhibits a compact design three sensing devices are packaged together facilitating the line scanning of Braille code the incorporation of electronic modules for audio and display conversion enhances user accessibility in portable applications The proposed system enhances human-machine interaction and serves as a valuable aid for visually impaired individuals The fabrication process starts with the growth of GaN epitaxial layers which include the unintentionally doped GaN The LED and the PD regions are defined by exposing the n-GaN through photolithography and inductively coupled plasma (ICP) etching The SiO2 current barrier layer and the indium-tin-oxide (ITO) current spreading layer are deposited on the p-GaN layer The LED and PD areas are isolated by photolithography and ICP etching to remove GaN layers between them till the sapphire substrate is exposed The Cr/Al/Ti/Pt/Au electrodes are e-beam evaporated on the n-GaN and ITO followed by deposition of SiO2 passivation and SiO2/TiO2 distributed Bragg reflector (DBR) ICP etching of the DBR and passivation layer results in the exposure of the underlying electrodes thereby enabling the formation of an electrical interconnection through the e-beam evaporation of the metal pad the fabricated wafer is laser-diced into small chips The PDMS gels are mixed with the pre-polymer and curing agent in a ratio of 10:1 and the mixture is poured into a designed 3D-printed resin mold the PDMS film is detached from the resin mold serving as a secondary mold with an inversed lens-shape geometry The epoxy adhesive with a known refractive index of 1.59 (QUINSON-H808) is injected into the mold and the packaged GaN chip is placed on top The mold is maintained in an inverted position to ensure the adhesives contact with the chip and PCB After curing with UV irradiation and detaching from the PDMS mold The data supporting plots within this paper and other findings of this study are available from the corresponding author upon request Global causes of blindness and distance vision impairment 1990-2020: a systematic review and meta-analysis Skin-inspired highly stretchable and conformable matrix networks for multifunctional sensing Deformable conductors for human-machine interface Reverse-micelle-induced porous pressure-sensitive rubber for wearable human-machine interfaces wireless sensors for full-body pressure and temperature mapping Highly sensitive wearable pressure sensor over a wide sensing range enabled by the skin surface‐like 3D patterned interwoven structure Controllable graphene wrinkle for a high-performance flexible pressure sensor Flexible suspended gate organic thin-film transistors for ultra-sensitive pressure detection Performance and service behavior in 1-D nanostructured energy conversion devices Embedding pinhole vertical gold nanowire electronic skins for Braille recognition A skin-like sensor for intelligent Braille recognition E‐skin tactile sensor matrix pixelated by position‐registered conductive microparticles creating pressure‐sensitive selectors Transparent and flexible fingerprint sensor array with multiplexed detection of tactile pressure and skin temperature Bio-photocapacitive tactile sensors as a touch-to-audio Braille reader and solar capacitor Gradient architecture‐enabled capacitive tactile sensor with high sensitivity and ultrabroad linearity range transparent triboelectric nanogenerator as electronic skin for biomechanical energy harvesting and tactile sensing A phonic Braille recognition system based on a self-powered sensor with self-healing ability Machine learning‐enabled tactile sensor design for dynamic touch decoding Self-powered porous polymer sensors with high sensitivity for machine learning-assisted motion and rehabilitation monitoring Diaphragm-micro-stylus-based fiber Bragg grating tactile sensor Prosthetic finger for fingertip tactile sensing via flexible chromatic optical waveguides Investigations on bond performances of GFRP/stainless steel reinforcements using distributed fiber optical sensors (DFOSs) Flexible capacitive tactile sensor based on micropatterned dielectric layer Magnetized micropillar-enabled wearable sensors for touchless and intelligent information communication Flexible magnetoreceptor with tunable intrinsic logic for on‐skin touchless human-machine interfaces Visual and tactile perception techniques for Braille recognition Lifelong robotic visual-tactile perception learning Advantages of 3D time-of-flight range imaging cameras in machine vision applications Real time motion capture using a single time-of-flight camera 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Structured-light 3D surface imaging: a tutorial Download references The authors acknowledge financial support from the National Natural Science Foundation of China under Grant 12074170 and in part by the Shenzhen Fundamental Research Program under Grant JCYJ20220530113201003 Southern University of Science and Technology performed the experiments and carried out the data analysis All authors contributed to the general discussion and approved the final manuscript Download citation DOI: https://doi.org/10.1038/s41378-025-00904-8 Metrics details frequently suffer damage from environmental and human factors necessitating effective restoration techniques This study introduces an innovative mural restoration approach using a generative adversarial network (GAN) within a UNet architecture The generator integrates Transformer and convolutional neural network (CNN) components effectively capturing and reconstructing complex mural features This work's novelty lies in integrating the Group-wise Multi-scale Self-Attention (GMSA) an Encoder-Decoder Feature Interaction (EDFI) module and a Local Feature Enhancement Block (LFEB) These components allow the model to better capture leading to a significant improvement over traditional restoration methods Tested on a dataset of Tang Dynasty murals the method demonstrated superior performance in PSNR and LPIPS metrics compared to seven other techniques Ablation studies confirmed the effectiveness of the heterogeneous network design and the critical contributions of the GMSA Practical restoration experiments showed the method's ability to handle various types of mural damage providing seamless and visually authentic restorations This novel approach offers a promising solution for the digital preservation and restoration of cultural heritage murals with potential applications in practical restoration projects Scientific restoration methods can restore murals to their original appearance as much as possible allowing future generations to better appreciate and study these precious cultural heritages existing deep learning-based mural restoration algorithms have shortcomings in detail recovery leading to the loss of detailed information or the presence of boundary artifacts in the restored images this paper introduces a novel mural restoration method based on generative adversarial networks consisting of a generator and a discriminator The generator performs the mural image restoration and the discriminator evaluates the quality of the restoration The ultimate goal is to restore the murals to their original state as closely as possible using an encoder-decoder information exchange module instead of the original UNet pixel-wise addition operation better utilizing the features of the encoder and decoder The module employs cross-attention to better integrate global and local information the heterogeneous design of the encoder and decoder better focuses on global information while restoring local details The contributions of this paper are summarized as follows: This paper proposes a novel mural restoration network based on generative adversarial networks (GANs) with a generator using an encoder-decoder structure that combines Transformer and convolutional neural networks By designing multi-scale sliding window attention it excels in handling global information and long-range dependencies while CNNs have advantages in local feature extraction and spatial information processing Combining these two allows the generator to more effectively capture and reconstruct the complex structures and texture features in murals improving restoration effects through multi-level encoder and decoder structures and skip connections The introduction of an encoder-decoder feature interaction module and a local feature enhancement block further enhances the ability to restore details and colors in mural images making the restoration more precise and realistic The discriminator adopts a PatchD discriminator and uses wavelet transforms to extract richer high-frequency components from the image Since murals often suffer from the loss or damage of high-frequency information (such as textures and fine details) using wavelet transforms to enhance the discriminator's perception ability helps more accurately evaluate the restoration effect and the degree of detail reconstruction Traditional image restoration methods focus on finding useful small image patches from the surrounding areas to repair the damaged regions these methods do not consider the semantic information contained in the image leading to suboptimal performance in complex scenarios Due to the limitations in feature extraction traditional methods have poor generalization capabilities and can only be used for simple and single-type mural restoration tasks the above methods often treat image restoration as a generative task resulting in images that look natural and realistic but have low similarity to real images The architecture of the proposed Network for image restoration Given a low-quality image \({I}{\prime}\in {\mathbb{R}}^{H\times W\times 3}\) a 3 × 3 convolution is used to extract shallow features \({X}_{0}\in {\mathbb{R}}^{H\times W\times C}\) where \(Con{v}_{3\times 3}(\cdot )\) denotes a 3 × 3 convolution operation and \(C\) denotes the number of channels for shallow features The shallow features \({X}_{0}\) are processed through a three-stage encoder-decoder with the encoder comprising GSwinTB and down-sampling modules For up-sampling and down-sampling operations pixel-shuffle and pixel-unshuffle operations are employed The features from the encoder can be represented as: where \({E}_{i}\) denotes the \(i\) th encoding stage and the feature \({X}_{i}\in {\mathbb{R}}^{\frac{H}{{2}^{i}}\times \frac{W}{{2}^{i}}\times {2}^{i}C},i=\text{1,2},3\)。\({X}_{i}\) is input to the corresponding decoder stage via a hopping connection and interacts with the decoder's features in a feature interaction By establishing feature interactions between the encoder and the decoder it allows the decoder at the next level to better utilize the high-level semantic information extracted by the encoder as well as more fine-grained image details to generate a more accurate and clearer output feature \({X}_{0}\) is used to extract global feature \({Y}_{3}\) using a GSwinTB after 3 encoder stages The decoder stage recovers image information layer by layer each of which includes an upsampling operation and a CNN-based local feature enhancement block In order to avoid the artifactual effect produced by the back-convolution the input features \({Y}_{i}\in {\mathbb{R}}^{\frac{H}{{2}^{i}}\times \frac{W}{{2}^{i}}\times {2}^{i}C},i=\text{3,2},1\) are up-sampled using pixel-shufflfle in the \(i\) th stage denoted as \({Y}_{i}^{up}\in {\mathbb{R}}^{\frac{H}{{2}^{i}}\times \frac{W}{{2}^{i}}\times {2}^{i}C},i=\text{3,2},1\) Output features and up-sampled features of the same level encoder are inputted into the EDFI at the same time where \(H(\cdot ,\cdot )\) denotes the codec feature interaction operation and \({\widetilde{Y}}_{i-1}\) is the output feature of EDFI。 The \(i\) th decoder stage \({D}_{i}\) can be expressed as: \({Y}_{0}\in {\mathbb{R}}^{H\times W\times C}\) extracts the features by 3 × 3 convolution and performs channel dimensionality reduction to obtain the features \({I}^{\prime}\in {\mathbb{R}}^{H\times W\times 3}\) and the final recovered result \(\widetilde{I}={I}^{\prime}+R\) Our network is structured as a Generative Adversarial Network (GAN) with the loss function comprising reconstruction loss and adversarial loss the reconstruction loss is computed using L1-Norm: \(\varpi\) is the discrimination matrix output by the PatchD discriminator and the smaller the elements in the discrimination matrix the better the recovered results GAN loss with generator (Net) loss defined as follows: The GAN loss used to train the discriminator has a symmetric form: where \(\alpha\) and \(\beta\) are hyperparameters that regulate the loss function where \({X}_{i-1}^{1}\) is the output feature of the first layer and the input feature of the second layer, \({\widetilde{X}}_{i}\) is the output feature of the second layer, GMSA denotes Group-wise Multi-scale Self-Attention operation, \(LN\) and \(MLP\) denote layer normalization and multilayer perceptron, respectively。 The architecture of the proposed Sub-module The architecture of the Group-wise Multi-scale Self-Attention and both \({\widetilde{X}}_{i}\) and \({Y}_{i}^{up}\) pass through the 1 × 1 convolution before global average pooling operation to obtain the feature vectors \({v}_{x}\in {\mathbb{R}}^{1\times 1\times {2}^{i}C} i=\text{0,1},2\) and \({v}_{y}\in {\mathbb{R}}^{1\times 1\times {2}^{i}C},i=\text{0,1},2\) where GAP denotes global average pooling and \(Con{v}_{1\times 1}(\cdot )\) denotes 1 × 1 convolution operation The vectors \({v}_{x}\) and \({v}_{y}\) go through a fully connected layer to map the feature vectors into a new space which is activated by a Sigmoid activation function after a nonlinear transformation of the features to obtain \({\widetilde{v}}_{x}\in {\mathbb{R}}^{1\times 1\times {2}^{i}C},i=\text{0,1},2\) and \({\widetilde{v}}_{y}\in {\mathbb{R}}^{1\times 1\times {2}^{i}C},i=\text{0,1},2\) The process of generating the vectors is represented as follows: where FC denotes the fully connected layer and \(\sigma\) denotes the Sigmoid activation function The resulting vectors \({\widetilde{v}}_{x}\) and \({\widetilde{v}}_{y}\) are multiplied element by element with \({Y}_{i}^{up}\) and \({\widetilde{X}}_{i}\) the Concatenation operation is performed and the spliced features are downscaled using a 1 × 1 convolution channel to obtain the output features \({\widetilde{Y}}_{i-1}\) of EDFI The output feature \({\widetilde{Y}}_{i-1}\) is represented as follows: where \(Con{v}_{1\times 1}(\cdot )\) denotes a 1 × 1 convolution operation \([\cdot ,\cdot ]\) denotes a splicing operation and \(\odot\) denotes element-by-element multiplication Local feature extraction module, as shown in Fig. 2(b) the \({\widetilde{Y}}_{i-1}\) features are used as inputs to the Local Feature Enhancement Block (LFEB) which consists of a cascade of two network layers with the same structure where the network layer is designed by designing the channel attention and the spatial attention in parallel to capture spatial correlation and channel correlation in the feature map \({\widetilde{Y}}_{i-1}\) is obtained by layer normalization and two-layer convolution operation to get the feature \({\widehat{Y}}_{i-1}\) \({\widehat{Y}}_{i-1}\) is spliced into the feature map by spatial attention and channel attention after respectively and then subsequently dimensionality reduction is done by using 1 × 1 convolutional channel the feature is compared with \({\widetilde{Y}}_{i-1}\) element by element to get the feature \({\widetilde{Y}}_{i-1}^{1}\) as the input of the second layer \({\widetilde{Y}}_{i-1}^{1}\) through the layer normalization and two − layer convolution operation to get the feature \({\widehat{Y}}_{i-1}^{1}\) \({\widehat{Y}}_{i-1}^{1}\) through the spatial attention and the channel attention spliced features using the 1 small constant 1 convolution channel dimensionality reduction and in with the original feature \({\widehat{Y}}_{i-1}\) are added element by element The final output of the second layer yields \({Y}_{i-1}\) and the augmented feature \({Y}_{i-1}\) obtained after LFEB is used as the input feature for the next decoding stage The specific representation is as follows: where \(Con{v}_{3\times 3}(\cdot )\) denotes a 3 × 3 convolution operation \(Con{v}_{1\times 1}(\cdot )\) is a 1 × 1 convolution operation and CA and SA are channel attention and spatial attention The heterogeneous networks for GAN-based ancient mural restoration is trained with complete mural images Given that we are targeting datasets with unique characteristics murals from the Tang Dynasty were chosen for building the dataset 50,704 images were used as the training set The number of images in the test set was 5634 and the ratio of images in the test set to the training set is 11.11% masks were applied to simulate damaged portions of the murals These methods were chosen for comparison because they have performed well in many real-world applications and therefore permit an accurate assessment of the performance and advantages of our proposed method across a wide range of metrics and aspects The above comparisons clearly show that our method is able to generate structures and textures that are closer to the original image while maintaining the quality of the restoration. Specific comparison results are described in Sect. "Comparison of Experimental Results" our method achieved a 32.49% increase in PSNR on the test set these results demonstrate the superiority of our method in ancient mural restoration Perceptual Similarity: In mural restoration the goal is not solely to replicate the original mural at the pixel level but to restore the visual perceptual quality of the mural LPIPS assists in quantifying the visual perceptual similarity between restoration results and original murals Handling Complexity: Mural restoration often involves intricate image challenges such as cracks While these issues might be difficult to address at the pixel level LPIPS offers an understanding and handling of these challenges Effect Evaluation: LPIPS provides a quantitative standard for assessing and comparing the effectiveness of different restoration methods By comparing the LPIPS scores of murals before and after restoration one can grasp their performance across various tasks Results from an LPIPS evaluation of the quality of all example images in Fig. 4 are displayed in Table 2 Our method achieved an LPIPS 56.61% lower than that of PUT These results highlight the significant advancements made in perceptual restoration by the proposed method the test parameters indicate that our method boasts impressive generalizability they exhibit significant disadvantages in sensory authenticity with noticeable smudging and clear distinctions between restored parts and the surrounding original regions effectively organizes the structures of lines and color blocks in various mural restoration scenarios restoring them in a style more akin to the original murals The local structures around the yellow circles in Fig. 6 demonstrate that EdgeConnect, PUT, and RFRNet fail to recover detailed structures. In contrast, our method successfully handles all the intricate details. Comparison of mural restoration results of EdgeConnect Comparison of mural restoration results of Uformer and GMSA blocks resulted in a performance decline and GMSA blocks contribute significantly to enhancing the model's performance Results of the restoration of the original broken mural The proposed mural restoration method employs a generative adversarial network (GAN) integrated with a UNet architecture incorporating Group-wise Multi-scale Self-Attention (GMSA) Encoder-Decoder Feature Interaction (EDFI) and Local Feature Enhancement Block (LFEB) This approach significantly enhances mural restoration quality and LPIPS metrics compared to state-of-the-art methods The GMSA module improves long-distance attention mechanisms EDFI facilitates better feature integration between the encoder and decoder Experimental results demonstrate the method's robustness and efficacy confirming its value for the digital preservation and restoration of cultural heritage murals No datasets were generated or analysed during the current study Ancient mural inpainting via structure information guided two-branch model Restoration of non-structural damaged murals in Shenzhen Bao’an based on a generator–discriminator network SeparaFill: Two generators connected mural image restoration based on generative adversarial network with skip connect Inpainting of dunhuang murals by sparsely modeling the texture similarity and structure continuity Line drawing guided progressive inpainting of mural damages End-to-end partial convolutions neural networks for dunhuang grottoes wall-painting restoration Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops Thanka mural inpainting based on multi-scale adaptive partial convolution and stroke-like mask Restoration of an ancient temple mural by a local search algorithm of an adaptive sample block Archaeology and restoration of costumes in tang tomb murals based on reverse engineering and human-computer interaction technology Region filling and object removal by exemplar-based image inpainting Real-ESRGAN: training real-world blind super-resolution with pure synthetic data Nontexture inpainting by curvature-driven diffusions Digital restoration of damaged mural images Proceedings of the Eighth Indian Conference on Computer Vision Virtual restoration of old mural paintings using patch matching technique 2012 Third International Conference on Emerging Applications of Information Technology Context encoders: feature learning by inpainting 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Image inpainting for irregular holes using partial convolutions Proceedings of the European conference on computer vision (ECCV) Deep generalized unfolding networks for image restoration Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Generative image inpainting with contextual attention Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Semantic image inpainting with progressive generative networks Proceedings of the 26th ACM international conference on Multimedia recurrent feature reasoning for image inpainting 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Ancient mural restoration based on a modified generative adversarial network EdgeConnect: structure guided image inpainting using edge prediction StructureFlow: image inpainting via structure-aware appearance flow Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Intelligent generation of Peking opera facial masks with deep learning frameworks Restormer: efficient transformer for high-resolution image restoration Retinal image restoration using transformer and cycle-consistent generative adversarial network 2022 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS) Deep transfer learning based classification model for COVID-19 disease Automated deep transfer learning-based approach for detection of COVID-19 infection in chest x-rays COVID-19 detection system using chest CT images and multiple kernels-extreme learning machine based on deep neural network Effect of the modification of CT scanner calibration curves on dose using density correction methods for chest cancer Segmentation of the pulmonary vascular trees in 3D CT images using variational region-growing Secure similar image search and copyright protection over encrypted medical image databases 3D coronary artery reconstruction by 2D motion compensation based on mutual information Automatic determination of optimal view for the visualization of coronary lesions by rotational X-ray angiography Early detection of coronary microvascular dysfunction using machine learning algorithm based on vectorcardiography and cardiodynamicsgram features Multiscale graph cuts based method for coronary artery segmentation in angiograms Spectral CT material decomposition in the presence of poisson noise: a kullback-leibler approach Assessment of qualitative and quantitative features in coronary artery MRA Coronary three-vessel disease with occlusion of the right coronary artery: What are the most important factors that determine the right territory perfusion An improved graph matching algorithm for the spatio-temporal matching of a coronary artery 3D tree sequence AI for COVID-19 detection from radiographs: incisive analysis of state of the art techniques key challenges and future directions Automatic detection of severely and mildly infected COVID-19 patients with supervised machine learning models The Unreasonable effectiveness of deep features as a perceptual metric Reduce information loss in transformers for pluralistic image inpainting JPGNet: joint predictive filtering and generative network for image inpainting Proceedings of the 29th ACM International Conference on Multimedia Uformer: a general U-shaped transformer for image restoration 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Learning enriched features for fast image restoration and enhancement Adam: A method for stochastic optimization Download references School of Information and Communication Engineering Key Laboratory of Acoustic Visual Technology and Intelligent Control System Beijing Key Laboratory of Modern Entertainment Technology All authors contributed to the current work RH proposed the research plan and supervised the whole process to provide constructive comments ZFH completed the method design and model construction and SK and ZX completed the dataset production and organized the experimental data All authors read and approved the final manuscript unless otherwise stated in a credit line to the data Download citation DOI: https://doi.org/10.1186/s40494-024-01517-6 singaporePAP's Gan Siow Huang wins Marymount SMC with 70.70% of votes over PSP's Jeffrey KhooPAP's Gan Siow Huang (left) received 14,868 votes while PSP's Jeffrey Khoo received 6,163 votes.PHOTO: AsiaOne filePUBLISHED ONMay 03, 2025 3:18 PMByKristy ChuaResults for Marymount SMC GE2025 The People's Action Party (PAP) has won Marymount SMC against the Progress Singapore Party (PSP) PAP's Gan Siow Huang received 14,868 votes while PSP's Jeffrey Khoo received 6,163 votes PAP won Marymount SMC with a vote share of 55.04 per cent against PSP For our GE2025 microsite, visit here editor@asiaone.com Metrics details Airflow sensing plays a pivotal role in numerous fields detecting bidirectional airflow using a single sensing unit poses significant challenges a miniature airflow sensing device is introduced utilizing a GaN optical chip integrated with a biomimetic hair structure The sensing device comprises a monolithic GaN chip that handles both light emission and detection constructed from nylon fibers and PDMS film undergo structural bending in converting airflow signals into optical changes modulating the light captured by the on-chip detector The intensity of the airflow directly correlates with the bending extent of the biomimetic hair facilitating the precise detection of airflow rates through changes in the photocurrent The integrated device can measure a wide range of airflow rates from −23.87 ms−1 to 21.29 ms−1 and exhibit a rapid response time of 13 ms and a detection limit of 0.1 ms−1 the developed device holds immense potential for applications in breath detection Fiber-optic airflow sensors involve assembling external components that require precise optical alignment and other conditions it may increase the complexity and size of the sensing system and limit practical applications a promising bi-directional airflow sensing device is introduced through a GaN-based chip integrated with a biomimetic hair structure The GaN chip acts as a core component for light emission and detection Nylon fibers and PDMS are employed as biomimetic hairs the PDMS can be directly and firmly integrated with the nylon fibers through a simple curing process which reduces structural complexity and enhances the compactness of the sensing device The hair structure responds to bidirectional airflow bending to introduce optical signals captured by the integrated photodetector A comprehensive investigation into the electrical and optical properties of the device alongside its ability to detect different airflow rates is conducted to validate the effectiveness of the proposed design Schematic diagrams of a the skin structure and b the airflow sensing device Microphotographs of the GaN chip in d non-luminous and e luminous states f Schematic diagrams showing the integration of nylon fibers Optical images of g the resultant sensing device and h the biomimetic hair structure The inset shows the light output power as a function of the LED drive current b Emission spectra of the LED biased at 10 mA and normalized spectral responsivity of unbiased PD c I–V curve of PD measured at different LED currents d Plot of the PD photocurrent as a function of the LED current The solid line represents the linear fit to the data points a Schematic diagram showing the experimental setup for airflow measurement b Optical images and schematic diagrams of the sensing device under positive and negative airflow c Photocurrent as a function of airflow rate for the devices with different ratios of PDMS elastomer and curing agent d Photocurrent as a function of airflow rate for the devices with different distances between the chip and reflective film e Plot of the relationship between the photocurrent and airflow rate when D is 0.3 mm and the PDMS ratio is 7.5:1 wt% Schematic diagrams in Fig. 3b illustrate the distribution of reflected light within the device under varying airflow conditions The light emitted from the LED reaches the PET reflective film through the transparent sapphire substrate the unbent nylon fibers maintain the reflective film in a nearly horizontal state reflecting a moderate amount of light onto the PD the nylon fibers bend and drive the reflective films closer to the chip allowing the PD to receive an increased amount of light a negative airflow direction causes the nylon fibers to lift the reflective film upwards resulting in diminished light reception by the PD The polarity of the photocurrent signal from the PD serves as an indicator of airflow direction while the photocurrent magnitude provides a quantitative representation of the airflow rate PDMS with higher stiffness requires a more substantial airflow to attain its maximum deformation thereby extending the detectable airflow range the optimal PDMS mixture ratio is selected as 7.5:1 wt% Although the measurement range can potentially be extended by further increasing the stiffness of the PDMS layer there exist tradeoff in reducing the sensitivity and detection limit of the device The initial distance of the reflective film from the chip is another crucial factor that affects the intensity of the reflected light. Figure 3d shows the relationship between airflow rate and photocurrent at different initial distances (D) a photocurrent variation of approximately 5.2% is observed within the airflow rate range of −24.35 ms−1 to 22.63 ms−1 the photocurrent remains nearly unchanged for airflow rates exceeding 13.21 ms−1 and falling below −11.78 ms−1 One potential approach to expand the measurement range is to reduce the D value by positioning the PET film closer to the chip surface the measurement range can be enhanced to −23.87 ms−1 to 21.29 ms−1 further reducing D to 0.1 mm restricts the ability of the device to detect airflow limiting it to a range of −10.21 ms−1 to 9.16 ms−1 and the D is set to 0.3 mm to ensure the widest possible measurement range which also explains why the device exhibits higher sensitivity under positive airflow than negative airflow the sensitivity in the negative airflow direction is lower than that in the positive airflow direction This is because the cantilever structure of PDMS film will cause the deformation in the negative direction to be smaller than in the positive direction the fitting data analyzed under the condition of decreasing airflow aligns well with the data of increasing airflow showing the stability of the sensing device Photocurrent response when measuring the stepwise airflow change in a positive and b negative directions Response time and recovery time when applying transient c positive and d negative airflow Photocurrent response of the device under small e positive airflow and f negative airflow g Reliability measurement of the device for 1000 cycles at a positive airflow velocity of 14.16 ms−1 This is attributed to the fast photoelectric conversion in the InGaN/GaN MQWs diode structure a Photograph of an individual speaking in front of the sensing device b Photocurrent response of the device to the airflow generated when pronouncing the letters ‘S’ c Photocurrent response of the device when humans pronounce ‘Sustech’ The inset shows a sound wave image of ‘Sustech’ collected with sound acquisition software d Diagram showing the patterns of Morse code for different letters Photocurrent response to airflow through nasal breathing is associated with Morse code patterns of e the letters ‘K’ g Optical image showing a person breathing through the nose into the device Photocurrent response of the device to h normal breathing and deep breathing which presents the waveform encoding the word ‘Safe’ highlighting its suitability for real-time monitoring a distinctive feature of this device is its ability to generate two opposing peaks within each breathing cycle representing both exhalation and inhalation a rare characteristic among reported airflow sensors a Schematic diagrams and optical images of the device serving as a buffer and a NOT gate The airflow applied to the device is defined as input X b Photocurrent output response measured when the device works as the buffer and the NOT gate c Schematic diagrams and optical images of the OR and NAND gate The airflows applied to the two devices are defined as input X and input Y d Photocurrent output response when the devices function as OR gate and NAND gate All LEDs are driven by a constant current of 10 mA Furthermore, by connecting a pair of devices in parallel, the OR gate and NAND gate functions can be implemented, as shown in Fig. 6c. Figure 6d exhibits the recorded photocurrent output when applying various airflow input combinations As the PDs of two devices are connected in parallel the resultant output signal represents the cumulative sum of the photocurrent changes from both PDs In the case of the presence of more than one airflow input the elevated photocurrent indicates the high-level state thus facilitating the implementation of an OR gate when reversing the airflow inputs to a negative direction an optical airflow sensing device utilizing a GaN chip integrated with a biomimetic hair is demonstrated The proposed design effectively converts the airflow dynamics into optical changes which are subsequently captured by the GaN chip Changes in photocurrent provide a reliable indication of airflow variations The developed device exhibits superior performance in terms of small footprint These advantages render the sensing device highly suitable for applications in speech recognition and information encoding its unique capability to detect bidirectional airflow shows great potential for breath detection and the execution of logic operations Epitaxial structures comprising InGaN/GaN multiple quantum wells (MQWs) are grown on a 4-inch c-plane sapphire substrate using metal-organic chemical vapor deposition (MOCVD) Photolithography is employed to pattern 16 hexagonal regions and the remaining region is defined as the PD n-GaN surfaces are exposed using inductively coupled plasma (ICP) etching Photolithography and ICP etching are used to completely remove the 20-μm-wide GaN region between the LED and the PD p- and n-electrodes are deposited on ITO and n-GaN After the deposition of a SiO2 passivation layer is deposited using an optical thin-film coater The p-pad and n-pad are deposited via electron beam evaporation The sapphire substrate is lapped and polished and then laser-cut into chips with dimensions of 0.9 × 1.1 × 0.16 mm3 The metal pads of the chip are soldered onto a printed circuit board (PCB) The mixture is poured into the mold surface and nylon fibers (8800-C1601A) with a diameter of 1.5 mm and a length of 10 mm are arranged on the top The assembly is subsequently heated at 100 °C for 10 min an integrated film of nylon fiber and PDMS with a thickness of 0.2 mm is obtained A PET reflective film (KIMT0042) with an area of 3 × 3 mm2 and thickness of 70 μm is attached on the bottom of the film The nylon fibers and PDMS integrated film are affixed to the U-shaped portion of the PCB using PDMS gel followed by a curing process to ensure adhesion Human body micro-environment: the benefits of controlling airflow interaction Analysis of airflow over building arrays for assessment of urban wind environment Airflow in courtyard and atrium buildings in the urban environment: a wind tunnel study Application of particle image velocimetry for the measurement of the airflow characteristics in an aircraft cabin Comparison of respiratory rates derived from heart rate variability A biomimetic multifunctional electronic hair sensor Compact biomimetic hair sensors based on single silicon nanowires for ultrafast and highly-sensitive airflow detection Polymeric piezoresistive airflow sensor to monitor respiratory patterns Artificial whisker sensor with undulated morphology and self-spread piezoresistors for diverse flow analyses bi-directional airflow sensor for turbulence detection Design and fabrication of a four-electrodes PVDF fiber for a flow sensor Enhanced performance of bionic ciliary piezoelectric microsensor for hydrodynamic perception Design and characterization of a novel biaxial bionic hair flow sensor based on resonant sensing Dipole-source localization using biomimetic flow-sensor arrays positioned as lateral-line system μ-biomimetic flow-sensors—Introducing light-guiding PDMS structures into MEMS Flexible out-of-plane wind sensors with a self-powered feature inspired by fine hairs of the spider A tactile and airflow motion sensor based on flexible double-layer magnetic cilia Ultrasensitive airflow sensor prepared by electrostatic flocking for sound recognition and motion monitoring Bioinspired fluffy fabric with in situ grown carbon nanotubes for ultrasensitive wearable airflow sensor Ultrasensitive airflow sensors based on suspended carbon nanotube networks Bioinspired ultrasensitive and flexible airflow sensor based on short carbon fiber network Design and fabrication of a graphene/polyvinylidene fluoride nanocomposite-based airflow sensor Highly sensitive airflow sensor based on Fabry–Perot interferometer and Vernier effect Overview of fiber optic sensor technologies for strain/temperature sensing applications in composite materials Differential-pressure fiber-optic airflow sensor for wind tunnel testing A multifunctional airflow sensor enabled by optical micro/nanofiber Recent development of fiber-optic chemical sensors and biosensors: mechanisms Progress in high performance III-nitride micro-light-emitting diodes GaN-based light-emitting diodes on various substrates: a critical review Intensity-stabilized LEDs with monolithically integrated photodetectors Performance of InGaN green light-emitting diodes with on-chip photodetectors based on wire-bonding and flip-chip configurations Extended PDMS stiffness range for flexible systems Large deflection of cantilever beams with geometric non-linearity: Analytical and numerical approaches Experimental measurements of airflow features and velocity distribution exhaled from sneeze and speech using particle image velocimetry Respiratory function and voice: the role for airflow measures Characterizing exhaled airflow from breathing and talking Minute volumes and inspiratory flow rates during exhaustive treadmill walking using respirators Wearable breath monitoring via a hot-film/calorimetric airflow sensing system Download references The authors acknowledge financial support from the National Natural Science Foundation of China (12074170) and the Shenzhen Fundamental Research Program (JCYJ20220530113201003) These authors contributed equally: Tianyu Ye Download citation DOI: https://doi.org/10.1038/s41378-024-00794-2 high performance gallium nitride (GaN) power conversion products—announced today that it has been awarded a contract by the Advanced Research Projects Agency-Energy (ARPA-E) Part of the ARPA-E CIRCUITS program and through a sub-contract from Illinois Institute of Technology the project covers the supply of GaN-based four-quadrant switches (FQSes) for use in various power conversion applications including novel ones like current source inverters cyclo-converters for drives and microinverters The initiative is the result of Transphorm's deep GaN engineering expertise (specifically its bidirectional GaN) along with industry and university interest in further exploring the possibilities of lateral GaN switches Transphorm will prototype the FQS platform using its 650 V GaN technology that continues to offer the industry's highest threshold voltage (4 V) in a 4-pin TO-247 package The project is expected to be completed in less than a year Transphorm's standard lateral GaN FETs inherently provide bidirectional current flow certain applications such as current source inverters for motor drives cyclo-converters and matrix converters also require bidirectional voltage control to effectively manage power flow This capability is traditionally achieved by placing two FETs in series using the devices' body diode to steer and control current flow or via two IGBTs and two diodes the FQS replaces the two FET or the two IGBT+two diode approaches with a single device capable of realizing bidirectional voltage control and bidirectional current flow The FQS uses two gates to block voltage of either polarity or pass current in either direction it reduces parts required to achieve the desired result thereby enabling higher power density "It is exciting to see the day approaching when GaN-based bidirectional switches will be ready for commercial production," says Emeritus Professor Tom Jahns "Power electronics engineers have been anxiously anticipating the day when MOS-gated bidirectional switches will become available because they are the key to implementing promising power converter topologies that offer exciting opportunities for improving efficiency They hold the potential to dramatically improve the commercial viability of new products including solid-state circuit breakers and integrated motor drives by making them significantly more compact and efficient than what is achievable using today's silicon-based switches." "GaN adoption is at a point today when bringing an FQS bidirectional device to market makes sense," said Dr "Lateral GaN technology enables compact FQS dies to be fabricated because the voltage blocking region can be shared This configuration cannot be realized with vertical power device technologies which gives GaN FQSes a clear edge in performance and cost one gets true bidirectionality in a fast low-loss switch which we believe will inspire next generation power conversion products through the CIRCUITS program driven partnerships." and solutions delivered straight to your inbox.