Pub Date : 2025-08-07DOI: 10.1038/s44287-025-00196-0
W.-Y. Woon, A. Kasperovich, J.-R. Wen, K. K. Hu, M. Malakoutian, J.-H. Jhang, S. Vaziri, I. Datye, C. C. Shih, J. F. Hsu, X. Y. Bao, Y. Wu, M. Nomura, S. Chowdhury, S. Sandy Liao
As transistor scaling approaches nanometre and even atomic scales, 3D stacking has become a critical enabler for advancement in the semiconductor industry, especially in high-performance computing and artificial intelligence (AI) applications. However, 3D integration introduces substantial thermal management challenges related to the increased power density and constrained heat dissipation pathways, particularly through low thermal conductivity interlayer dielectrics and complex interfaces. In this Review, we discuss state-of-the-art thermal management materials, covering their process compatibility, the critical integration challenges and the need for improved methods to enhance heat transport across interfaces. Advanced thermal characterization metrologies are introduced to highlight the need for non-destructive in-line metrologies. Finally, we provide a road map that outlines future research directions for material growth, integration and characterization methodologies to enable viable thermal solutions for 3D integration and beyond. The shrinking dimensions, the increased structural complexity and the 3D stacking of silicon-based semiconductor devices are intensifying challenges in thermal dissipation. This Review explores thermal management materials, integration challenges and characterization methods, and proposes a road map for efficient heat dissipation solutions in 3D integration.
{"title":"Thermal management materials for 3D-stacked integrated circuits","authors":"W.-Y. Woon, A. Kasperovich, J.-R. Wen, K. K. Hu, M. Malakoutian, J.-H. Jhang, S. Vaziri, I. Datye, C. C. Shih, J. F. Hsu, X. Y. Bao, Y. Wu, M. Nomura, S. Chowdhury, S. Sandy Liao","doi":"10.1038/s44287-025-00196-0","DOIUrl":"10.1038/s44287-025-00196-0","url":null,"abstract":"As transistor scaling approaches nanometre and even atomic scales, 3D stacking has become a critical enabler for advancement in the semiconductor industry, especially in high-performance computing and artificial intelligence (AI) applications. However, 3D integration introduces substantial thermal management challenges related to the increased power density and constrained heat dissipation pathways, particularly through low thermal conductivity interlayer dielectrics and complex interfaces. In this Review, we discuss state-of-the-art thermal management materials, covering their process compatibility, the critical integration challenges and the need for improved methods to enhance heat transport across interfaces. Advanced thermal characterization metrologies are introduced to highlight the need for non-destructive in-line metrologies. Finally, we provide a road map that outlines future research directions for material growth, integration and characterization methodologies to enable viable thermal solutions for 3D integration and beyond. The shrinking dimensions, the increased structural complexity and the 3D stacking of silicon-based semiconductor devices are intensifying challenges in thermal dissipation. This Review explores thermal management materials, integration challenges and characterization methods, and proposes a road map for efficient heat dissipation solutions in 3D integration.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 9","pages":"598-613"},"PeriodicalIF":0.0,"publicationDate":"2025-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123151","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-28DOI: 10.1038/s44287-025-00200-7
Silvia Conti
A article in Nature Electronics presents a metal-stamp imprinting technique for the fabrication of wafer-scale, high-quality, residue-free two-dimensional semiconductor arrays.
Pub Date : 2025-07-25DOI: 10.1038/s44287-025-00201-6
Miranda L. Vinay
An article in Nature Cities assesses the deployable potential of rooftop solar photovoltaics across Chinese cities, finding that only 42% of the national technical potential is realistically deployable.
{"title":"Realizing rooftop photovoltaics for China’s carbon neutrality goals","authors":"Miranda L. Vinay","doi":"10.1038/s44287-025-00201-6","DOIUrl":"10.1038/s44287-025-00201-6","url":null,"abstract":"An article in Nature Cities assesses the deployable potential of rooftop solar photovoltaics across Chinese cities, finding that only 42% of the national technical potential is realistically deployable.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 8","pages":"519-519"},"PeriodicalIF":0.0,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123189","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-18DOI: 10.1038/s44287-025-00191-5
Kyungseo Park, Hwayeong Jeong, Yoontae Jung, Ji-Hoon Suh, Minkyu Je, Jung Kim
The rising interest in robotics and virtual reality has driven a growing demand for intuitive interfaces that enable seamless human–robot interaction (HRI). Bio-signal-based solutions, using biopotential and bio-impedance, offer a promising approach for estimating human motion intention thanks to their ability to capture physiological neuromuscular activity in real time. This Review discusses the potential of biopotential and bio-impedance sensing systems for advancing HRI focusing on the role of integrated circuits in enabling practical applications. Biopotential and bio-impedance can be used to monitor human physiological states and motion intention, making them highly suitable for enhancing motion recognition in HRI. However, as stand-alone modalities, they face limitations related to inter-subject variability and susceptibility to noise, highlighting the need for hybrid sensing techniques. The performance of these sensing modalities is closely tied to the development of integrated circuits optimized for low-noise, low-power operation and accurate signal acquisition in a dynamic environment. Understanding the complementary strengths and limitations of biopotential and bio-impedance signals, along with the advances in integrated circuit technologies for their acquisition, highlights the potential of hybrid, multimodal systems to enable robust, intuitive and scalable HRI. The growing interest in robotics in daily life has increased the demand for intuitive interfaces for human–robot interaction (HRI). This Review examines the potential, challenges and innovations of bio-signal analysis to enhance HRI and facilitate broader applications.
{"title":"Using biopotential and bio-impedance for intuitive human–robot interaction","authors":"Kyungseo Park, Hwayeong Jeong, Yoontae Jung, Ji-Hoon Suh, Minkyu Je, Jung Kim","doi":"10.1038/s44287-025-00191-5","DOIUrl":"10.1038/s44287-025-00191-5","url":null,"abstract":"The rising interest in robotics and virtual reality has driven a growing demand for intuitive interfaces that enable seamless human–robot interaction (HRI). Bio-signal-based solutions, using biopotential and bio-impedance, offer a promising approach for estimating human motion intention thanks to their ability to capture physiological neuromuscular activity in real time. This Review discusses the potential of biopotential and bio-impedance sensing systems for advancing HRI focusing on the role of integrated circuits in enabling practical applications. Biopotential and bio-impedance can be used to monitor human physiological states and motion intention, making them highly suitable for enhancing motion recognition in HRI. However, as stand-alone modalities, they face limitations related to inter-subject variability and susceptibility to noise, highlighting the need for hybrid sensing techniques. The performance of these sensing modalities is closely tied to the development of integrated circuits optimized for low-noise, low-power operation and accurate signal acquisition in a dynamic environment. Understanding the complementary strengths and limitations of biopotential and bio-impedance signals, along with the advances in integrated circuit technologies for their acquisition, highlights the potential of hybrid, multimodal systems to enable robust, intuitive and scalable HRI. The growing interest in robotics in daily life has increased the demand for intuitive interfaces for human–robot interaction (HRI). This Review examines the potential, challenges and innovations of bio-signal analysis to enhance HRI and facilitate broader applications.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 8","pages":"555-571"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123067","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-17DOI: 10.1038/s44287-025-00192-4
Yuan-Chi Yang, Hsiu-Hau Lin, Szuya Sandy Liao
In nanoscale semiconductor devices, not only do electron–electron interactions require proper treatment but heat transport must also be integrated coherently. In this Perspective, we propose a paradigm shift: to treat electron transport using a three-part phase diagram that includes diffusive, ballistic and viscous electron-fluid regimes and to adopt a statistical-field approach to extend the tools for analysis, including the drift–diffusion model. The statistical-field approach posits that semiconductor devices — as open quantum systems characterized by fluctuating energy and particle numbers — can achieve local equilibrium through frequent microscopic collisions of electrons. The corresponding statistical fields emerge — specifically, spatial and temporal variations in temperature and chemical potential, which dictate the flows of energy and particles. The quantum nature of these statistical fields enables a seamless integration of quantum complexities, and the approach naturally incorporates heat dissipation in a self-consistent theoretical framework (although the proper modelling of boundary conditions requires further attention). We highlight the critical need to identify the transport regime in which short-channel nanodevices operate, to be able to build accurate simulators that will drive device design and optimization. This Perspective sets out an approach to electron transport in nanoscale devices based on statistical fields — specifically the spatial and temporal variations in temperature and chemical potential that drive energy and particle flow — and highlights the importance of identifying the transport regime, which might be diffusive, ballistic or viscous.
{"title":"A statistical-field approach to electron transport in semiconductor nanodevices","authors":"Yuan-Chi Yang, Hsiu-Hau Lin, Szuya Sandy Liao","doi":"10.1038/s44287-025-00192-4","DOIUrl":"10.1038/s44287-025-00192-4","url":null,"abstract":"In nanoscale semiconductor devices, not only do electron–electron interactions require proper treatment but heat transport must also be integrated coherently. In this Perspective, we propose a paradigm shift: to treat electron transport using a three-part phase diagram that includes diffusive, ballistic and viscous electron-fluid regimes and to adopt a statistical-field approach to extend the tools for analysis, including the drift–diffusion model. The statistical-field approach posits that semiconductor devices — as open quantum systems characterized by fluctuating energy and particle numbers — can achieve local equilibrium through frequent microscopic collisions of electrons. The corresponding statistical fields emerge — specifically, spatial and temporal variations in temperature and chemical potential, which dictate the flows of energy and particles. The quantum nature of these statistical fields enables a seamless integration of quantum complexities, and the approach naturally incorporates heat dissipation in a self-consistent theoretical framework (although the proper modelling of boundary conditions requires further attention). We highlight the critical need to identify the transport regime in which short-channel nanodevices operate, to be able to build accurate simulators that will drive device design and optimization. This Perspective sets out an approach to electron transport in nanoscale devices based on statistical fields — specifically the spatial and temporal variations in temperature and chemical potential that drive energy and particle flow — and highlights the importance of identifying the transport regime, which might be diffusive, ballistic or viscous.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 9","pages":"614-620"},"PeriodicalIF":0.0,"publicationDate":"2025-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-15DOI: 10.1038/s44287-025-00193-3
Wenlong Ma, Yaxue Sun, Congyu Wang, Peng Wang
The real-world deployment of the Internet of Things (IoT) infrastructures faces high energy demands. To tackle this demand, triboelectric nanogenerators and field-effect transistors (FETs) led to the emergence of tribotronic transistors that enable active mechanosensation by converting mechanical stimuli into tribo-potential, and droplet electricity generators (DEGs) that enhance the efficiency of raindrop energy harvesting through the bulk effect of FET-inspired architectures. In this Review, we explore the working mechanisms and design principles of tribotronic transistors and DEGs, highlighting the key scientific and technical challenges that must be overcome for their seamless integration into global IoT networks. We highlight the development of advanced devices for IoT data collection, memory and processing, and ambient energy harvesting in near-perpetual IoT networks, facilitating advancements in IoT applications including tactile sensors, artificial synapses, energy harvesters and self-powered sensors. Finally, we discuss key areas requiring further study, including understanding fundamental mechanisms, optimizing system design and addressing practical challenges in the application of tribotronic transistors and DEGs for large-scale IoT networks and self-powered sensors. This Review outlines the co-development of triboelectric nanogenerators and field-effect transistors into tribotronic transistors and droplet energy generators, which can harvest energy from small mechanical motion to power the Internet of Things.
{"title":"Mutual promotion of triboelectric nanogenerators and field-effect transistors towards the IoT","authors":"Wenlong Ma, Yaxue Sun, Congyu Wang, Peng Wang","doi":"10.1038/s44287-025-00193-3","DOIUrl":"10.1038/s44287-025-00193-3","url":null,"abstract":"The real-world deployment of the Internet of Things (IoT) infrastructures faces high energy demands. To tackle this demand, triboelectric nanogenerators and field-effect transistors (FETs) led to the emergence of tribotronic transistors that enable active mechanosensation by converting mechanical stimuli into tribo-potential, and droplet electricity generators (DEGs) that enhance the efficiency of raindrop energy harvesting through the bulk effect of FET-inspired architectures. In this Review, we explore the working mechanisms and design principles of tribotronic transistors and DEGs, highlighting the key scientific and technical challenges that must be overcome for their seamless integration into global IoT networks. We highlight the development of advanced devices for IoT data collection, memory and processing, and ambient energy harvesting in near-perpetual IoT networks, facilitating advancements in IoT applications including tactile sensors, artificial synapses, energy harvesters and self-powered sensors. Finally, we discuss key areas requiring further study, including understanding fundamental mechanisms, optimizing system design and addressing practical challenges in the application of tribotronic transistors and DEGs for large-scale IoT networks and self-powered sensors. This Review outlines the co-development of triboelectric nanogenerators and field-effect transistors into tribotronic transistors and droplet energy generators, which can harvest energy from small mechanical motion to power the Internet of Things.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 8","pages":"541-554"},"PeriodicalIF":0.0,"publicationDate":"2025-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123132","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-07DOI: 10.1038/s44287-025-00189-z
Sen Tan, Peilin Xie, Juan C. Vasquez, Josep M. Guerrero
Microgrids are a cornerstone of modern energy infrastructure, but the increase in digitalization presents security challenges. Cyberattacks can target various microgrid components and have the potential to disrupt operations and compromise data integrity, leading to faults, power blackouts or even physical damage. To safeguard the operation and reliability of microgrids, defence mechanisms, including detection and mitigation strategies, are being advanced. The complexity of the cybersecurity landscape, owing to the diversity of microgrid models, configurations and control algorithms, as well as the types of cyberattack and target locations, presents numerous challenges and opportunities. This Review surveys the key developments and challenges in securing microgrids against cyber threats, with a focus on microgrid control. The fundamental structure and vulnerabilities of cyber–physical microgrids are outlined, and the potential impact of cyberattacks is examined. Methods for attack detection and mitigation are identified and categorized based on microgrid modelling approaches and control objectives. Finally, emerging defence technologies and promising research opportunities in microgrid cybersecurity are highlighted. Digitalization is increasing the cyber threat to microgrids. This Review discusses the vulnerabilities of cyber–physical microgrids and examines the cyberattack detection methods and mitigation strategies being developed to increase the cybersecurity of microgrid control.
{"title":"Developments, challenges and future opportunities in cybersecure microgrid control","authors":"Sen Tan, Peilin Xie, Juan C. Vasquez, Josep M. Guerrero","doi":"10.1038/s44287-025-00189-z","DOIUrl":"10.1038/s44287-025-00189-z","url":null,"abstract":"Microgrids are a cornerstone of modern energy infrastructure, but the increase in digitalization presents security challenges. Cyberattacks can target various microgrid components and have the potential to disrupt operations and compromise data integrity, leading to faults, power blackouts or even physical damage. To safeguard the operation and reliability of microgrids, defence mechanisms, including detection and mitigation strategies, are being advanced. The complexity of the cybersecurity landscape, owing to the diversity of microgrid models, configurations and control algorithms, as well as the types of cyberattack and target locations, presents numerous challenges and opportunities. This Review surveys the key developments and challenges in securing microgrids against cyber threats, with a focus on microgrid control. The fundamental structure and vulnerabilities of cyber–physical microgrids are outlined, and the potential impact of cyberattacks is examined. Methods for attack detection and mitigation are identified and categorized based on microgrid modelling approaches and control objectives. Finally, emerging defence technologies and promising research opportunities in microgrid cybersecurity are highlighted. Digitalization is increasing the cyber threat to microgrids. This Review discusses the vulnerabilities of cyber–physical microgrids and examines the cyberattack detection methods and mitigation strategies being developed to increase the cybersecurity of microgrid control.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 8","pages":"522-540"},"PeriodicalIF":0.0,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123191","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-03DOI: 10.1038/s44287-025-00190-6
Martina Corsi, Elena Bellotti, Salvatore Surdo, Giuseppe Barillaro
Implantable bioresorbable electronic systems — comprising miniature devices that sense, process and respond to physiological cues — are reshaping precision medicine. These systems provide real-time monitoring of vital signs, biochemical markers and disease-specific indicators within the body and wirelessly transmit data that enable timely, personalized interventions. Constructed from biodegradable materials, the devices safely dissolve after completing their function, which eliminates the need for surgical removal and reduces complications. These factors position bioresorbable electronics at the forefront of sustainable and environmentally conscious technologies for personalized medicine. This Perspective describes advances in implantable bioresorbable electronics and highlights the transformative potential of these systems, their diverse medical applications and their substantial effects on healthcare. An ideal architecture for implantable bioresorbable systems is outlined and the components that provide sensing, stimulation, electronic processing, power generation and system encapsulation are described. The materials, architectures and integration strategies of each component type are discussed in detail to highlight current capabilities and emerging innovations. Critical challenges of biocompatibility, data fidelity, energy sustainability and triggered degradation that must be addressed to unlock the full potential of these technologies are also discussed. Bioresorbable electronic systems that overcome these hurdles could revolutionize patient-centred healthcare and extend the reach of sustainable electronic technologies. Implantable bioresorbable electronic systems are revolutionizing precision medicine with real-time health monitoring, targeted interventions and biodegradability. This Perspective discusses architectural designs, key functional components, challenges and the transformative potential of sustainable electronics in patient-centred care.
{"title":"Implantable bioresorbable electronic systems for sustainable precision medicine","authors":"Martina Corsi, Elena Bellotti, Salvatore Surdo, Giuseppe Barillaro","doi":"10.1038/s44287-025-00190-6","DOIUrl":"10.1038/s44287-025-00190-6","url":null,"abstract":"Implantable bioresorbable electronic systems — comprising miniature devices that sense, process and respond to physiological cues — are reshaping precision medicine. These systems provide real-time monitoring of vital signs, biochemical markers and disease-specific indicators within the body and wirelessly transmit data that enable timely, personalized interventions. Constructed from biodegradable materials, the devices safely dissolve after completing their function, which eliminates the need for surgical removal and reduces complications. These factors position bioresorbable electronics at the forefront of sustainable and environmentally conscious technologies for personalized medicine. This Perspective describes advances in implantable bioresorbable electronics and highlights the transformative potential of these systems, their diverse medical applications and their substantial effects on healthcare. An ideal architecture for implantable bioresorbable systems is outlined and the components that provide sensing, stimulation, electronic processing, power generation and system encapsulation are described. The materials, architectures and integration strategies of each component type are discussed in detail to highlight current capabilities and emerging innovations. Critical challenges of biocompatibility, data fidelity, energy sustainability and triggered degradation that must be addressed to unlock the full potential of these technologies are also discussed. Bioresorbable electronic systems that overcome these hurdles could revolutionize patient-centred healthcare and extend the reach of sustainable electronic technologies. Implantable bioresorbable electronic systems are revolutionizing precision medicine with real-time health monitoring, targeted interventions and biodegradability. This Perspective discusses architectural designs, key functional components, challenges and the transformative potential of sustainable electronics in patient-centred care.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 8","pages":"572-583"},"PeriodicalIF":0.0,"publicationDate":"2025-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123068","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-02DOI: 10.1038/s44287-025-00187-1
Corey Lammie, Hadjer Benmeziane, William Simon, Elena Ferro, Athanasios Vasilopoulos, Julian Büchel, Manuel Le Gallo, Irem Boybat, Abu Sebastian
Analogue in-memory computing (AIMC) is an emerging computational paradigm that can efficiently accelerate the key operations in deep learning (DL) inference workloads. Heterogeneous architectures, which integrate both AIMC tiles and digital processing units, have been proposed to enable the end-to-end execution of various deep neural network models. However, developing a software stack for these architectures is challenging, owing to their distinct characteristics — such as the need for extensive or complete weight stationarity and pipelined execution across layers, if maximum performance is to be achieved. Moreover, AIMC tiles are inherently stochastic and hence introduce a combination of stochastic and deterministic noise, which adversely affects accuracy. As a result, existing tools for software stack development are not directly applicable. In this Perspective, we give an overview of the key attributes of DL software stacks and AIMC-based accelerators, outline the challenges associated with designing DL software stacks for AIMC-based accelerators and present opportunities for future research. Analogue in-memory computing (AIMC), with digital processing, forms a useful architecture for performant end-to-end execution of deep neural network models, but requires the development of sophisticated software stacks. This Perspective outlines the challenges in designing deep learning software stacks for AIMC-based accelerators, and suggests directions for future research.
{"title":"Deep learning software stacks for analogue in-memory computing-based accelerators","authors":"Corey Lammie, Hadjer Benmeziane, William Simon, Elena Ferro, Athanasios Vasilopoulos, Julian Büchel, Manuel Le Gallo, Irem Boybat, Abu Sebastian","doi":"10.1038/s44287-025-00187-1","DOIUrl":"10.1038/s44287-025-00187-1","url":null,"abstract":"Analogue in-memory computing (AIMC) is an emerging computational paradigm that can efficiently accelerate the key operations in deep learning (DL) inference workloads. Heterogeneous architectures, which integrate both AIMC tiles and digital processing units, have been proposed to enable the end-to-end execution of various deep neural network models. However, developing a software stack for these architectures is challenging, owing to their distinct characteristics — such as the need for extensive or complete weight stationarity and pipelined execution across layers, if maximum performance is to be achieved. Moreover, AIMC tiles are inherently stochastic and hence introduce a combination of stochastic and deterministic noise, which adversely affects accuracy. As a result, existing tools for software stack development are not directly applicable. In this Perspective, we give an overview of the key attributes of DL software stacks and AIMC-based accelerators, outline the challenges associated with designing DL software stacks for AIMC-based accelerators and present opportunities for future research. Analogue in-memory computing (AIMC), with digital processing, forms a useful architecture for performant end-to-end execution of deep neural network models, but requires the development of sophisticated software stacks. This Perspective outlines the challenges in designing deep learning software stacks for AIMC-based accelerators, and suggests directions for future research.","PeriodicalId":501701,"journal":{"name":"Nature Reviews Electrical Engineering","volume":"2 9","pages":"621-633"},"PeriodicalIF":0.0,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145123054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：