Recurrent neural networks(RNNs)have proven to be indispensable for processing sequential and temporal data,with extensive applications in language modeling,text generation,machine translation,and time-series forecasti...Recurrent neural networks(RNNs)have proven to be indispensable for processing sequential and temporal data,with extensive applications in language modeling,text generation,machine translation,and time-series forecasting.Despite their versatility,RNNs are frequently beset by significant training expenses and slow convergence times,which impinge upon their deployment in edge AI applications.Reservoir computing(RC),a specialized RNN variant,is attracting increased attention as a cost-effective alternative for processing temporal and sequential data at the edge.RC’s distinctive advantage stems from its compatibility with emerging memristive hardware,which leverages the energy efficiency and reduced footprint of analog in-memory and in-sensor computing,offering a streamlined and energy-efficient solution.This review offers a comprehensive explanation of RC’s underlying principles,fabrication processes,and surveys recent progress in nano-memristive device based RC systems from the viewpoints of in-memory and in-sensor RC function.It covers a spectrum of memristive device,from established oxide-based memristive device to cutting-edge material science developments,providing readers with a lucid understanding of RC’s hardware implementation and fostering innovative designs for in-sensor RC systems.Lastly,we identify prevailing challenges and suggest viable solutions,paving the way for future advancements in in-sensor RC technology.展开更多
The utilization of processing capabilities within the detector holds significant promise in addressing energy consumption and latency challenges. Especially in the context of dynamic motion recognition tasks, where su...The utilization of processing capabilities within the detector holds significant promise in addressing energy consumption and latency challenges. Especially in the context of dynamic motion recognition tasks, where substantial data transfers are necessitated by the generation of extensive information and the need for frame-by-frame analysis. Herein, we present a novel approach for dynamic motion recognition, leveraging a spatial-temporal in-sensor computing system rooted in multiframe integration by employing photodetector. Our approach introduced a retinomorphic MoS_(2) photodetector device for motion detection and analysis. The device enables the generation of informative final states, nonlinearly embedding both past and present frames. Subsequent multiply-accumulate (MAC) calculations are efficiently performed as the classifier. When evaluating our devices for target detection and direction classification, we achieved an impressive recognition accuracy of 93.5%. By eliminating the need for frame-by-frame analysis, our system not only achieves high precision but also facilitates energy-efficient in-sensor computing.展开更多
AI development has brought great success to upgrading the information age.At the same time,the large-scale artificial neural network for building AI systems is thirsty for computing power,which is barely satisfied by ...AI development has brought great success to upgrading the information age.At the same time,the large-scale artificial neural network for building AI systems is thirsty for computing power,which is barely satisfied by the conventional computing hardware.In the post-Moore era,the increase in computing power brought about by the size reduction of CMOS in very large-scale integrated circuits(VLSIC)is challenging to meet the growing demand for AI computing power.To address the issue,technical approaches like neuromorphic computing attract great attention because of their feature of breaking Von-Neumann architecture,and dealing with AI algorithms much more parallelly and energy efficiently.Inspired by the human neural network architecture,neuromorphic computing hardware is brought to life based on novel artificial neurons constructed by new materials or devices.Although it is relatively difficult to deploy a training process in the neuromorphic architecture like spiking neural network(SNN),the development in this field has incubated promising technologies like in-sensor computing,which brings new opportunities for multidisciplinary research,including the field of optoelectronic materials and devices,artificial neural networks,and microelectronics integration technology.The vision chips based on the architectures could reduce unnecessary data transfer and realize fast and energy-efficient visual cognitive processing.This paper reviews firstly the architectures and algorithms of SNN,and artificial neuron devices supporting neuromorphic computing,then the recent progress of in-sensor computing vision chips,which all will promote the development of AI.展开更多
The artificial intelligence era has witnessed a surge of demand in detection and recognition of biometric information,with applications from financial services to information security.However,the physical separation o...The artificial intelligence era has witnessed a surge of demand in detection and recognition of biometric information,with applications from financial services to information security.However,the physical separation of sensing,memory,and computational units in traditional biometric systems introduces severe decision latency and operational power consumption.Herein,an in-sensor reservoir computing(RC)system based on MoTe_(2)/BaTiO_(3)optical synapses is proposed to detect and recognize the faces and fingerprints information.In optical operation mode,the device exhibits low energy consumption of 41.2 pJ,long retention time of 3×10^(4)s,high endurance of 10^(4)switching cycles,and multifunctional sensing-memory-computing visual simulations.The light intensity-dependent optical sensing and multilevel optical storage properties are exploited to achieve sunburned eye simulation and image memory functions.These nonlinear,multi-state,short-term storage,and long-term memory characteristics make MoTe_(2)/BaTiO_(3)optical synapses a suitable reservoir layer and readout layer,with short-term properties to project complicated input features into high-dimensional output features,and long-term properties to be used as a readout layer,thus further building an in-sensor RC system for face and fingerprint recognition.Under the 40%Gaussian noise environment,the system achieves 91.73%recognition accuracy for face and 97.50%for fingerprint images,and experimental verification is carried out,which shows potential in practical applications.These results provide a strategy for constructing a high-performance in-sensor RC system for high-accuracy biometric identification.展开更多
In-optical-sensor computing architectures based on neuro-inspired optical sensor arrays have become key milestones for in-sensor artificial intelligence(AI)technology,enabling intelligent vision sensing and extensive ...In-optical-sensor computing architectures based on neuro-inspired optical sensor arrays have become key milestones for in-sensor artificial intelligence(AI)technology,enabling intelligent vision sensing and extensive data processing.These architectures must demonstrate potential advantages in terms of mass production and complementary metal oxide semiconductor compatibility.Here,we introduce a visible-light-driven neuromorphic vision system that integrates front-end retinomorphic photosensors with a back-end artificial neural network(ANN),employing a single neuro-inspired indium-g allium-zinc-oxide photo transistor(NIP)featuring an aluminum sensitization layer(ASL).By methodically adjusting the ASL coverage on IGZO phototransistors,a fast-switching response-type and a synaptic response-type of IGZO photo transistors are successfully developed.Notably,the fabricated NIP shows a remarkable retina-like photoinduced synaptic plasticity under wavelengths up to 635 nm,with over256-states,weight update nonlinearity below 0.1,and a dynamic range of 64.01.Owing to this technology,a 6×6 neuro-inspired optical image sensor array with the NIP can perform highly integrated sensing,memory,and preprocessing functions,including contrast enhancement,and handwritten digit image recognition.The demonstrated prototype highlights the potential for efficient hardware implementations in in-sensor AI technologies.展开更多
Reconfigurable devices can be used to achieve multiple logic operation and intelligent optical sensing with low power consumption,which is promising candidates for new generation electronic and optoelectronic integrat...Reconfigurable devices can be used to achieve multiple logic operation and intelligent optical sensing with low power consumption,which is promising candidates for new generation electronic and optoelectronic integrated circuits.However,the versatility is still limited and need to be extended by the device architectures design.Here,we report an asymmetrically gate two-dimensional(2D)van der Waals heterostructure with hybrid dielectric layer SiO_(2)/hexagonal boron nitride(h-BN),which enable rich function including reconfigurable logic operation and in-sensor information encryption enabled by both volatile and non-volatile optoelectrical modulation.When the partial gate is grounded,the non-volatile light assisted electrostatic doping endowed partially reconfigurable doping between n-type and p-type,which allow the switching of logic XOR and not implication(NIMP).When the global gate is grounded,additionally taking the optical signal as another input signal,logic AND and OR is realized by combined regulation of the light and localized gate voltage.Depending on the high on/off current ratio approaching 105 and reliable&switchable logic gate,in-sensor information encryption and decryption is demonstrated by manipulating the logic output.Hence,these results provide strong extension for current reconfigurable electronic and optoelectronic devices.展开更多
Rapid developments in the Internet of Things and Artificial Intelligence trigger higher requirements for image perception and learning of external environments through visual systems.However,limited by von Neumann'...Rapid developments in the Internet of Things and Artificial Intelligence trigger higher requirements for image perception and learning of external environments through visual systems.However,limited by von Neumann's bottleneck,the physical separation of sense,memory,and processing units in a conventional personal computer-based vision system tend to consume a significant amount of energy,time latency,and additional hardware costs.By integrating computational tasks of multiple functionalities into the sensors themselves,the emerging bio-inspired neuromorphic visual systems provide an opportunity to overcome these limitations.With high speed,ultralow power and strong adaptability,it is highly desirable to develop a neuromorphic vision system that is based on highly precise in-sensor computing devices,namely retinomorphic devices.We here present a timely review of retinomorphic devices for visual in-sensor computing.We begin with several types of physical mechanisms of photoelectric sensors that can be constructed for artificial vision.The potential applications of retinomorphic hardware are,thereafter,thoroughly summarized.We also highlight the possible strategies to existing challenges and give a brief perspective of retinomorphic architecture for in-sensor computing.展开更多
With the development of artificial intelligence and the Internet of Things,the number of sensory nodes is growing rapidly,leading to the exchange of large quantities of redundant data between sensors and computing uni...With the development of artificial intelligence and the Internet of Things,the number of sensory nodes is growing rapidly,leading to the exchange of large quantities of redundant data between sensors and computing units.Insensor computing schemes,which integrate sensing and processing,have provided a promising route to addressing the sensing/processing bottleneck by reducing power consumption,time delay and hardware redundancy.In this study,an in-sensor computing architecture involving a photoelectronic cell based on a wafer-scale two-dimensional MoS_(2)thin film was demonstrated.The MoS_(2)photodetector cell used a top-gate device structure with in-dium tin oxide(ITO)as the transparent gate electrode,which exhibited high air-stability and a high photoresponsivity(R)up to 555.8 A W^(-1) at an illumination power density(P_(in))of 16.0μW cm^(-2)(λ=500 nm).Additionally,a MoS_(2)photodetector array with uniform photoresponsive characteristics was achieved.Furthermore,logic gates,including inverter,NAND,and NOR,were achieved based on MoS_(2)photodetector cells.Such multifunctional and robust in-sensor computing was ascribed to the uniform wafer-scale MoS_(2)film grown by atomic layer deposition(ALD)and the unique device structure.Because the detection of optical signals and logic operations were achieved through MoS_(2)photodetector cells with area efficiency,the proposed in-sensor computing device paves the way for potential applications in high-performance,integrated sensing and processing systems.展开更多
Memtransistors in which the source-drain channel conductance can be nonvolatilely manipulated through the gate signals have emerged as promising components for implementing neuromorphic computing.On the other side,it ...Memtransistors in which the source-drain channel conductance can be nonvolatilely manipulated through the gate signals have emerged as promising components for implementing neuromorphic computing.On the other side,it is known that the complementary metal-oxide-semiconductor(CMOS)field effect transistors have played the fundamental role in the modern integrated circuit technology.Therefore,will complementary memtransistors(CMT)also play such a role in the future neuromorphic circuits and chips?In this review,various types of materials and physical mechanisms for constructing CMT(how)are inspected with their merits and need-to-address challenges discussed.Then the unique properties(what)and poten-tial applications of CMT in different learning algorithms/scenarios of spiking neural networks(why)are reviewed,including super-vised rule,reinforcement one,dynamic vision with in-sensor computing,etc.Through exploiting the complementary structure-related novel functions,significant reduction of hardware consuming,enhancement of energy/efficiency ratio and other advan-tages have been gained,illustrating the alluring prospect of design technology co-optimization(DTCO)of CMT towards neuro-morphic computing.展开更多
The latest developments in bio-inspired neuromorphic vision sensors can be summarized in 3 keywords:smaller,faster,and smarter.(1)Smaller:Devices are becoming more compact by integrating previously separated component...The latest developments in bio-inspired neuromorphic vision sensors can be summarized in 3 keywords:smaller,faster,and smarter.(1)Smaller:Devices are becoming more compact by integrating previously separated components such as sensors,memory,and processing units.As a prime example,the transition from traditional sensory vision computing to in-sensor vision computing has shown clear benefits,such as simpler circuitry,lower power consumption,and less data redundancy.(2)Swifter:Owing to the nature of physics,smaller and more integrated devices can detect,process,and react to input more quickly.In addition,the methods for sensing and processing optical information using various materials(such as oxide semiconductors)are evolving.(3)Smarter:Owing to these two main research directions,we can expect advanced applications such as adaptive vision sensors,collision sensors,and nociceptive sensors.This review mainly focuses on the recent progress,working mechanisms,image pre-processing techniques,and advanced features of two types of neuromorphic vision sensors based on near-sensor and in-sensor vision computing methodologies.展开更多
随着位置服务(location based service,LBS)应用需求的日益增加以及多部位微机电系统(micro electro mechanical system,MEMS)导航传感器的广泛普及,行人航位推算(pedestrian dead reckoning,PDR)越来越受关注,成为行人导航研究中主流...随着位置服务(location based service,LBS)应用需求的日益增加以及多部位微机电系统(micro electro mechanical system,MEMS)导航传感器的广泛普及,行人航位推算(pedestrian dead reckoning,PDR)越来越受关注,成为行人导航研究中主流的技术之一。但是,低成本的MEMS传感器测量噪声大,PDR解算误差积累严重;且PDR算法的普适性差,不同穿戴位置的MEMS导航传感器约束条件的可用性差异明显。提出了一种基于穿戴式MEMS传感器状态识别的多部位PDR算法。首先,采用支持向量机(support vector machine,SVM)进行全监督训练,实现了静止状态及运动状态下手部、腿部、腰部、足部4种穿戴位置的准确识别;然后,分析了不同穿戴位置下PDR算法的适用性,根据适用性分析结果提出了多部位PDR的综合解算策略。实测结果表明,该方法能够动态、准确地实现穿戴式MEMS传感器的状态识别,正确率达97%以上;应用PDR综合解算策略后,足部PDR能够实现高精度解算,累计误差为0.74%,而其他位置(手部、腿部、腰部)解算效果得到显著改善,累计误差从识别前的6.76%~21.19%减小为2.92%~5.62%。展开更多
基金supported by National Key Research and Development Program of China(Grant No.2022YFA1405600)Beijing Natural Science Foundation(Grant No.Z210006)+3 种基金National Natural Science Foundation of China—Young Scientists Fund(Grant No.12104051,62122004)Hong Kong Research Grant Council(Grant Nos.27206321,17205922,17212923 and C1009-22GF)Shenzhen Science and Technology Innovation Commission(SGDX20220530111405040)partially supported by ACCESS—AI Chip Center for Emerging Smart Systems,sponsored by Innovation and Technology Fund(ITF),Hong Kong SAR。
文摘Recurrent neural networks(RNNs)have proven to be indispensable for processing sequential and temporal data,with extensive applications in language modeling,text generation,machine translation,and time-series forecasting.Despite their versatility,RNNs are frequently beset by significant training expenses and slow convergence times,which impinge upon their deployment in edge AI applications.Reservoir computing(RC),a specialized RNN variant,is attracting increased attention as a cost-effective alternative for processing temporal and sequential data at the edge.RC’s distinctive advantage stems from its compatibility with emerging memristive hardware,which leverages the energy efficiency and reduced footprint of analog in-memory and in-sensor computing,offering a streamlined and energy-efficient solution.This review offers a comprehensive explanation of RC’s underlying principles,fabrication processes,and surveys recent progress in nano-memristive device based RC systems from the viewpoints of in-memory and in-sensor RC function.It covers a spectrum of memristive device,from established oxide-based memristive device to cutting-edge material science developments,providing readers with a lucid understanding of RC’s hardware implementation and fostering innovative designs for in-sensor RC systems.Lastly,we identify prevailing challenges and suggest viable solutions,paving the way for future advancements in in-sensor RC technology.
基金supported by the National Natural Science Foundation of China (52322210, 52172144, 22375069, 21825103, and U21A2069)National Key R&D Program of China (2021YFA1200501)+2 种基金Shenzhen Science and Technology Program (JCYJ20220818102215033, JCYJ20200109105422876)the Innovation Project of Optics Valley Laboratory (OVL2023PY007)Science and Technology Commission of Shanghai Municipality (21YF1454700)。
文摘The utilization of processing capabilities within the detector holds significant promise in addressing energy consumption and latency challenges. Especially in the context of dynamic motion recognition tasks, where substantial data transfers are necessitated by the generation of extensive information and the need for frame-by-frame analysis. Herein, we present a novel approach for dynamic motion recognition, leveraging a spatial-temporal in-sensor computing system rooted in multiframe integration by employing photodetector. Our approach introduced a retinomorphic MoS_(2) photodetector device for motion detection and analysis. The device enables the generation of informative final states, nonlinearly embedding both past and present frames. Subsequent multiply-accumulate (MAC) calculations are efficiently performed as the classifier. When evaluating our devices for target detection and direction classification, we achieved an impressive recognition accuracy of 93.5%. By eliminating the need for frame-by-frame analysis, our system not only achieves high precision but also facilitates energy-efficient in-sensor computing.
基金Project supported in part by the National Key Research and Development Program of China(Grant No.2021YFA0716400)the National Natural Science Foundation of China(Grant Nos.62225405,62150027,61974080,61991443,61975093,61927811,61875104,62175126,and 62235011)+2 种基金the Ministry of Science and Technology of China(Grant Nos.2021ZD0109900 and 2021ZD0109903)the Collaborative Innovation Center of Solid-State Lighting and Energy-Saving ElectronicsTsinghua University Initiative Scientific Research Program.
文摘AI development has brought great success to upgrading the information age.At the same time,the large-scale artificial neural network for building AI systems is thirsty for computing power,which is barely satisfied by the conventional computing hardware.In the post-Moore era,the increase in computing power brought about by the size reduction of CMOS in very large-scale integrated circuits(VLSIC)is challenging to meet the growing demand for AI computing power.To address the issue,technical approaches like neuromorphic computing attract great attention because of their feature of breaking Von-Neumann architecture,and dealing with AI algorithms much more parallelly and energy efficiently.Inspired by the human neural network architecture,neuromorphic computing hardware is brought to life based on novel artificial neurons constructed by new materials or devices.Although it is relatively difficult to deploy a training process in the neuromorphic architecture like spiking neural network(SNN),the development in this field has incubated promising technologies like in-sensor computing,which brings new opportunities for multidisciplinary research,including the field of optoelectronic materials and devices,artificial neural networks,and microelectronics integration technology.The vision chips based on the architectures could reduce unnecessary data transfer and realize fast and energy-efficient visual cognitive processing.This paper reviews firstly the architectures and algorithms of SNN,and artificial neuron devices supporting neuromorphic computing,then the recent progress of in-sensor computing vision chips,which all will promote the development of AI.
基金supported by the National Key R&D Plan“Nano Frontier”Key Special Project(Grant No.2021YFA1200502)Cultivation Projects of National Major R&D Project(Grant No.92164109)+13 种基金the National Natural Science Foundation of China(Grant Nos.61874158,62004056,and 62104058)the Special Project of Strategic Leading Science and Technology of Chinese Academy of Sciences(Grant No.XDB44000000-7)Key R&D Plan Projects in Hebei Province(Grant No.22311101D)Hebei Basic Research Special Key Project(Grant No.F2021201045)the Support Program for the Top Young Talents of Hebei Province(Grant No.70280011807)the Supporting Plan for 100 Excellent Innovative Talents in Colleges and Universities of Hebei Province(Grant No.SLRC2019018)the Interdisciplinary Research Program of Natural Science of Hebei University(No.DXK202101)the Institute of Life Sciences and Green Development(No.521100311)the Natural Science Foundation of Hebei Province(Nos.F2022201054 and F2021201022)the Outstanding Young Scientific Research and Innovation Team of Hebei University(Grant No.605020521001)the Special Support Funds for National High Level Talents(Grant No.041500120001)the Advanced Talents Incubation Program of the Hebei University(Grant Nos.521000981426,521100221071,and 521000981363)the Science and Technology Project of Hebei Education Department(Grant Nos.QN2020178 and QN2021026)Postgraduate's Innovation Fund Project of Hebei Province(CXZZBS2024004).
文摘The artificial intelligence era has witnessed a surge of demand in detection and recognition of biometric information,with applications from financial services to information security.However,the physical separation of sensing,memory,and computational units in traditional biometric systems introduces severe decision latency and operational power consumption.Herein,an in-sensor reservoir computing(RC)system based on MoTe_(2)/BaTiO_(3)optical synapses is proposed to detect and recognize the faces and fingerprints information.In optical operation mode,the device exhibits low energy consumption of 41.2 pJ,long retention time of 3×10^(4)s,high endurance of 10^(4)switching cycles,and multifunctional sensing-memory-computing visual simulations.The light intensity-dependent optical sensing and multilevel optical storage properties are exploited to achieve sunburned eye simulation and image memory functions.These nonlinear,multi-state,short-term storage,and long-term memory characteristics make MoTe_(2)/BaTiO_(3)optical synapses a suitable reservoir layer and readout layer,with short-term properties to project complicated input features into high-dimensional output features,and long-term properties to be used as a readout layer,thus further building an in-sensor RC system for face and fingerprint recognition.Under the 40%Gaussian noise environment,the system achieves 91.73%recognition accuracy for face and 97.50%for fingerprint images,and experimental verification is carried out,which shows potential in practical applications.These results provide a strategy for constructing a high-performance in-sensor RC system for high-accuracy biometric identification.
基金supported by the National Research Foundation of Korea(NRF)Grant funded by the Korea government(MSIT)(Grant No.RS-2023-00256917)Samsung Display。
文摘In-optical-sensor computing architectures based on neuro-inspired optical sensor arrays have become key milestones for in-sensor artificial intelligence(AI)technology,enabling intelligent vision sensing and extensive data processing.These architectures must demonstrate potential advantages in terms of mass production and complementary metal oxide semiconductor compatibility.Here,we introduce a visible-light-driven neuromorphic vision system that integrates front-end retinomorphic photosensors with a back-end artificial neural network(ANN),employing a single neuro-inspired indium-g allium-zinc-oxide photo transistor(NIP)featuring an aluminum sensitization layer(ASL).By methodically adjusting the ASL coverage on IGZO phototransistors,a fast-switching response-type and a synaptic response-type of IGZO photo transistors are successfully developed.Notably,the fabricated NIP shows a remarkable retina-like photoinduced synaptic plasticity under wavelengths up to 635 nm,with over256-states,weight update nonlinearity below 0.1,and a dynamic range of 64.01.Owing to this technology,a 6×6 neuro-inspired optical image sensor array with the NIP can perform highly integrated sensing,memory,and preprocessing functions,including contrast enhancement,and handwritten digit image recognition.The demonstrated prototype highlights the potential for efficient hardware implementations in in-sensor AI technologies.
基金supported by the Beijing Natural Science Foundation(No.JQ20027)the National Science Foundation of China(No.62305013)+2 种基金China Postdoctoral Science Foundation(No.2023M730137)the China National Postdoctoral Program for Innovative Talents(No.BX20230033)Beijing Postdoctoral Research Foundation(No.2023-zz-95).
文摘Reconfigurable devices can be used to achieve multiple logic operation and intelligent optical sensing with low power consumption,which is promising candidates for new generation electronic and optoelectronic integrated circuits.However,the versatility is still limited and need to be extended by the device architectures design.Here,we report an asymmetrically gate two-dimensional(2D)van der Waals heterostructure with hybrid dielectric layer SiO_(2)/hexagonal boron nitride(h-BN),which enable rich function including reconfigurable logic operation and in-sensor information encryption enabled by both volatile and non-volatile optoelectrical modulation.When the partial gate is grounded,the non-volatile light assisted electrostatic doping endowed partially reconfigurable doping between n-type and p-type,which allow the switching of logic XOR and not implication(NIMP).When the global gate is grounded,additionally taking the optical signal as another input signal,logic AND and OR is realized by combined regulation of the light and localized gate voltage.Depending on the high on/off current ratio approaching 105 and reliable&switchable logic gate,in-sensor information encryption and decryption is demonstrated by manipulating the logic output.Hence,these results provide strong extension for current reconfigurable electronic and optoelectronic devices.
基金supported by National Key Research and Development Program of China(2021YFA1200700)The National Natural Science Foundation of China(No.T2222025 and 62174053)+1 种基金Open Research Projects of Zhejiang Lab(2021MD0AB03),Shanghai Science and Technology Innovation Action Plan(21JC1402000 and 21520714100)the Fundamental Research Funds for the Central Universities.The authors would like to express their gratitude to EditSprings(https://www.editsprings.cn)for the expert linguistic services provided.
文摘Rapid developments in the Internet of Things and Artificial Intelligence trigger higher requirements for image perception and learning of external environments through visual systems.However,limited by von Neumann's bottleneck,the physical separation of sense,memory,and processing units in a conventional personal computer-based vision system tend to consume a significant amount of energy,time latency,and additional hardware costs.By integrating computational tasks of multiple functionalities into the sensors themselves,the emerging bio-inspired neuromorphic visual systems provide an opportunity to overcome these limitations.With high speed,ultralow power and strong adaptability,it is highly desirable to develop a neuromorphic vision system that is based on highly precise in-sensor computing devices,namely retinomorphic devices.We here present a timely review of retinomorphic devices for visual in-sensor computing.We begin with several types of physical mechanisms of photoelectric sensors that can be constructed for artificial vision.The potential applications of retinomorphic hardware are,thereafter,thoroughly summarized.We also highlight the possible strategies to existing challenges and give a brief perspective of retinomorphic architecture for in-sensor computing.
基金supported by the young scientist project of MOE innovation platform,the Science and Technology Commission of Shanghai Municipality(21DZ1100700)China Postdoctoral Science Foundation(Grant BX2021070,2021M700026)the Zhejiang Lab’s International Talent Fund for Young Professionals and Jiaxing Science and Technology Project(Grants No.2021AY10057).
文摘With the development of artificial intelligence and the Internet of Things,the number of sensory nodes is growing rapidly,leading to the exchange of large quantities of redundant data between sensors and computing units.Insensor computing schemes,which integrate sensing and processing,have provided a promising route to addressing the sensing/processing bottleneck by reducing power consumption,time delay and hardware redundancy.In this study,an in-sensor computing architecture involving a photoelectronic cell based on a wafer-scale two-dimensional MoS_(2)thin film was demonstrated.The MoS_(2)photodetector cell used a top-gate device structure with in-dium tin oxide(ITO)as the transparent gate electrode,which exhibited high air-stability and a high photoresponsivity(R)up to 555.8 A W^(-1) at an illumination power density(P_(in))of 16.0μW cm^(-2)(λ=500 nm).Additionally,a MoS_(2)photodetector array with uniform photoresponsive characteristics was achieved.Furthermore,logic gates,including inverter,NAND,and NOR,were achieved based on MoS_(2)photodetector cells.Such multifunctional and robust in-sensor computing was ascribed to the uniform wafer-scale MoS_(2)film grown by atomic layer deposition(ALD)and the unique device structure.Because the detection of optical signals and logic operations were achieved through MoS_(2)photodetector cells with area efficiency,the proposed in-sensor computing device paves the way for potential applications in high-performance,integrated sensing and processing systems.
基金supported by the National Key Research and Development Program of China(No.2023YFB4502200)Natural Science Foundation of China(Nos.92164204 and 62374063)the Science and Technology Major Project of Hubei Province(No.2022AEA001).
文摘Memtransistors in which the source-drain channel conductance can be nonvolatilely manipulated through the gate signals have emerged as promising components for implementing neuromorphic computing.On the other side,it is known that the complementary metal-oxide-semiconductor(CMOS)field effect transistors have played the fundamental role in the modern integrated circuit technology.Therefore,will complementary memtransistors(CMT)also play such a role in the future neuromorphic circuits and chips?In this review,various types of materials and physical mechanisms for constructing CMT(how)are inspected with their merits and need-to-address challenges discussed.Then the unique properties(what)and poten-tial applications of CMT in different learning algorithms/scenarios of spiking neural networks(why)are reviewed,including super-vised rule,reinforcement one,dynamic vision with in-sensor computing,etc.Through exploiting the complementary structure-related novel functions,significant reduction of hardware consuming,enhancement of energy/efficiency ratio and other advan-tages have been gained,illustrating the alluring prospect of design technology co-optimization(DTCO)of CMT towards neuro-morphic computing.
基金This work was supported by the National Research Foundation of Korea(NRF)grant funded by the Korea government(MSIT)(No.NRF-2019R1A2C2002447)This research also was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF)funded by the Ministry of Education(No.NRF-2014R1A6A1030419)This work also was supported by Korea Institute for Advancement of Technology(KIAT)grant funded by the Korea Government(MOTIE)(P0020967,Advanced Training Program for Smart Sensor Engineers).
文摘The latest developments in bio-inspired neuromorphic vision sensors can be summarized in 3 keywords:smaller,faster,and smarter.(1)Smaller:Devices are becoming more compact by integrating previously separated components such as sensors,memory,and processing units.As a prime example,the transition from traditional sensory vision computing to in-sensor vision computing has shown clear benefits,such as simpler circuitry,lower power consumption,and less data redundancy.(2)Swifter:Owing to the nature of physics,smaller and more integrated devices can detect,process,and react to input more quickly.In addition,the methods for sensing and processing optical information using various materials(such as oxide semiconductors)are evolving.(3)Smarter:Owing to these two main research directions,we can expect advanced applications such as adaptive vision sensors,collision sensors,and nociceptive sensors.This review mainly focuses on the recent progress,working mechanisms,image pre-processing techniques,and advanced features of two types of neuromorphic vision sensors based on near-sensor and in-sensor vision computing methodologies.
文摘随着位置服务(location based service,LBS)应用需求的日益增加以及多部位微机电系统(micro electro mechanical system,MEMS)导航传感器的广泛普及,行人航位推算(pedestrian dead reckoning,PDR)越来越受关注,成为行人导航研究中主流的技术之一。但是,低成本的MEMS传感器测量噪声大,PDR解算误差积累严重;且PDR算法的普适性差,不同穿戴位置的MEMS导航传感器约束条件的可用性差异明显。提出了一种基于穿戴式MEMS传感器状态识别的多部位PDR算法。首先,采用支持向量机(support vector machine,SVM)进行全监督训练,实现了静止状态及运动状态下手部、腿部、腰部、足部4种穿戴位置的准确识别;然后,分析了不同穿戴位置下PDR算法的适用性,根据适用性分析结果提出了多部位PDR的综合解算策略。实测结果表明,该方法能够动态、准确地实现穿戴式MEMS传感器的状态识别,正确率达97%以上;应用PDR综合解算策略后,足部PDR能够实现高精度解算,累计误差为0.74%,而其他位置(手部、腿部、腰部)解算效果得到显著改善,累计误差从识别前的6.76%~21.19%减小为2.92%~5.62%。
