Resistance Random Access Memory(ReRAM)crossbar arrays have been used in compute in-memory(CIM)application owing to its high bit-density,non-volatility,and capability to perform multiplyaccumulate(MAC)calculations effi...Resistance Random Access Memory(ReRAM)crossbar arrays have been used in compute in-memory(CIM)application owing to its high bit-density,non-volatility,and capability to perform multiplyaccumulate(MAC)calculations efficiently.The expansion of the size of the crossbars has led to the emerging challenge of high IR voltage drop and more complex logic control devices.In this paper,we propose a progressive weight pruning strategy based on gradient sensitivity analysis to reduce redundant parameters and enhance overall sparsity.Building upon this sparsity-enhanced structure,we further introduce two complementary weight quantization-mapping methods tailored for high-bit and low-bit quantization scenarios.The proposed method utilizes group quantization for clustering to merge weights in higher bits and leverages differential properties to conduct spectral clustering for merging weights in lower bits.Experimental results indicate notable savings in crossbar resources with minimal loss of precision.Moreover,we designed a carrier board-FPGA testing platform and deployed a neural network on a 32×32 size ReRAM crossbar.The results show that the proposed algorithm saves 42%of units,and the recognition accuracy of the MNIST dataset is within an acceptable range(91.5%to 88.3%).展开更多
The paper presents the results of experimental studies of synaptic plasticity in a memristive memory element based on nanocrystalline ZnO films grown by pulsed laser deposition.The obtained results can be used in the ...The paper presents the results of experimental studies of synaptic plasticity in a memristive memory element based on nanocrystalline ZnO films grown by pulsed laser deposition.The obtained results can be used in the development of technological bases for the formation of high-performance multi-level artificial synapses for elements of neuroelectronics and hardware neural networks.展开更多
Artificial intelligence(AI)advancements are driving the need for highly paral-lel and energy-efficient computing analogous to the human brain and visualsystem.Inspired by the human brain,resistive random-access memori...Artificial intelligence(AI)advancements are driving the need for highly paral-lel and energy-efficient computing analogous to the human brain and visualsystem.Inspired by the human brain,resistive random-access memories(ReRAMs)have recently emerged as an essential component of the intelligentcircuitry architecture for developing high-performance neuromorphic comput-ing systems.This occurs due to their fast switching with ultralow power con-sumption,high ON/OFF ratio,excellent data retention,good endurance,andeven great possibilities for altering resistance analogous to their biologicalcounterparts for neuromorphic computing applications.Additionally,with theadvantages of photoelectric dual modulation of resistive switching,ReRAMsallow optically inspired artificial neural networks and reconfigurable logicoperations,promoting innovative in-memory computing technology forneuromorphic computing and image recognition tasks.Optoelectronicneuromorphic computing architectured ReRAMs can simulate neural func-tionalities,such as light-triggered long-term/short-term plasticity.They can beused in intelligent robotics and bionic neurological optoelectronic systems.Metal oxide(MOx)–polymer hybrid nanocomposites can be beneficial as anactive layer of the bistable metal–insulator–metal ReRAM devices,which holdpromise for developing high-performance memory technology.This reviewexplores the state of the art for developing memory storage,advancement inmaterials,and switching mechanisms for selecting the appropriate materials asactive layers of ReRAMs to boost the ON/OFF ratio,flexibility,and memorydensity while lowering programming voltage.Furthermore,material designcum-synthesis strategies that greatly influence the overall performance of MOx–polymer hybrid nanocomposite ReRAMs and their performances arehighlighted.Additionally,the recent progress of multifunctional optoelectronicMOx–polymer hybrid composites-based ReRAMs are explored as artificial syn-apses for neural networks to emulate neuromorphic visualization and memo-rize information.Finally,the challenges,limitations,and future outlooks ofthe fabrication of MOx–polymer hybrid composite ReRAMs over the conven-tional von Neumann computing systems are discussed.展开更多
The monolithic three-dimensional integration of memory and logic circuits could dramatically improve the performance and energy efficiency of computing systems. Some conventional and emerging memories are suitable for...The monolithic three-dimensional integration of memory and logic circuits could dramatically improve the performance and energy efficiency of computing systems. Some conventional and emerging memories are suitable for vertical integration, including highly scalable metal-oxide resistive switching devices （＂memristors＂）. However, the integration of logic circuits has proven to be much more challenging than expected. In this study, we demonstrated memory and logic functionality in a monolithic three-dimensional circuit by adapting the recently proposed memristor-based stateful material implication logic. By modifying the original circuit to increase its robustness to device imperfections, we experimentally showed, for the first time, a reliable multi-cycle multi-gate material implication logic operation and half-adder circuit within a three- dimensional stack of monolithically integrated memristors. Direct data manipulation in three dimensions enables extremely compact and high-throughput logic- in-memory computing and, remarkably, presents a viable solution for the Feynman Grand Challenge of implementing an 8-bit adder at the nanoscale.展开更多
Monte Carlo （MC） simulations, including multiple physical and chemical mechanisms, were performed to investigate the microstructure evolution of a conducting metal filament in a typical oxide-electrolyte-based ReRAM...Monte Carlo （MC） simulations, including multiple physical and chemical mechanisms, were performed to investigate the microstructure evolution of a conducting metal filament in a typical oxide-electrolyte-based ReRAM. It has been revealed that the growth direction and geometry of the conductive filament are controlled by the ion migration rate in the electrolyte layer during the formation procedure. When the migration rate is rela- tive high, the filament is shown to grow from cathode to anode. When the migration rate is low, the growth direction is expected to start from the anode. Simulated conductive filament （CF） geometries and I-V characteristics are also illustrated and analyzed. A good agreement between the simulation results and experiment data is obtained.展开更多
Modern recommendation systems are widely used in modern data centers.The random and sparse embedding lookup operations are the main performance bottleneck for processing recommendation systems on traditional platforms...Modern recommendation systems are widely used in modern data centers.The random and sparse embedding lookup operations are the main performance bottleneck for processing recommendation systems on traditional platforms as they induce abundant data movements between computing units and memory.ReRAM-based processing-in-memory(PIM)can resolve this problem by processing embedding vectors where they are stored.However,the embedding table can easily exceed the capacity limit of a monolithic ReRAM-based PIM chip,which induces off-chip accesses that may offset the PIM profits.Therefore,we deploy the decomposed model on-chip and leverage the high computing efficiency of ReRAM to compensate for the decompression performance loss.In this paper,we propose ARCHER,a ReRAM-based PIM architecture that implements fully yon-chip recommendations under resource constraints.First,we make a full analysis of the computation pattern and access pattern on the decomposed table.Based on the computation pattern,we unify the operations of each layer of the decomposed model in multiply-and-accumulate operations.Based on the access observation,we propose a hierarchical mapping schema and a specialized hardware design to maximize resource utilization.Under the unified computation and mapping strategy,we can coordinatethe inter-processing elements pipeline.The evaluation shows that ARCHER outperforms the state-of-the-art GPU-based DLRM system,the state-of-the-art near-memory processing recommendation system RecNMP,and the ReRAM-based recommendation accelerator REREC by 15.79×,2.21×,and 1.21× in terms of performance and 56.06×,6.45×,and 1.71× in terms of energy savings,respectively.展开更多
It is investigated for the effect ofa ferroelectric Si：Hf02 thin film on the resistive switching in a stacked Pt/Si：HfO2/highly-oxygen-deficient HfO2-x/Pt structure. Improved resistance performance was observed. It ...It is investigated for the effect ofa ferroelectric Si：Hf02 thin film on the resistive switching in a stacked Pt/Si：HfO2/highly-oxygen-deficient HfO2-x/Pt structure. Improved resistance performance was observed. It was concluded that the observed resistive switching behavior was related to the modulation of the width and height of a depletion barrier in the HfO2-x layer, which was caused by the Si：HfO2 ferroelectric polarization field effect. Reliable switching reproducibility and long data retention were observed in these memory cells, suggesting their great potential in non-volatile memories applications with full compatibility and simplicity.展开更多
基金supported in part by the National Key Research and Development Program of China under Grant 2021YFA0717700in part by Nanjing University-China Mobile Communications Group Company,Ltd.Joint Institute.
文摘Resistance Random Access Memory(ReRAM)crossbar arrays have been used in compute in-memory(CIM)application owing to its high bit-density,non-volatility,and capability to perform multiplyaccumulate(MAC)calculations efficiently.The expansion of the size of the crossbars has led to the emerging challenge of high IR voltage drop and more complex logic control devices.In this paper,we propose a progressive weight pruning strategy based on gradient sensitivity analysis to reduce redundant parameters and enhance overall sparsity.Building upon this sparsity-enhanced structure,we further introduce two complementary weight quantization-mapping methods tailored for high-bit and low-bit quantization scenarios.The proposed method utilizes group quantization for clustering to merge weights in higher bits and leverages differential properties to conduct spectral clustering for merging weights in lower bits.Experimental results indicate notable savings in crossbar resources with minimal loss of precision.Moreover,we designed a carrier board-FPGA testing platform and deployed a neural network on a 32×32 size ReRAM crossbar.The results show that the proposed algorithm saves 42%of units,and the recognition accuracy of the MNIST dataset is within an acceptable range(91.5%to 88.3%).
基金funded by the Russian Science Foundation Grant No.23-79-10272,https://rscf.ru/project/23-79-10272/at the Southern Federal University(regarding the growth of nanocrystalline zinc oxide films)financial support of the scientific program of the[National Center for Physics and Mathematics],section 9“Artificial intelligence and big data in technical,industrial,natural and social systems”(regarding electrical measurements of nanocrystalline zinc oxide films).
文摘The paper presents the results of experimental studies of synaptic plasticity in a memristive memory element based on nanocrystalline ZnO films grown by pulsed laser deposition.The obtained results can be used in the development of technological bases for the formation of high-performance multi-level artificial synapses for elements of neuroelectronics and hardware neural networks.
基金Council of Scientific and Industrial Research,India,Grant/Award Number:08/096(0012)/2020-EMR-IGovernment of Uttar Pradesh,India,Grant/Award Numbers:108/2021/2585/Sattar-4-2021-4(28)/2021/20,78/2022/1984/Sattar-4-2022-003-70-4099/7/022/19,CST/D-1524+1 种基金Chaudhary Charan Singh University,India,Grant/Award Number:Dev./1043/29.06.2022National Research Foundation of Korea,Grant/Award Numbers:2019R1A2C1085448,2023R1A2C1005421。
文摘Artificial intelligence(AI)advancements are driving the need for highly paral-lel and energy-efficient computing analogous to the human brain and visualsystem.Inspired by the human brain,resistive random-access memories(ReRAMs)have recently emerged as an essential component of the intelligentcircuitry architecture for developing high-performance neuromorphic comput-ing systems.This occurs due to their fast switching with ultralow power con-sumption,high ON/OFF ratio,excellent data retention,good endurance,andeven great possibilities for altering resistance analogous to their biologicalcounterparts for neuromorphic computing applications.Additionally,with theadvantages of photoelectric dual modulation of resistive switching,ReRAMsallow optically inspired artificial neural networks and reconfigurable logicoperations,promoting innovative in-memory computing technology forneuromorphic computing and image recognition tasks.Optoelectronicneuromorphic computing architectured ReRAMs can simulate neural func-tionalities,such as light-triggered long-term/short-term plasticity.They can beused in intelligent robotics and bionic neurological optoelectronic systems.Metal oxide(MOx)–polymer hybrid nanocomposites can be beneficial as anactive layer of the bistable metal–insulator–metal ReRAM devices,which holdpromise for developing high-performance memory technology.This reviewexplores the state of the art for developing memory storage,advancement inmaterials,and switching mechanisms for selecting the appropriate materials asactive layers of ReRAMs to boost the ON/OFF ratio,flexibility,and memorydensity while lowering programming voltage.Furthermore,material designcum-synthesis strategies that greatly influence the overall performance of MOx–polymer hybrid nanocomposite ReRAMs and their performances arehighlighted.Additionally,the recent progress of multifunctional optoelectronicMOx–polymer hybrid composites-based ReRAMs are explored as artificial syn-apses for neural networks to emulate neuromorphic visualization and memo-rize information.Finally,the challenges,limitations,and future outlooks ofthe fabrication of MOx–polymer hybrid composite ReRAMs over the conven-tional von Neumann computing systems are discussed.
文摘The monolithic three-dimensional integration of memory and logic circuits could dramatically improve the performance and energy efficiency of computing systems. Some conventional and emerging memories are suitable for vertical integration, including highly scalable metal-oxide resistive switching devices （＂memristors＂）. However, the integration of logic circuits has proven to be much more challenging than expected. In this study, we demonstrated memory and logic functionality in a monolithic three-dimensional circuit by adapting the recently proposed memristor-based stateful material implication logic. By modifying the original circuit to increase its robustness to device imperfections, we experimentally showed, for the first time, a reliable multi-cycle multi-gate material implication logic operation and half-adder circuit within a three- dimensional stack of monolithically integrated memristors. Direct data manipulation in three dimensions enables extremely compact and high-throughput logic- in-memory computing and, remarkably, presents a viable solution for the Feynman Grand Challenge of implementing an 8-bit adder at the nanoscale.
基金Project supported by the Ministry of Science and Technology of China(Nos.2010CB934200,2011CBA00602,2009CB930803,2011CB921804,2011AA010401,2011AA010402,XDA06020102)the National Natural Science Foundation of China(Nos.61221004,61274091,60825403,61106119,61106082,61306117)
文摘Monte Carlo （MC） simulations, including multiple physical and chemical mechanisms, were performed to investigate the microstructure evolution of a conducting metal filament in a typical oxide-electrolyte-based ReRAM. It has been revealed that the growth direction and geometry of the conductive filament are controlled by the ion migration rate in the electrolyte layer during the formation procedure. When the migration rate is rela- tive high, the filament is shown to grow from cathode to anode. When the migration rate is low, the growth direction is expected to start from the anode. Simulated conductive filament （CF） geometries and I-V characteristics are also illustrated and analyzed. A good agreement between the simulation results and experiment data is obtained.
基金This work was supported by the National Key R&D Program of China(No.2022YFB4501403)the National Natural Science Foundation of China(Grant Nos.62322205,62072195,61825202,and 61832006)the Zhejiang Lab(No.2022PI0AC02).
文摘Modern recommendation systems are widely used in modern data centers.The random and sparse embedding lookup operations are the main performance bottleneck for processing recommendation systems on traditional platforms as they induce abundant data movements between computing units and memory.ReRAM-based processing-in-memory(PIM)can resolve this problem by processing embedding vectors where they are stored.However,the embedding table can easily exceed the capacity limit of a monolithic ReRAM-based PIM chip,which induces off-chip accesses that may offset the PIM profits.Therefore,we deploy the decomposed model on-chip and leverage the high computing efficiency of ReRAM to compensate for the decompression performance loss.In this paper,we propose ARCHER,a ReRAM-based PIM architecture that implements fully yon-chip recommendations under resource constraints.First,we make a full analysis of the computation pattern and access pattern on the decomposed table.Based on the computation pattern,we unify the operations of each layer of the decomposed model in multiply-and-accumulate operations.Based on the access observation,we propose a hierarchical mapping schema and a specialized hardware design to maximize resource utilization.Under the unified computation and mapping strategy,we can coordinatethe inter-processing elements pipeline.The evaluation shows that ARCHER outperforms the state-of-the-art GPU-based DLRM system,the state-of-the-art near-memory processing recommendation system RecNMP,and the ReRAM-based recommendation accelerator REREC by 15.79×,2.21×,and 1.21× in terms of performance and 56.06×,6.45×,and 1.71× in terms of energy savings,respectively.
基金supported by the National Natural Science Foundation of China(No.11374182)the Natural Science Foundation of Shandong Province(No.ZR2012FQ012)the Jinan Independent Innovation Projects of Universities(No.201303019)
文摘It is investigated for the effect ofa ferroelectric Si：Hf02 thin film on the resistive switching in a stacked Pt/Si：HfO2/highly-oxygen-deficient HfO2-x/Pt structure. Improved resistance performance was observed. It was concluded that the observed resistive switching behavior was related to the modulation of the width and height of a depletion barrier in the HfO2-x layer, which was caused by the Si：HfO2 ferroelectric polarization field effect. Reliable switching reproducibility and long data retention were observed in these memory cells, suggesting their great potential in non-volatile memories applications with full compatibility and simplicity.
