2023
H. Padberg, A. Regev, G. Piccolboni, A. Bricalli, G. Molas, J. F. Nodin, and S. Kvatinsky,
"Experimental Demonstration of Non-Stateful In-Memory Logic with 1T1R SiOx Valence Change Mechanism Memristors", IEEE Transactions on Circuits and Systems II: Express Briefs (in press).
O. Leitersdorf, Y. Boneh, G. Gazit, R. Ronen, and S. Kvatinsky,
"FourierPIM: High-throughput in-memory Fast Fourier Transform and polynomial multiplication", Memories - Materials, Devices, Circuits and Systems, Volume 4, July 2023.
M. Khalifa, B. Hoffer, O. Leitersdorf, R. Hanhan, L. Yavits, and S. Kvatinsky,
"ClaPIM: Scalable Sequence CLAssification using Processing-In-Memory", IEEE Transactions on Very Large Scale Integration (VLSI) Systems, July 2023 (in press).
O. Leitersdorf, D. Leitersdorf, J. Gal, M. Dahan, R. Ronen, and S. Kvatinsky,
"AritPIM: High-Throughput In-Memory Arithmetic", IEEE Transactions on Emerging Topics in Computing (TETC), April 2023.
M. A. Hadish, S. Kvatinsky, and A. Gero,
"Learning and Instruction that Combine Multiple Levels of Abstraction in Engineering: Attitudes of Students and Faculty", International Journal of Engineering Education, Vol. 39, No. 1, pp. 154–162, 2023
2022
W. Wang, L. Danial, Y. Li, E. Herbelin, E. Pikhay, Y. Roizin, B. Hoffer, Z. Wang, and S. Kvatinsky,
"A memristive deep belief neural network based on silicon synapses", Nature Electronics, December 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
M. Zou, N. Du, and S. Kvatinsky,
"Review of Security Techniques for Memristor Computing Systems", Frontiers in Electronic Materials, December 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
B. Hoffer, N. Wainstein, C. M. Neumann, E. Pop, E. Yalon, and S. Kvatinsky,
"Stateful Logic using Phase Change Memory", IEEE Journal of Exploratory Solid-State Computational Devices and Circuits Transactions on Electronic Devices, Vol. 8, No. 2, pp. 77-83, December 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
Yang Li, Wei Wang, Di Zhang, Maria Baskin, Aiping Chen, Shahar Kvatinsky, Eilam Yalon, and Lior Kornblum,
"Scalable Al2O3-TiO2 Conductive Oxide Interfaces as Defect Reservoirs for Resistive Switching Devices", Advanced Electronic Material, November 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
Z. Chen, G. Zhang, H. Cai, C. Bengel, F. Liu, X. Zhao, S. Kvatinsky, H. Schmidt, R. Waser, S. Menzel, and N. Du,
"Study on Sneak Path Effect in the Self-rectifying Crossbar Arrays based on Emerging Memristive Devices", Frontiers in Electronic Materials, October 2022, (in press).
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
I. Salameh, E. G. Friedman, and S. Kvatinsky,
"Superconducting Logic Using 2Φ Josephson Junctions with Half Flux Quantum Pulses", IEEE Transactions on Circuits and Systems II: Express Briefs, Vol. 69, No. 5, pp. 2533-2537, May 2022.
Superconductive logic based on Josephson junctions (JJ) is a promising technology for energy efficient supercomputers and cloud computing. This technology can deliver significant improvements in performance and energy efficiency as compared to CMOS. Superconductive circuits, however, suffer from low density integration as compared to CMOS, primarily due to the limited scalability of the inductors. To improve the scalability of superconductive logic, a logic family based on a novel JJ technology, 2ϕ -JJ, has been proposed that eliminates the inductors. In this brief, three circuits are presented which exploit this scalable inductor-less technology. This novel 2ϕ -JJ technology represents the data as half flux quantum (HFQ) pulses, which improves the energy efficiency and speed as compared to standard superconductive logic such as rapid single flux quantum (RSFQ). Unlike RSFQ, the proposed circuits dynamically switch upon receiving an HFQ pulse, saving energy. These 2ϕ -JJ logic circuits operate 2.25X faster and require 2.6X less energy as compared to RSFQ.
W. Wang, B. Hoffer, T. Greenberg-Toledo, Y. Li, E. Herbelin, R. Ronen, X. Xu, Y. Zhao, J. Yang, and S. Kvatinsky,
"Efficient Training of the Memristive Deep Belief Net Immune to Non-Idealities of the Synaptic Devices", Advanced Intelligent Systems, Vol. 4, No. 5, pp. 22100249, May 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
M. M. Dahan, E. T. Breyer, S. Slesazeck, T. Mikolajick, and S. Kvatinsky,
"C-AND: Mixed Writing Scheme for Disturb Reduction in 1T Ferroelectric FET Memory", IEEE Transactions on Circuits and Systems I: Regular Papers, Vol. 69, No. 4, pp. 1595-1605, April 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
O. Leitersdorf, R. Ronen, and S. Kvatinsky,
"MultPIM: Fast Stateful Multiplication for Processing-in-Memory", IEEE Transactions on Circuits and Systems II: Express Briefs, Vol. 69, No. 3, pp. 1647-1651, March 2022.
Superconductive logic based on Josephson junctions (JJ) is a promising technology for energy efficient supercomputers and cloud computing. This technology can deliver significant improvements in performance and energy efficiency as compared to CMOS. Superconductive circuits, however, suffer from low density integration as compared to CMOS, primarily due to the limited scalability of the inductors. To improve the scalability of superconductive logic, a logic family based on a novel JJ technology, 2ϕ -JJ, has been proposed that eliminates the inductors. In this brief, three circuits are presented which exploit this scalable inductor-less technology. This novel 2ϕ -JJ technology represents the data as half flux quantum (HFQ) pulses, which improves the energy efficiency and speed as compared to standard superconductive logic such as rapid single flux quantum (RSFQ). Unlike RSFQ, the proposed circuits dynamically switch upon receiving an HFQ pulse, saving energy. These 2ϕ -JJ logic circuits operate 2.25X faster and require 2.6X less energy as compared to RSFQ.
2021
W. Wang, L. Danial, E. Herbelin, B. Hoffer, B. Oved, T. Greenberg-Toledo, E. Pikhay, Y. Roizin, and S. Kvatinsky,
"Physical-Based Compact Model of Y-Flash Memristor for Neuromorphic Computation", Applied Physics Letters, Vol. 119, No. 26, December 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
Y. Li, S. Kvatinsky, and L. Kornblum,
"Harnessing Conductive Oxide Interfaces for Resistive Random-Access Memories", Frontiers in Physics, (in press).
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
T. Greenberg-Toledo, B. Perach, I. Hubara, D. Soudry, S. Kvatinsky,
"Training of Quantized Deep Neural Networks using a Magnetic Tunnel Junction-Based Synapse", Semiconductor Science and Technology, Vol. 36, No. 11, October 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
M. Zou, J. Zhou, J. Sun, C. Ji, C. Wang, and S. Kvatinsky,
"Improving Efficiency and Lifetime of Logic-in-Memory by Combining IMPLY and MAGIC Families", Journal of Systems Architecture, Vol. 119, October 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
K. Stern, N. Wainstein, Y. Keller, C. M. Neumann, E. Pop, S. Kvatinsky, and E. Yalon,
"Sub-Nanosecond Pulses Enable Partial Reset for Analog Phase Change Memory", IEEE Electron Device Letters, Vol. 42, No. 9, pp. 1291-1294, September 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
R. Ronen, A. Eliahu, O. Leitersdorf, N. Peled, K. Korgaonkar, A. Chattopadhyay, B. Perach, and S. Kvatinsky,
"The Bitlet Model: A Parameterized Analytical Model to Compare PIM and CPU Systems", ACM Journal on Emerging Technologies in Computing Systems, Vol. 18, No. 2, Article No. 43, pp. 1-29, April 2022.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Wainstein, G. Ankonina, T. Swoboda, M. Muñoz Rojo, S. Kvatinsky, and E. Yalon,
"Indirectly Heated Switch as a Platform for Nanosecond Probing of Phase Transition Properties in Chalcogenides", IEEE Transactions on Electron Devices, Vol. 68, Issue 3, pp. 1298-1303, March 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
D. Biolek, Z. Kolka, V. Biolkova, Z. Biolek, and S. Kvatinsky,
"(V)TEAM for SPICE Simulation of Memristive Devices with Improved Numerical Performance", IEEE Access, Vol. 9, No. 9, pp. 30242-30255, February 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Wainstein, E. Yalon, G. Adam, and S. Kvatinsky,
"Radiofrequency Switches Based on Emerging Resistive Memory Technologies - A Survey", Proceedings of the IEEE, Vol 109, No. 1, pp. 77-95, January 2021.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2020
N. Wainstein ,G. Ankonina, S. Kvatinsky, and E. Yalon,
"Compact Modeling and Electro-Thermal Measurements of Indirectly-Heated Phase Change RF Switches", IEEE Transactions on Electron Devices, Vol. 67, Issue 11, pp. 5182-5187, November 2020.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
B. Hoffer, V. Rana, S. Menzel R. Waser, and S. Kvatinsky,
"Experimental Demonstration of Memristor-Aided Logic (MAGIC) Using Valence Change Memory (VCM)", IEEE Transactions on Electron Devices, Vol. 67, pp. 3115-3122, August 2020.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
D. Miron, D. Cohen Azarzar, B. Hoffer, M. Baskin, S. Kvatinsky, E. Yalon and L. Kornblum,
"Oxide 2D Electron Gases as a Reservoir of Defects for Resistive Switching", Applied Physics Letters, Vol.116, Issue 22, June 2020
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
H. Abo Hanna, L. Danial, S. Kvatinsky, and R. Daniel,
"Cytomorphic Electronics with Memristors for Modeling Fundamental Genetic Circuits", IEEE Transactions on Biomedical Circuits and Systems, Vol. 14, pp. 386-401, June 2020.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
A. Eliahu, R. Ronen, P. E. Gaillardon, and S. Kvatinsky,
"multiPULPly: A Multiplication Engine for Accelerating Neural Networks on Ultra-Low-Power Architectures", ACM Journal on Emerging Technologies in Computing Systems, Vol. 1, No. 1, Article 1, January 2020.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2019
L. Danial, E. Pikhay, E. Herbelin, N. Wainstein, V. Gupta, N. Wald, Y. Roizin, R. Daniel, and S. Kvatinsky,
"Two-terminal floating-gate transistors with a low-power memristive operation mode for analogue neuromorphic computing", Nature Electronics, Vol. 2, pp. 596-605, December 2019.
Jacobs Best Paper Award
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
R. Ben-Hur, R. Ronen, A. Haj-Ali, D. Bhattacharjee, A. Eliahu, N. Peled, and S. Kvatinsky,
"SIMPLER MAGIC: Synthesis and Mapping of In-Memory Logic Executed in a Single Row to Improve Throughput", IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, Vol. 39, No. 10, pp. 2434-2447, October 2020.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
B. Perach and S. Kvatinsky,
"An Asynchronous and Low-Power True Random Number Generator using STT-MTJ", IEEE Transactions on Very Large Scale Integration Systems, Vol. 27, No. 11, pp. 2473-2484, November 2019.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
M. Ramadan, N. Wainstein, R. Ginosar, S. Kvatinsky,
"Adaptive Programming in Multi-Level Cell ReRAM", Microelectronics Journal, Vol. 90, pp. 169-180, August 2019.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Wald and S. Kvatinsky,
"Understanding the influence of device, circuit and environmental variations on real processing in memristive memory using Memristor Aided Logic", Microelectronics Journal, Vol. 86, pp. 22-33, April 2019
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
T. Greenberg, R. Mazor, A. Haj-Ali, and S. Kvatinsky,
"Supporting the Momentum Training Algorithm Using a Memristor-Based Synapse", IEEE Transactions on Circuits and Systems I: Regular Papers, Vol. 66, No. 4, pp. 1571-1583, April 2019.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Talati, H. Ha, B. Perach, R. Ronen, and S. Kvatinsky,
"CONCEPT: A Column Oriented Memory Controller for Efficient Memory and PIM Operations in RRAM", IEEE Micro, Vol. 39, No. 1, pp. 33-43, January/February 2019.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2018
E. Giacomin, T. Greenberg, S. Kvatinsky, and P.-E. Gaillardon,
"A Robust Digital RRAM-based Convolutional Block for Low-Power Image Processing and Learning Applications", IEEE Transactions on Circuits and Systems I: Regular Papers, Vol. 62, No. 2, pp. 643-654, February 2019.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
A. Haj-Ali, R. Ben-Hur, N. Wald, R. Ronen, and S. Kvatinsky,
"IMAGING-In-Memory AlGorithms for Image processiNG", IEEE Transactions on Circuits and Systems I: Regular Papers, Vol. 65, No 12, pp. 4258-4271, December 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
L. Danial, N. Wainstein, S. Kraus, and S. Kvatinsky,
"Breaking Through the Speed-Power-Accuracy Tradeoff in ADCs using a Memristive Neuromorphic Architecture", IEEE Transactions on Emerging Topics in Computational Intelligence, Vol. 2, No.5, pp. 396-409, October 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
A. Haj Ali, R. Ben Hur, N. Wald, R. Ronen, and S. Kvatinsky,
"Not in Name Alone: A Memristive Memory Processing Unit for Real In-Memory Processing", IEEE Micro, Vol. 38, No. 5, pp. 13-21, September/October 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Wainstein and S. Kvatinsky,
"A Lumped RF Model for Nanoscale Memristive Devices and Non-Volatile Single-Pole Double-Throw Switches", IEEE Transactions on Nanotechnology, vol. 17, no. 5, pp. 873-883, September 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Wainstein and S. Kvatinsky,
"TIME – Tunable Inductors using MEmristors", IEEE Transactions on Circuits and Systems I: Regular Papers, Vol. 65, No. 5, pp. 1505-1515, May 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
L. Danial, N. Wainstein, S. Kraus, and S. Kvatinsky,
"DIDACTIC: A Data-Intelligent Digital-to-Analog Converter with a Trainable Integrated Circuit using Memristors", IEEE Journal on Emerging and Selected Topics in Circuits and Systems, Vol. 8, No. 1, pp. 146-158, March 2018.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
A. Doz, I. Goldstein, and S. Kvatinsky,
"Analysis of the Row Grounding Method in a Memristor-Based Crossbar Array", International Journal of Circuit Theory and Applications, Vol. 46, No. 1, pp. 122-137, January 2018
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2017
A. Pedram, S. Richardson, S. Galal, S. Kvatinsky and M. Horowitz,
"Dark Memory and Accelerator-Rich System Optimization in the Dark Silicon Era", IEEE Design and Test, Vol. 34, No. 2, pp. 39-50, April 2017
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2016
Y. Cassuto, S. Kvatinsky, and E. Yaakobi,
"Information-Theoretic Sneak Path Mitigation in Memristor Crossbar Arrays", IEEE Transaction on Information Theory, Vol. 62, No. 9, pp. 4801-4814, September 2016.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
N. Talati, S. Gupta, P. Mane, and S. Kvatinsky,
"Logic Design within Memristive Memories Using Memristor Aided loGIC (MAGIC)", IEEE Transactions on Nanotechnology, Vol. 15, No. 4, pp. 635-650, July 2016
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
A. Morad, L. Yavits, S. Kvatinsky, and R. Ginosar,
"Resistive GP-SIMD Processing In-Memory", ACM Transactions on Architecture and Code Optimization, Vol. 12, No. 4, Article 57, January 2016
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2015
L. Yavits, S. Kvatinsky, A. Morad, and R. Ginosar,
"Resistive Associative Processor", IEEE Computer Architecture Letters, Vol. 14, No. 2, July-December 2015
Best of CAL winner 2015
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
D. Soudry, D. Di Castro, A. Gal, A. Kolodny, and S. Kvatinsky,
"Memristor-based Multilayer Neural Networks with Online Gradient Descent Training", IEEE Transactions on Neural Networks and Learning Systems , Vol. 26, No. 10, pp. 2408-2421, October 2015
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
R. Patel, S. Kvatinsky, E. G. Friedman, and A. Kolodny,
"Multistate Register Based on Resistive RAM", IEEE Transactions on Very Large Scale Integration (VLSI), Vol. 23, No. 9, pp. 1750-1759, September 2015
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
S. Kvatinsky, M. Ramadan, E. G. Friedman, and A. Kolodny,
"VTEAM – A General Model for Voltage Controlled Memristor", Transactions on Circuits and Systems II: Express Briefs, Vol. 62, No. 8, pp. 786-790, August 2015
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2014
Y. Levy, J. Bruk, Y. Cassuto, E. G. Friedman, A. Kolodny, E. Yaacobi, and S. Kvatinsky,
"Logic Operation in Memory Using a Memristive Akers Array", Microelectronics Journal, Vol. 45, No. 11, pp. 1429-1437, November 2014
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
S. Kvatinsky, D. Belousov, S. Liman, G. Satat, N. Wald, E. G. Friedman, A. Kolodny, and U. C. Weiser,
"MAGIC – Memristor Aided LoGIC", IEEE Transactions on Circuits and Systems II: Express Briefs, Vol. 61, No. 11, pp. 895- 899, November 2014.
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
S. Kvatinsky, N. Wald, G. Satat, E. G. Friedman, A. Kolodny, and U. C. Weiser,
"Memristor-based Material Implication (IMPLY) Logic: Design Principles and Methodologies", IEEE Transactions on Very Large Scale Integration (VLSI), Vol. 22, No. 10, pp. 2054-2066, October 2014
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
S. Kvatinsky, Y. H. Nacson, Y. Etsion, E. G. Friedman, A. Kolodny, and U. C. Weiser,
"Memristor-based Multithreading", IEEE Computer Architecture Letters, Vol. 13, No. 1, pp. 41-44, January-June 2014
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
2013
S. Kvatinsky, E. G. Friedman, A. Kolodny, and U. C. Weiser,
"The Desired Memristor for Circuit Designers", IEEE Circuits and Systems Magazine, second quarter, Vol. 13, No. 2, pp. 17-22, second quarter 2013
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
S. Kvatinsky, E. G. Friedman, A. Kolodny, and U. C. Weiser,
"TEAM - ThrEshold Adaptive Memristor Model", IEEE Transactions on Circuits and Systems I: Regular Paper, Vol. 60, No. 1, pp. 211-221, January 2013
2015 Guillemin-Cauer Best Paper Award
The tunability of conductance states of various emerging nonvolatile memristive devices emulates the plasticity of biological synapses, making it promising in the hardware realization of large-scale neuromorphic systems. The inference of the neural network can be greatly accelerated by the vector-matrix multiplication (VMM) performed within a crossbar array of memristive devices in one step.
Nevertheless, the implementation of the VMM needs complex peripheral circuits, and the complexity further increases as non-idealities of memristive devices prevent precise conductance tuning (especially for the online training) and largely degrade the performance of the deep neural networks (DNNs). Herein, an efficient online training method of the memristive deep belief net (DBN) is presented. The proposed memristive DBN uses stochastically binarized activations, reducing the complexity of peripheral circuits, and uses the contrastive divergence (CD)-based gradient descent learning algorithm. The analog VMM and digital CD are performed separately in a mixed-signal hardware arrangement,
making the memristive DBN highly immune to non-idealities of synaptic devices.
The number of write operations on memristive devices is reduced by two orders of magnitude. The recognition accuracy of 95–97% can be achieved for the MNIST dataset using pulsed synaptic behaviors of various memristive synaptic devices.
Journals
