STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM

This paper proposes an in-memory binary spiking neural network (BSNN) based on spin-transfer-torque magnetoresistive RAM (STT-MRAM). We propose residual BSNN learning using a surrogate gradient that shortens the time steps in the BSNN while maintaining sufficient accuracy. At the circuit level, pres...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Van-Tinh Nguyen, Quang-Kien Trinh, Renyuan Zhang, Yasuhiko Nakashima
Formato: article
Lenguaje:EN
Publicado: IEEE 2021
Materias:
Acceso en línea:https://doaj.org/article/fdc82db0a0f44b1380ee2886de2abf87
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
id oai:doaj.org-article:fdc82db0a0f44b1380ee2886de2abf87
record_format dspace
spelling oai:doaj.org-article:fdc82db0a0f44b1380ee2886de2abf872021-11-17T00:01:17ZSTT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM2169-353610.1109/ACCESS.2021.3125685https://doaj.org/article/fdc82db0a0f44b1380ee2886de2abf872021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9605611/https://doaj.org/toc/2169-3536This paper proposes an in-memory binary spiking neural network (BSNN) based on spin-transfer-torque magnetoresistive RAM (STT-MRAM). We propose residual BSNN learning using a surrogate gradient that shortens the time steps in the BSNN while maintaining sufficient accuracy. At the circuit level, presynaptic spikes are fed to memory units through differential bit lines (BLs), while binarized weights are stored in a subarray of nonvolatile STT-MRAM. When the common inputs are fed through BLs, vector-to-matrix multiplication can be performed in a single memory sensing phase, hence achieving massive parallelism with low power and low latency. We further introduce the concept of a dynamic threshold to reduce the implementation complexity of synapses and neuron circuitry. This adjustable threshold also permits a nonlinear batch normalization (BN) function to be incorporated into the integrate-and-fire (IF) neuron circuit. The circuitry greatly improves the overall performance and enables high regularity in circuit layouts. Our proposed netlist circuits are built on a 65-nm CMOS with a fitted magnetic tunnel junction (MTJ) model for performance evaluation. The hardware/software co-simulation results indicate that the proposed design can deliver a performance of 176.6 TOPS/W for an in-memory computing (IMC) subarray size of <inline-formula> <tex-math notation="LaTeX">$1\times 288$ </tex-math></inline-formula>. The classification accuracy reaches 97.92&#x0025; (83.85&#x0025;) on the MNIST (CIFAR-10) dataset. The impacts of the device non-idealities and process variations are also thoroughly covered in the analysis.Van-Tinh NguyenQuang-Kien TrinhRenyuan ZhangYasuhiko NakashimaIEEEarticleBinary spiking neural networkemerging memory technologyin-memory computingneuromorphic computingprocess variationSTT-MRAMElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 9, Pp 151373-151385 (2021)
institution DOAJ
collection DOAJ
language EN
topic Binary spiking neural network
emerging memory technology
in-memory computing
neuromorphic computing
process variation
STT-MRAM
Electrical engineering. Electronics. Nuclear engineering
TK1-9971
spellingShingle Binary spiking neural network
emerging memory technology
in-memory computing
neuromorphic computing
process variation
STT-MRAM
Electrical engineering. Electronics. Nuclear engineering
TK1-9971
Van-Tinh Nguyen
Quang-Kien Trinh
Renyuan Zhang
Yasuhiko Nakashima
STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
description This paper proposes an in-memory binary spiking neural network (BSNN) based on spin-transfer-torque magnetoresistive RAM (STT-MRAM). We propose residual BSNN learning using a surrogate gradient that shortens the time steps in the BSNN while maintaining sufficient accuracy. At the circuit level, presynaptic spikes are fed to memory units through differential bit lines (BLs), while binarized weights are stored in a subarray of nonvolatile STT-MRAM. When the common inputs are fed through BLs, vector-to-matrix multiplication can be performed in a single memory sensing phase, hence achieving massive parallelism with low power and low latency. We further introduce the concept of a dynamic threshold to reduce the implementation complexity of synapses and neuron circuitry. This adjustable threshold also permits a nonlinear batch normalization (BN) function to be incorporated into the integrate-and-fire (IF) neuron circuit. The circuitry greatly improves the overall performance and enables high regularity in circuit layouts. Our proposed netlist circuits are built on a 65-nm CMOS with a fitted magnetic tunnel junction (MTJ) model for performance evaluation. The hardware/software co-simulation results indicate that the proposed design can deliver a performance of 176.6 TOPS/W for an in-memory computing (IMC) subarray size of <inline-formula> <tex-math notation="LaTeX">$1\times 288$ </tex-math></inline-formula>. The classification accuracy reaches 97.92&#x0025; (83.85&#x0025;) on the MNIST (CIFAR-10) dataset. The impacts of the device non-idealities and process variations are also thoroughly covered in the analysis.
format article
author Van-Tinh Nguyen
Quang-Kien Trinh
Renyuan Zhang
Yasuhiko Nakashima
author_facet Van-Tinh Nguyen
Quang-Kien Trinh
Renyuan Zhang
Yasuhiko Nakashima
author_sort Van-Tinh Nguyen
title STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
title_short STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
title_full STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
title_fullStr STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
title_full_unstemmed STT-BSNN: An In-Memory Deep Binary Spiking Neural Network Based on STT-MRAM
title_sort stt-bsnn: an in-memory deep binary spiking neural network based on stt-mram
publisher IEEE
publishDate 2021
url https://doaj.org/article/fdc82db0a0f44b1380ee2886de2abf87
work_keys_str_mv AT vantinhnguyen sttbsnnaninmemorydeepbinaryspikingneuralnetworkbasedonsttmram
AT quangkientrinh sttbsnnaninmemorydeepbinaryspikingneuralnetworkbasedonsttmram
AT renyuanzhang sttbsnnaninmemorydeepbinaryspikingneuralnetworkbasedonsttmram
AT yasuhikonakashima sttbsnnaninmemorydeepbinaryspikingneuralnetworkbasedonsttmram
_version_ 1718426086572818432