Abstract:The brain performs intelligent tasks with extremely low energy consumption. This work takes its inspiration from two strategies used by the brain to achieve this energy efficiency: the absence of separation between computing and memory functions and reliance on low-precision computation. The emergence of resistive memory technologies indeed provides an opportunity to tightly co-integrate logic and memory in hardware. In parallel, the recently proposed concept of a Binarized Neural Network, where multiplications are replaced by exclusive NOR (XNOR) logic gates, offers a way to implement artificial intelligence using very low precision computation. In this work, we therefore propose a strategy for implementing low-energy Binarized Neural Networks that employs brain-inspired concepts while retaining the energy benefits of digital electronics. We design, fabricate, and test a memory array, including periphery and sensing circuits, that is optimized for this in-memory computing scheme. Our circuit employs hafnium oxide resistive memory integrated in the back end of line of a 130-nm CMOS process, in a two-transistor, two-resistor cell, which allows the exclusive NOR operations of the neural network to be performed directly within the sense amplifiers. We show, based on extensive electrical measurements, that our design allows a reduction in the number of bit errors on the synaptic weights without the use of formal error-correcting codes. We design a whole system using this memory array. We show on standard machine learning tasks (MNIST, CIFAR-10, ImageNet, and an ECG task) that the system has inherent resilience to bit errors. We evidence that its energy consumption is attractive compared to more standard approaches and that it can use memory devices in regimes where they exhibit particularly low programming energy and high endurance. We conclude the work by discussing how it associates biologically plausible ideas with more traditional digital electronics concepts.

Binary Convolutional Neural Network Based on 3d Field-Programmable Ferroelectric Diode Array

DaDianNao: A Machine-Learning Supercomputer

On-chip 4F-System Based on Concave Mirrors for Optical Neural Networks

Complementary Memory Cell Based on Field-Programmable Ferroelectric Diode for Ultra-Low Power Current-SA Free BNN Applications

Design of Ferroelectric FET-Based Capacitive-Coupling Computing-In-Memory for Binary Neural Networks

A Proposal for Energy-Efficient Cellular Neural Network based on Spintronic Devices

Monolithically Integrated Complementary Ferroelectric FET XNOR Synapse for the Binary Neural Network

Clements-Enhanced Complex-Valued Coherent Mesh with Balanced Detection Units for Photonic Neural Networks

Binary Convolutional Neural Network on RRAM.

Analog Deep Neural Network Based On Nor Flash Computing Array For High Speed/Energy Efficiency Computation

Space-efficient optical computing with an integrated chip diffractive neural network

Single chip photonic deep neural network with accelerated training

A Convolution Neural Network Implemented by Three 3 × 3 Photonic Integrated Reconfigurable Linear Processors

Design And Optimization Of Fefet-Based Crossbars For Binary Convolution Neural Networks

High area efficiency binary neural processing elements with time-domain signals using SFQ circuits

Binary Neural Network with 16 Mb Rram Macro Chip for Classification and Online Training

Hardware Implementation of Energy Efficient Deep Learning Neural Network Based on Nanoscale Flash Computing Array

Hadamard Product-Based In-Memory Computing Design for Floating Point Neural Network Training

Ultra-high Density and Nonvolatile Nanophotonic Convolution Processing Unit

CMOS-compatible compute-in-memory accelerators based on integrated ferroelectric synaptic arrays for convolution neural networks

Digital Biologically Plausible Implementation of Binarized Neural Networks With Differential Hafnium Oxide Resistive Memory Arrays