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.

Ultra Low Power Associative Computing with Spin Neurons and Resistive Crossbar Memory

Spin-NeuroMem: A Low-Power Neuromorphic Associative Memory Design Based on Spintronic Devices

Leveraging Spintronic Devices for Ultra-Low Power In-Memory Computing: Logic and Neural Network.

Low Power In-Memory Computing Based On Dual-Mode Sot-Mram

Linear symmetric self-selecting 14-bit kinetic molecular memristors

Energy Efficient and High Performance Current-Mode Neural Network Circuit using Memristors and Digitally Assisted Analog CMOS Neurons

Proposal for an All-Spin Artificial Neural Network: Emulating Neural and Synaptic Functionalities Through Domain Wall Motion in Ferromagnets

IMCS2: Novel Device-to-Architecture Co-Design for Low-Power In-Memory Computing Platform Using Coterminous Spin Switch

Neuromorphic Computing Based on Resistive RAM.

Physical reservoir computing built by spintronic devices for temporal information processing

A Perovskite Memristor with Large Dynamic Space for Analog-Encoded Image Recognition

Enabling Normally-Off In Situ Computing With a Magneto-Electric FET-Based SRAM Design

Energy Scaling Advantages of Resistive Memory Crossbar Based Computation and Its Application to Sparse Coding

All Spin Artificial Neural Networks Based on Compound Spintronic Synapse and Neuron.

Resistance-Sum Architecture for Voltage-Controlled SOT-MRAM based Computing-in-Memory With Hybrid References

Enabling Normally-off In-Situ Computing with a Magneto-Electric FET-based SRAM Design

Ultralow-power in-memory computing based on ferroelectric memcapacitor network

A High-Accuracy and Low-Power Emerging Technology-Based Associative Memory

HXNOR-PBNN: A Scalable and Parallel Spintronics Synaptic Architecture for Probabilistic Binary Neural Networks

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

CMOS Back-End Compatible Memristors for in Situ Digital and Neuromorphic Computing Applications.