NEAT: Non-linearity Aware Training for Accurate and Energy-Efficient Implementation of Neural Networks on 1T-1R Memristive Crossbars

TL;DR

This paper introduces NEAT, a training method that accounts for transistor non-linearities in 1T-1R memristive crossbars, improving accuracy and energy efficiency in DNN implementations.

Contribution

The paper proposes a novel non-linearity aware training approach that regularizes network weights and employs heterogeneous gate control to mitigate non-idealities in 1T-1R crossbars.

Findings

Achieves ~20% energy savings with less than 1% accuracy loss.

Effectively recovers classification accuracy degraded by non-linearity.

Demonstrates improvements on CIFAR10 and CIFAR100 datasets.

Abstract

Memristive crossbars suffer from non-idealities (such as, sneak paths) that degrade computational accuracy of the Deep Neural Networks (DNNs) mapped onto them. A 1T-1R synapse, adding a transistor (1T) in series with the memristive synapse (1R), has been proposed to mitigate such non-idealities. We observe that the non-linear characteristics of the transistor affect the overall conductance of the 1T-1R cell which in turn affects the Matrix-Vector-Multiplication (MVM) operation in crossbars. This 1T-1R non-ideality arising from the input voltage-dependent non-linearity is not only difficult to model or formulate, but also causes a drastic performance degradation of DNNs when mapped onto crossbars. In this paper, we analyse the non-linearity of the 1T-1R crossbar and propose a novel Non-linearity Aware Training (NEAT) method to address the non-idealities. Specifically, we first identify the range of network weights, which can be mapped into the 1T-1R cell within the linear operating region of the transistor. Thereafter, we regularize the weights of the DNNs to exist within the linear operating range by using iterative training algorithm. Our iterative training significantly recovers the classification accuracy drop caused by the non-linearity. Moreover, we find that each layer has a different weight distribution and in turn requires different gate voltage of transistor to guarantee linear operation. Based on this observation, we achieve energy efficiency while preserving classification accuracy by applying heterogeneous gate voltage control to the 1T-1R cells across different layers. Finally, we conduct various experiments on CIFAR10 and CIFAR100 benchmark datasets to demonstrate the effectiveness of our non-linearity aware training. Overall, NEAT yields ~20% energy gain with less than 1% accuracy loss (with homogeneous gate control) when mapping ResNet18 networks on 1T-1R crossbars.