MemSE: Fast MSE Prediction for Noisy Memristor-Based DNN Accelerators

TL;DR

This paper presents a theoretical and fast analytical method to predict the mean squared error in memristor-based DNN accelerators, accounting for hardware noise sources, enabling efficient optimization of system parameters.

Contribution

It introduces a novel analytical approach to estimate MSE in memristor-based DNNs, significantly faster than traditional simulation methods.

Findings

Analytical predictions closely match simulation results.

The method is nearly 100 times faster than Monte-Carlo simulations.

Optimization of parameters reduces error under power constraints.

Abstract

Memristors enable the computation of matrix-vector multiplications (MVM) in memory and, therefore, show great potential in highly increasing the energy efficiency of deep neural network (DNN) inference accelerators. However, computations in memristors suffer from hardware non-idealities and are subject to different sources of noise that may negatively impact system performance. In this work, we theoretically analyze the mean squared error of DNNs that use memristor crossbars to compute MVM. We take into account both the quantization noise, due to the necessity of reducing the DNN model size, and the programming noise, stemming from the variability during the programming of the memristance value. Simulations on pre-trained DNN models showcase the accuracy of the analytical prediction. Furthermore the proposed method is almost two order of magnitude faster than Monte-Carlo simulation, thus making it possible to optimize the implementation parameters to achieve minimal error for a given power constraint.