NeuralFuse: Learning to Recover the Accuracy of Access-Limited Neural Network Inference in Low-Voltage Regimes

TL;DR

NeuralFuse is an add-on module that improves the energy efficiency of neural networks in low-voltage regimes by learning input transformations to generate error-resistant data representations, recovering accuracy without retraining.

Contribution

NeuralFuse introduces a novel, easy-to-implement method that enhances DNN robustness to SRAM bit errors in low-voltage conditions without requiring model retraining.

Findings

Reduces SRAM access energy by up to 24% at 1% bit-error rate.

Recovers up to 57% of accuracy lost due to low-voltage errors.

Applicable to remote and non-configurable hardware without retraining.

Abstract

Deep neural networks (DNNs) have become ubiquitous in machine learning, but their energy consumption remains problematically high. An effective strategy for reducing such consumption is supply-voltage reduction, but if done too aggressively, it can lead to accuracy degradation. This is due to random bit-flips in static random access memory (SRAM), where model parameters are stored. To address this challenge, we have developed NeuralFuse, a novel add-on module that handles the energy-accuracy tradeoff in low-voltage regimes by learning input transformations and using them to generate error-resistant data representations, thereby protecting DNN accuracy in both nominal and low-voltage scenarios. As well as being easy to implement, NeuralFuse can be readily applied to DNNs with limited access, such cloud-based APIs that are accessed remotely or non-configurable hardware. Our experimental results demonstrate that, at a 1% bit-error rate, NeuralFuse can reduce SRAM access energy by up to 24% while recovering accuracy by up to 57%. To the best of our knowledge, this is the first approach to addressing low-voltage-induced bit errors that requires no model retraining.