Random and Adversarial Bit Error Robustness: Energy-Efficient and Secure DNN Accelerators

TL;DR

This paper proposes a robust training approach for DNN accelerators that enhances energy efficiency and security against bit errors and adversarial attacks without hardware modifications.

Contribution

It introduces a combination of robust quantization, weight clipping, and bit error training methods that improve robustness across various voltages and accelerators.

Findings

Reduces energy consumption by up to 30% at low bit-precision.

Maintains test accuracy within 0.8%/2% while reducing energy use.

Achieves significant robustness against adversarial bit errors, lowering test error to 26.22%.

Abstract

Deep neural network (DNN) accelerators received considerable attention in recent years due to the potential to save energy compared to mainstream hardware. Low-voltage operation of DNN accelerators allows to further reduce energy consumption, however, causes bit-level failures in the memory storing the quantized weights. Furthermore, DNN accelerators are vulnerable to adversarial attacks on voltage controllers or individual bits. In this paper, we show that a combination of robust fixed-point quantization, weight clipping, as well as random bit error training (RandBET) or adversarial bit error training (AdvBET) improves robustness against random or adversarial bit errors in quantized DNN weights significantly. This leads not only to high energy savings for low-voltage operation as well as low-precision quantization, but also improves security of DNN accelerators. In contrast to related work, our approach generalizes across operating voltages and accelerators and does not require hardware changes. Moreover, we present a novel adversarial bit error attack and are able to obtain robustness against both targeted and untargeted bit-level attacks. Without losing more than 0.8%/2% in test accuracy, we can reduce energy consumption on CIFAR10 by 20%/30% for 8/4-bit quantization. Allowing up to 320 adversarial bit errors, we reduce test error from above 90% (chance level) to 26.22%.