Zero-Space Cost Fault Tolerance for Transformer-based Language Models on ReRAM

TL;DR

This paper introduces a zero-space cost fault tolerance mechanism for transformer-based language models on ReRAM, combining structure pruning, weight duplication, and MSB embedding to enhance robustness without additional storage.

Contribution

It proposes a novel fault protection method that achieves fault tolerance in ReRAM-based transformers without extra space, using innovative weight and structure modifications.

Findings

Effective fault tolerance on nine GLUE tasks with BERT

No additional storage overhead achieved

Improved robustness against hardware faults

Abstract

Resistive Random Access Memory (ReRAM) has emerged as a promising platform for deep neural networks (DNNs) due to its support for parallel in-situ matrix-vector multiplication. However, hardware failures, such as stuck-at-fault defects, can result in significant prediction errors during model inference. While additional crossbars can be used to address these failures, they come with storage overhead and are not efficient in terms of space, energy, and cost. In this paper, we propose a fault protection mechanism that incurs zero space cost. Our approach includes: 1) differentiable structure pruning of rows and columns to reduce model redundancy, 2) weight duplication and voting for robust output, and 3) embedding duplicated most significant bits (MSBs) into the model weight. We evaluate our method on nine tasks of the GLUE benchmark with the BERT model, and experimental results prove its effectiveness.