Fault-Tolerant Dot-Product Engines

TL;DR

This paper introduces coding schemes for dot-product engines that enable error correction and detection in integer vector-matrix multiplication, enhancing fault tolerance in computational hardware.

Contribution

It proposes novel coding schemes for fault-tolerant dot-product engines that handle errors under both $L_1$-metric and Hamming metric.

Findings

Coding schemes successfully correct errors in dot-product computations.

The schemes detect errors efficiently under different metrics.

Enhanced reliability in hardware-based vector-matrix multiplication.

Abstract

Coding schemes are presented that provide the ability to correct and detect computational errors while using dot-product engines for integer vector--matrix multiplication. Both the $L_1$-metric and the Hamming metric are considered.