Long short term memory based hardware prefetcher

Abstract

Hardware prefetching is an efficient mechanism to hide cache miss penalties. Accuracy, coverage, and timeliness are three primary metrics in evaluating prefetcher performance. Highly accurate hardware prefetches are desired to predict complex memory access patterns in multicore systems. In this paper, we propose a long short term memory (LSTM) prefetcher---a neural network based hardware prefetcher. Offline experiment shows that the proposed LSTM prefetcher achieves higher accuracy and better coverage on a set of evaluated traces.