A Low-Energy Set-Associative I-Cache Design with Last Accessed Way Based Replacement and Predicting Access Policy

Predicting access is known as an approach to reduce the set-associative I-cache's energy consumption. This paper proposes a LAW (Last Accessed Way) based low-energy I-cache, taking both the replacement and predicting access policy into account. By replacing and accessing instructions speculatively following the last accessed way, the PHR (predict-hit-rate) is significantly improved. Simulator results show that compared with the conventional LRU (Least Recently Used) replacement based set-associative I-cache, such a 16 KB, 4-way and 32B-line-size I-cache can reduce about 47%~57% energy consumption, with only 2.63% performance overhead. In contrast with the way-prediction cache, the proposed I-cache could improve nearly 5.32% performance overhead and 5.95% energy consumption at little cost of hardware.