Memory accesses typically consume the majority of power in computer systems and there is an active research community developing more energy efficient memories. Most of the proposed emerging memories promise significant efficiency improvements, but suffer from limitations such as bad write endurance or restricted access schemes. Software programming machinery of processors contributes heavily to memory hierarchy power consumption due to the need to deliver instruction words at almost every clock cycle of execution from the memory hierarchy. This project aims to minimize this overhead by developing essential techniques for efficiently utilizing emerging memory technologies to implement energy efficient instruction streams. The results will be useful for the whole spectrum of computing from tiny IoT devices to datacenter servers. Ecological and economical impact potential is also major since growing part of greenhouse emissions are being caused by computing.

2020

2024