Profiling of lossless-compression algorithms for a novel biomedical-implant architecture

Abstract

In view of a booming market for microelectronic implants, our ongoing research work is focusing on the specification and design of a novel biomedical microprocessor core targeting a large subset of existing and future biomedical applications. Towards this end, we have taken steps in identifying various tasks commonly required by such applications and profiling their behavior and requirements. A prominent family of such tasks is lossless data compression. In this work we profile a large collection of compression algorithms on suitably selected biomedical workloads. Compression ratio, average and peak power consumption, total energy budget, compression rate and program-code size metrics have been evaluated. Findings indicate the best-performing algorithms across most metrics to be mlzo (scores high in 5 out of 6 imposed metrics) and fin (present in 4 out of 6 metrics). Further mlzo profiling reveals the dominance of i) address-generation, load, branch and compare instructions, and ii) interdependent logical-logical and logical-compare instructions combinations.