Optimizing Bandwidth Utilization Through Word Based Compression in Main Memories
N Aswathy, Harsh Verma, Hemangee K. Kapoor · 2025
Memory compression maximizes memory capacity, lowers bandwidth requirements, and reduces energy consumption. State-of-the-art compression techniques mostly focus on enhancing memory capacity, which involves complex reasoning for memory-based data retrieval. In this paper, we propose conditional frequent word compression (Cond-FWC) to reduce bandwidth requirements and energy consumption. The proposed Cond-FWC, as its name implies, eliminates the words that appear the most frequently from the cache line to compress the block. Furthermore, the method accurately places compressed data and metadata for compression with the help of an existing rank subsetting policy. Cond-FWC yields a 16.32% performance improvement and 55% energy reduction compared to an uncompressed non-sub-ranked baseline through rank subsetting and frequent word compression.