CoolPression - a hybrid significance compression technique for reducing energy in caches

This paper describes CoolPression, a hybrid hardware-based significance compression technique for reducing energy in caches. The scheme exploits data compression opportunities at bit granularity by employing two compression schemes: a significance compression scheme that counts leading ones and zeros, and a dynamic zero compression technique. Based on actual data, the more energy-efficient scheme is selected dynamically to minimize bitline drives needed for each cache access. Using SPECint2000 benchmark, our experiments show that the CoolPression improves dynamic energy consumption by more than 35% against a baseline cache, while having a saving ranged from 5 to 15% compared to each scheme applied alone.