Dynamic Cache Tuning for Efficient Memory Based Computing in Multicore Architectures

Memory-based computing (MBC) is a promising approach to improve overall system reliability when few functional units are defective or unreliable under process-induced or thermal variations. A major challenge in using MBC for reliability improvement is that it can introduce significant energy and performance overhead. In this paper, we present an efficient dynamic cache reconfiguration and partitioning technique to improve performance and energy efficiency in MBC-enabled reliable multicore systems. We use genetic algorithm to search effectively in a large and complex design space. Experimental results demonstrate that the proposed cache reconfiguration and partitioning approach can significantly improve both performance and energy efficiency for on-demand memory based computing without sacrificing reliability.