We demonstrate a system-level CMP architecture where optical cache memories are shared among multiple processing cores through optical buses. System-level simulations show 25-45% execution time improvement and significant ...