The Memory-Processor Gap Processor speed increases by 60% per year while memory access speed increase is 10% Most techniques focus on latency tolerance which exposes bandwidth limitations The most obvious solution is to reduce traffic to off-chip memory
IRAM Architecture
FlexRAM Design
DIVA Project
Active Pages
Experimental Plans On-chip DRAM cache FPGA-based chip Compatible with current architecture Allows bigger caches in same die area Reduces processor-memory gap problem by reducing traffic to off-chip memory FPGA-based chip Uses standard FPGA chips Allows good memory and logic locality Eliminates difference in fabrication problem Integrates memory and logic onto a single chip easily
Conclusions Memory-logic integration is possibly the best solution Much more research still needs to be done in this area There are several options that are relatively unexplored Research must continue in this area as the memory-processor gap widens every year