Non-volatile Content Addressable Memory for Computing Acceleration
Abstract: We propose a resistive content addressable memory (CAM) accelerator, called RCA, which exploits data locality to have an approximate memory-based computation. RCA stores frequent patterns and performs computation inside CAM without using processing cores. During execution time, RCA searches an input operand among values on a CAM and returns the closest row. RCA can accelerate CPU computation by 12.6x and improve the energy efficiency by 6.6x as compared to a traditional CPU, with minimal error.
Bio: Daniel is a Masters student at UCSD studying Computer Science and Engineering. He is a member of the System Energy Efficiency Lab (SEElab) focused applications of nonvolatile memory and approximate computing.