Resistive Computation: A Critique

Resistive Computation was suggested by [6] as an idea for tacking the power wall by replacing conventional CMOS logic with Magnetic Tunnel Junction (MTJ) based Look-Up Tables (LUTs). Spin Transfer Torque RAM (STTRAM) is an emerging CMOS-compatible non-volatile memory technology based on Magnetic Tunnel Junctions as a memory bit [3]. The principal advantage of STTRAM is that it is leakage-resistant, which is an important characteristic beyond the 45nm technology node, where leakage concerns are becoming a limiting factor in microprocessor performance. Although STTRAM is a good candidate for replacing SRAM for on-chip memory, we argue in this article MTJ-based LUTs are unnecessarily expensive in terms of area, power, and performance when implementing fixed combinational logic that does not require the reprogramming ability provided by MTJs.