Fixed-point arithmetic on a budget: Comparing probabilistic and reduced-precision addition

Probabilistic CMOS technology or PCMOS, is an ultra-low-power solution particularly well suited to multimedia applications. Novel results are presented here showing that PCMOS computes with less error and less power than reduced-bit-width, or truncated, deterministic arithmetic due to quantization noise injected by limited-precision, fxed-point processing. Simulation results, derived from physical layout of a PCMOS, fixed-point, ripple-carry, adder, are given showing energy savings of up to 2X can be achieved while surpassing the quantization performance of reduced-precision solutions.