Comparing Stochastic and Deterministic Computing

摘要: Technology scaling has raised the specter of myriads cheap, but unreliable and/or stochastic devices that must be creatively combined to create a reliable computing system. This renewed interest in exploits stochasticity—embracing, not combating device physics. If representation is used implement programmable general-purpose architecture akin CPUs, GPUs, or FPGAs, preponderance evidence indicates most system energy will expended communication and storage as opposed computation. paper presents an analytical treatment benefits drawbacks adopting approach by examining cost representing value. We show both laws costs for low precision representations. also analyze multiplication implemented using versus deterministic approaches, since prototypical inexpensive operation. compares favorably when holding reliability constant.