Inexact computing using probabilistic circuits

Abstract

Numerous computing applications can tolerate low error rates. In such applications, inexact approaches provide the ability to achieve significantly lower power. This work demonstrates the power-error trade-offs that can be achieved. Using probabilistic modeling in sub-50-nm silicon transistor technology, the relationship between statistical uncertainties and errors are elucidated for different configurations and topologies and the trade-offs quantified. Gate-level implementation of the probabilistic CMOS logic is validated by circuit simulations of a commercial 45-nm SOI CMOS process technology. Using a practical ALU architecture where voltages can be scaled from most significant to least significant bit blocks as an example, the potential benefits of this technique are shown. A calculation error of 10 −6 , an error rate quite tolerable for many computational tasks, is shown to be possible with a total power reduction of more than 40%.