Design and Exploration of a General Purpose Stochastic Computing Processor

The rise of Internet of Things (IoT) with millions of edge devices created per year drives the need for computing platforms that have lower power, better fault tolerance and greater computing flexibility. With more and more novel applications and algorithms being rapidly deployed, the traditional approach of using custom hardware to achieve a low power budget will not scale. This project explores the design of an ultra low power, fault tolerant stochastic computing general purpose processor that satisfies the key metrics that IoT edge devices care about while being able to support the rapid shift in applications. Not only will the realization of a stochastic general purpose processor be able to maintain a low power envelope and be resilient through bit-flip errors, stochastic computing’s ability to provide progressive precision (i.e. compute results that start as a rough approximation and become more accurate over time) is often a desirable property for many of the applications in this space (e.g. machine learning).

Faculty Supervisor:

Jason Anderson

Student:

Partner:

Tokyo Institute of Technology

Discipline:

Engineering

Sector:

Education

University:

University of Toronto

Program: