Machine Learning at the Edge Using Spintronic Stochastic Computing
Abstract
Stochastic computing (SC) has seen a renaissance in recent years as a means for machine learning acceleration duet o its compact arithmetic and approximation properties. Still, SC accuracy remains an issue, with prior works either not fully utilizing the computational density or suffering from significant accuracy losses. In this work, we propose a set of optimizations targeting stream generation and execution components of SC, that bridges the accuracy gap between stochastic computing and fixed-point neural networks. Our techniques improve accuracy by coupling controlled stream sharing with training and balancing OR and binary accumulations. They further optimizes the SC execution through progressive shadow buffering, spintronic nonvolatile memory and architectural optimizations. We show that they can improve accuracy compared to state-of-the-art SC by 2.2-4.0% points while being up to 4.4X faster and 6.8X more energy efficient. Our optimizations eliminate the accuracy gap between SC and fixed-point architectures while delivering up to 5.6X higher throughput and 2.9X lower energy.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 29, 2021
- Accession Number
- AD1136388
Entities
People
- Albert Lee
- Di Wu
- Jiyue Yang
- Kang L. Wang
- Puneet Gupta
- Sudhakar Pamarti
- Tianmu Li
- Wojciech Romaszkan
Organizations
- University of California, Los Angeles