Machine Learning-Based Dynamic Voltage and Frequency Scaling Error Detection
Abstract
Modern microprocessors are more and more optimized for speed and power efficiency. A system that regulates these parameters is Dynamic Voltage and Frequency Scaling (DVFS). Generally, all bugs or errors in a microarchitecture fall under two categories: performance and logical. These errors can apply to any component of the microarchitecture. A performance error is an error that results not in a logically incorrect output of a system, but a slowdown in the production of that output. Most broadly, errors related to DVFS would be not increasing voltage and frequency leading to slower execution in real time, or the inverse (increasing voltage and frequency) leading to wasteful power consumption and chip degradation. The first slows down the machine unnecessarily, and the second decreases expected battery time. Using machine learning to analyze data extracted from gem5 to detect these errors is the objective of this project.
Citation
Muschinske, John C (2022). Machine Learning-Based Dynamic Voltage and Frequency Scaling Error Detection. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /196503.