Real Time On-Die Power Thermal Profiling for Machine Learning Design Applications

摘要

Many Advanced Computing System-on-Chip (SOC) implement special computing elements such as Artificial Intelligence Engine (AIE) to support AI Machine Learning (ML) Applications. Unlike traditional computing architecture, machine learning usages such as data flow with different algorithms are still evolving. Therefore, the power and thermal conditions of the machine learning computing SOC is not clearly understood. Also, the voltage and temperature conditions strongly depend on the end user specific designs.