Machine Learning Based Thermal Management

Background:

CPUs and TPUs generate a lot of heat 

Traditional temperature sensors can not easily acquire temperature data at any given point of the cooling system. They can either:

•  measure temperature at a single point or

•  for the overall system 

Goals and Objectives:

• Develop a model for a TPU chip cooling system that takes advantage of AI vision by measuring various properties of bubbles in the flow of a water cooling system

• Train an AI model to recognize and interpret bubble/droplet dynamics

• Gain a better understanding of how AI works

• Learn how to use AI in real world applications

Sponsor

•  Yoonjin Won (won@uci.edu)