High Dimensional Optimization and Inverse Methods for Electronic Design – Phase 2, CAEML Core Project

Our approach is to be able to solve large design problems by expanding on the capabilities of our work, which have been tested on problems of up to 24 dimensions at NC State University (NCSU).

This project seeks to extend Bayesian Optimization and Inverse Neural Networks with the capability to solve problems of over 40 dimensions. Random embeddings are used in several of these techniques, which will be demonstrated using analog design and Signal Integrity examples. Our approach expands on prior work tested at NC State, which was able to solve problems with up to 24 dimensions.


Principle Investigators

Paul D. Franzon
Brian Allan Floyd
Dror Zeev Baron

More Details

The objective of this project is to extend Bayesian optimization and Inverse Neural Networks design space generation to be able to handle higher numbers of dimensions. The goal is to solve problems with over 40 dimensions. A common feature in several of the techniques is the use of random embeddings. We will demonstrate these techniques using examples from analog design (Bayesian Optimization) and Signal Integrity (Inverse Neural Networks).