REGISTER | LOGIN
Breaking News
News & Analysis

Machine Learning Comes to Chip Design

Solido to launch Machine Learning Labs
4/6/2017 09:00 AM EDT
3 comments
NO RATINGS
Page 1 / 3 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
HankWalker
User Rank
Author
Re: Misleading Title
HankWalker   4/8/2017 1:48:45 PM
NO RATINGS
New approaches such as BDD, SAT solvers, simulated annealing, spectral partitioning, genetic search, etc. became available, and proved more efficient on EDA problems. One of the challenges in neural networks, now, as then, was the need to train them. Similarly, expert systems development required the iteration between rule development and experimentation to include all of the important rules. One non-EDA system I helped with had 2500 rules. In the end these are all algorithms, but some of them have gone through a hype cycle.

junko.yoshida
User Rank
Author
Re: Misleading Title
junko.yoshida   4/8/2017 7:48:11 AM
NO RATINGS
Thanks, HankWalker. Much appreciate your perspective here. You wrote: Many papers were published on using neural networks for a variety of EDA problems. Then it all died down. Why did it all die down?

HankWalker
User Rank
Author
Misleading Title
HankWalker   4/7/2017 12:34:23 PM
NO RATINGS
The title of the article implies that machine learning in chip design is a new idea. There was a huge bubble of using AI-based techniques for electronic design automation in the late 1980s. Companies such as Trimeter Technologies sold expert-system based tools for logic synthesis. Many papers were published on using neural networks for a variety of EDA problems. Then it all died down. Meanwhile, machine learning based outlier analysis techniques have been widely used in IC testing for many years. I am not knocking Solido, just the idea that machine learning in EDA and test is new.

 

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed