algofoogle (Anton Maurovic)
Folks, we have a select group of people here in this wafer.space community who might be interested in this, and @Tim 'mithro' Ansell maybe this would be of interest to you for various reasons in your current explorations...Will Salcedo, who worked with us on customer design support for a while at Efabless, is developing GPU & AI acceleration for OpenLane. This is a blurb about their work:
Will Salcedo (https://www.linkedin.com/in/william-salcedo/) and Vamshi Balangala (https://www.linkedin.com/in/vamshibalanaga/) are building enhanced EDA tools on top of OpenLane that improve the speed, scalability, and quality of existing open-source tools. We are using GPU-accelerated approaches and AI to get at least 100x runtime improvements over legacy tools. If you are struggling with closing out designs, we would love to help you get to tapeout faster.
Will and Vamshi are hoping to find people interested in what they're doing, and possibly willing to experiment with what they've done so far, so if anyone would be prepared to at least reach out to them and find out more, please let me know. From what Will was telling me, their approach was to greatly accelerate parts of the resizing/repair, with the intent initially of doing far greater design-space exploration (since for larger designs or extreme timing/density requirements, this can often be a bottleneck). They have also contemplated adapting their work to things like IR drop analysis.
It sounds interesting (and a few of us at Efabless started talking to them late 2024 and it was already showing promise then), and I'm just sharing this around as a favour. I would like to know more about this!
