Close

Presentation

Machine Learning Techniques for PDK Development Efficiency
TimeMonday, July 11th5pm - 6pm PDT
LocationLevel 2 Exhibit Hall
Event Type
Engineering Track Poster
Engineering Tracks
Topics
AI
Back-End Design
Cloud
Embedded Systems
Front-End Design
IP
DescriptionThe Process Design Kit (PDK) is a key component for IC design. As the very first interaction a customer design team has with chip manufacturers, it is crucial to first pass silicon success. The covid pandemic has created both a major disruption in the worldwide supply chain and a drastic acceleration of our world’s digitalization. These factors have created an unprecedented chip shortage and subsequently a huge demand for new PDKs.
Parameterized Cells (PCells) are a central part of the PDKs. However, their development is time consuming, and their high quality is very important to guarantee good time to market (by avoiding possible silicon re-spins or tape-out delays). To increase quality and standardization, we use our “best practice” document, but it contains more than 100 guidelines, and it can become challenging for developers to know which ones apply to their specific pcells. This is where our Machine Learning approach adds value: it helps the developer focus on the relevant best practices for a given pcell.
This presentation will discuss our research approach to developing this proof of concept, the hurdles we encountered and how we handled them as well as the current limits and our future steps for that tool.