ML Based DRV Prediction and Optimization for DTCO
DescriptionDue to the limitations of intrinsic scaling of the process, it has become important to consider design side in process scaling. In these DTCO works, optimizing Design Rules (DR) to improve PPA requires huge resources. This is because as the number of DR parameters increases, the search space grows exponentially and the TAT becomes tight. To find a PPA-optimized DR, we propose ML-based platform that can search for a wider range of DRs in less time than before.
Our ML-based platform optimizes DTCO work with automated/parallelized systems and can conduct a wide range of DR multivariate analysis that engineers could not. With this platform, we performed Random Search for DR Sets using Boosting-based regression model. Through this, it was possible to check FI, and it was able to analyze and suggest the tendency of each DR. Using an optimized DR sets, we improved block area by 0.94% on average with an evaluation time 2.0x faster than the manual DTCO work at the SF3 process.
We think that our platform can replace or contribute to tasks that the engineer cannot do in DTCO work. In addition, We are planning to expand to a platform targeting Power and Performance.
Event Type
Back-End Design
TimeWednesday, July 12th3:45pm - 4:00pm PDT
Location2008, Level 2
Back-End Design