Close

Presentation

PPATuner: Pareto-driven Tool Parameter Auto-tuning in Physical Design via Gaussian Process Transfer Learning
TimeThursday, July 14th5:10pm - 5:30pm PDT
Location3007, Level 3
Event Type
Research Manuscript
Keywords
Timing and Low Power Design
Topics
EDA
DescriptionNowadays, design complexity makes the VLSI design flow increasingly rely on EDA tools. Meanwhile, IC designers are suffering the pressure from the time-to-market.
For a new design, oceans of attempts to navigate high QoR will be made via multiple tool runs with numerous combinations of tool parameters.
Besides, designers often puzzle over simultaneously considering multiple objectives.
To tackle the dilemma, we propose a Pareto-driven physical design tool parameter tuning methodology with incorporating the adaptive transfer Gaussian process model.
It can learn the transfer knowledge from the existing tool parameter combinations.
The experimental results demonstrate the merits of our framework.