Abstract

Deep reinforcement learning can generate complex control policies, but requires large amounts of training data to work effectively. Recent work has attempted to address this issue by leveraging differentiable simulators. However, inherent problems such as local minima and exploding/vanishing numerical gradients prevent these methods from being generally applied to control tasks with complex contact-rich dynamics, such as humanoid locomotion in classical RL benchmarks.

Machine learning (ML) has been applied to many EDA problems in recent years. We can classify these applications into three major categories: Predictor, Optimizer and Generator based on the role of ML played in these applications and the ML techniques used. Ideally one would like to adopt the Optimizer and Generator approaches to solve a hard EDA problem directly with ML, and we call these ML-alone approach. It is very challenging, however, to scale the ML-alone approach to solve real world EDA problems.

High-throughput RTL simulation is critical for verifying today’s highly complex SoCs. Recent research has explored accelerating RTL simulation by leveraging event-driven approaches or partitioning heuristics to speed up simulation on a single stimulus.

In this paper, we present the first Power, Performance, and Area (PPA)-directed, end-to-end placement optimization framework that provides cell clustering constraints as placement guidance to advance commercial placers. Specifically, we formulate PPA metrics as Machine Learning (ML) loss functions, and use graph clustering techniques to optimize them by improving clustering assignments.