METRICS 2.0: A machine-learning based optimization system for IC design

Despite advancements in quality of results of design automation tools, many challenges remain to be solved. In particular, there is lack of a unified and standardized system for collecting, storing, and communicating design process metrics.

The lack of such system hampers the use of machine learning techniques, especially across the entire design flow. In this work, we present our current thinking toward a “METRICS 2.0” system for systematic data collection and machine learning in design automation flows.

Inspired by the METRICS infrastructure, published nearly twenty years ago, we re-imagine a unified system better suited for recent developments in databases and machine learning. Our proposed system spans a wide range of components in integrated circuit design and offers (i) an updated dictionary of generic, tool and flow metrics,(ii) a standard modernized data storage approach that allows for easier data sharing, and (iii) enablement for design process outcome prediction and debugging through machine learning techniques.