Process-aware design profiling with machine learning based Calibre SONR

In the past, design rule checks alone could be used to predict and prevent systematic hotspots in IC layouts. Later DRC was supplemented by optical rule checks and layout rule checks using trained lithography and etch models, which reliably found most systematic hotspots.

For today's designs on advanced manufacturing nodes, new ways of detecting hotspots and suppressing nuisance defects are needed. Leading-edge fabs face challenges and are turning to machine learning to solve them. We introduce a feature-vector-based machine learning tool, Calibre SONR from Siemens EDA, to predict hotspots on different process stages and cross products based on known hotspots.

This paper, originally given as a presentation at the 2020 Design Automation Conference, describes Calibre SONR, a machine-learning based software that uses feature vectors to boost the productivity and accuracy of fab defect detection and diagnosis.