- This event has passed.
Active Learning for Fast, Comprehensive SPICE Verification
July 8 @ 8:00 AM - 9:00 AM
The scope of SPICE-level verification has increased massively with new requirements for safety critical applications, statistical timing characterization, wider FinFET voltage domains, and tighter product margins. We now have many more PVT corners to verify against, and many types of IP need to be verified to high sigma, requiring millions or billions of Monte Carlo samples. Simulation budgets have ballooned and brute-force simulation methods no longer deliver the coverage required within production runtime constraints.
For the past 14 years, Solido (now part of Mentor, a Siemens Business) has been using active learning technologies to accelerate SPICE verification by 10X to 1MX, while maintaining SPICE-level accuracy. This talk reviews Solido’s active learning techniques used within tools. It provides a deeper dive into the all-new High-Sigma Verifier technology, as well as provides production usage updates from ML Characterization Suite Analytics and Generator.
What You Will Learn
You will become familiar with the methods Solido has used to improve speed, accuracy, and coverage of SPICE batches by orders of magnitude. You will understand where they apply, how they work, and why they work. You will learn how Solido products can speed up your PVT, Monte Carlo, and characterization by orders of magnitude.
Who Should Attend
Anyone who runs (or enables those who run) a lot of PVT, Monte Carlo, or high-sigma Monte Carlo analysis and would like more speed or coverage. Anyone who does cell characterization and wants to decrease CPU, engineering, and EDA license resources, to deliver on a faster schedule, and to make better quality libraries.
- The SPICE verification explosion
- Common challenges with faster algorithmic approaches
- How the Solido active learning solution works and why it is better
- Active learning for speeding up PVT, 3-sigma Monte Carlo, and high-sigma Monte Carlo
- Active learning for speeding up cell characterization and .lib verification