Utilizing the Cloud to Increase Library Characterization Throughput and Reduce Schedule Bottlenecks
TimeMonday, December 6th4:06pm - 4:24pm PST
Event Type
Designer, IP and Embedded Systems Track Presentations
Virtual Programs
Presented In-Person
DescriptionToday, SPICE-based characterization of Liberty (.lib) models for modern standard cell libraries involves running hundreds of millions of SPICE simulations, with the amount of simulations needed increasing exponentially over the past few years, due to added quantity and complexity of data in .lib models.

Due to the high amount of computation required, library characterization teams routinely utilize hundreds to thousands of CPUs during each characterization run, and require days or weeks of schedule time to complete each characterization iteration for a given set of .libs. This presents a significant challenge for library teams and IT infrastructure teams trying to meet design schedules while balancing computational hardware costs.

In this paper, we present the steps taken and the choices made in setting up a cloud environment for characterization workloads, and quantify scaling efficiency results from a validated testcase. Overall, we find that with the correct configuration, cloud is a valid method providing “burst” compute power to accelerate high priority workloads, while being able to off-load on-premises tasks when needed.