Unified library characterisation tool leverages machine learning in the cloud
Employing advanced machine learning techniques, this tool uses smart interpolation to help determine the critical corners that need to be characterized. By making characterization processes thoroughly distributed and massively parallel, it has been fully optimized for running on cloud-based servers.
The new tool offers up to 3X performance increase by running corners in parallel and natively running statistical and nominal characterization together, claims the company.
“Characterization is an extremely time-consuming activity with increasing corners, larger libraries, and new data formats,” said Ron Moore, vice president of business planning, Physical Design Group, Arm. “By using Arm’s Artisan Physical IP, we validated Cadence’s Liberate Trio Characterization Suite and saw a notable improvement in turnaround time using the same number of CPUs. This is an important step in continuing to deliver high-performance libraries to our mutual customers.”
The machine learning algorithms in the Liberate Trio suite help guide the designer, predicting critical corners and helping designers decide what corners need to be characterized. The tool uses smart interpolation, not just linear interpolation, to ensure accuracy. Cadence created a unified graphical user interface (GUI) cockpit that lets designers use a single script to efficiently launch and monitor characterization. This addresses the challenge of maintaining consistency of data across the large number of process, voltage and temperature (PVT) corners. It also helps designers properly mine all the data that is collected.
The Liberate Trio suite has been optimized to be fully cloud ready, whether employed in a public or company private cloud, and scalable to over a thousand CPUs. Characterization of a library containing over 1000 cells that would normally take weeks now can be turned around in days.
Cadence – www.cadence.com