The chip verification industry could adopt technique from the software business to improve productivity, said a leading engineer in a keynote to the DVcon Europe conference in Munich
Techniques such as machine learning, agile processes and languages such as Python could be incorporated more effectively alongside cloud technologies, says Moshe Zalcberg, CEO of Veriest (above).
“The software guys are not imagining how to use AI, it’s happening,” he said. “Microsoft and Facebook are using AI-aware tools for catching bugs and suggesting snippets of code, searching for security vulnerabilities. These are already on [software repository] GitHub.”
He points to releases in the last year from Cadence Design Systems and Synopsys on AI, as well as Intel using machine learning techniques for ‘intent-driven validation’ and startups such as VerifAI but says there is a longway to go in verification.
He also points to the need to adopt native cloud technologies for scaling and to provide more data to improve the verification and validation process, but there are still issues with EDA licenses on the cloud. “We are still not there on the EDA license scheme,” he said.
“Besides the IT resources there’s all the different data we have in verification, there’s a lot of things to capture and work on,” he said. “We already do descriptive and diagnostic analytics but beyond that its less common to do predictive analytics and prescriptive analytics. Today it tends to be based on experience and gut feeling when we have the data on the clr enable us to analyse the data in a different way.”
Then there are more graduates skilled in newer languages such as Python that could be used alongside SystemVerilog and e for building frameworks. “For the first time we see people are using Python for verification and the fact that this is part of the design means the industry is looking for other solutions.”