At the Cadence VIP dinner at Korea CDNLive last month, Paul Cunningham spoke about the state of the EDA industry, especially in the digital and signoff world.
It all comes down to the productivity gap between the design diversity and complexity rising faster than the design schedules and other limited resources. This is our fundamental challenge.
To overcome those challenges, designers must approach the challenge using a four-pronged approach:
- Using core innovation
- Leveraging trends in software
- Designing at a higher level of abstraction
- Streamlining tool interoperability
The lesson that stayed with me, though? Designing at a higher level of abstraction. This means Machine Learning, or as Lip-Bu says, “machinelearningdeeplearning”. In other words: if we can’t solve the problem—or solve it fast enough—then we need to train a neural network to predict the answer for us.
Machine learning opportunities in EDA cover the following four areas:
- Increasing coverage ramp in verification
- Predicting downstream flow effects
- Predicting where to focus optimization to improve PPA
- Turning batch-mode tasks into real-time interactive tasks
Also using Virtuoso EAD: Machine Learning-based models are created for integration with optimization methods; even offline learning and adaptation is in beta.
These innovations bring this gap to a more manageable end, to designing diverse and complex systems within a reasonable schedule and with limited resources.
