Issue link: https://iconnect007.uberflip.com/i/1028393
20 DESIGN007 MAGAZINE I SEPTEMBER 2018 2009-2010 timeframe, with two persons and myself. Our motivation came from observing the scale and complexity that grew with the increase in data such as larger designs, larger simulations, etc. To address these problems, we began to look at data-driven solutions such as analytics and machine learning. When we began the work, there was not the same buzz around machine or deep learning, and we just found it to be a useful tool to create fast models of complex non-linear problems that required long com- pute times using more traditional methods. Two machine learning-based solutions were released in 2013 in the Virtuoso Electrically Aware Design (EAD) environment and Virtu- oso Analog Design Environment (ADE). Shaughnessy: Tell us about Cadence's current research into AI. Are you continuously work- ing to implement more AI in EDA tools? White: As you can understand, I can't go into details on research and development that has not been released in our products. First, we are fortunate to have a brilliant team of engineers who continually push the edge. Our machine learning team includes members of our Virtuoso technology team, which focuses on placement, routing, and analog design and electrically aware design as well as our OrbitIO package and board solu- tion—so they have hands-on experience with the applications we are trying to automate. We are constantly evaluating data-driven solutions such as machine and deep learn- ing, as well as analytics, optimization and distributed processing. You always hear about machine and deep learning, but those other solutions are often required as well. Optimiza- tion is often overlooked in discussions about building intelligent, adaptive systems. From a decision systems perspective, you want to use optimization in conjunction with ML/DL to drive the system or decision sequence to some desired state (e.g., a placement and rout- ing alternative that meets design intent). One solution is no more important than the other. The biggest challenge in productizing AI is addressing the verification, deployment and support issues that every new technology faces. Most of the available open source software focuses on the creation of ML or DL models and not as much on how to verify, calibrate or adapt those models in environments that change. However, those factors are critical in CAD/EDA tools where design intent may not be fully observable or the environment changes Figure 1: A graphic showing the development of machine learning and EDA over the years.