Smarter electronics design through machine learning

Machine learning experts at Carnegie Mellon University are collaborating with Cadence Design Systems, Inc. and NVIDIA on a Defense Advanced Research Projects Agency (DARPA) sponsored project to automate the design process of electronic circuits and chips.

Burak Kara, a professor of mechanical engineering and Barnabas Poczos, associate professor of machine learning, will apply advanced machine learning techniques to develop integrated and intelligent design system flows. These flows, or processes, will optimize how systems on chip (SoCs), systems in package (SiPs), and printed circuit boards (PCBs) are made to improve their performance.

One machine learning technique that the team will use is deep reinforcement learning, which creates and uses efficient algorithms to solve complex problems in a series of steps. With the computational heft to test and reject millions of possible actions, the machine or network is able to learn which combinations work the best while receiving positive or negative reinforcement, much like a puppy learning to sit for a treat. This artificial intelligence tool can be applied to a range of areas from healthcare to robotics—and electronics design.

“The design of circuit layouts and routing wires between the components is an extremely challenging problem, and impossible for humans to do without computers, said Kara. "Using advanced machine learning approaches, we are demonstrating that a computer can learn how to best accomplish these tasks. Our current results are quite promising in that they can already outperform conventional design automation approaches commonly used in this space.”

Using advanced machine learning approaches, we are demonstrating that a computer can learn how to best accomplish these tasks.

Burak Kara, Professor, Department of Mechanical Engineering, Carnegie Mellon University

The collaboration is through Cadence’s Machine learning-driven Automatic Generation of Electronic Systems Through Intelligent Collaboration (MAGESTIC) research and development program. When the company was selected by DARPA to develop more automated design capabilities for electronics in 2018, it created MAGESTIC and sought out machine learning leaders from Carnegie Mellon and NVIDIA to round out their team.

“Our collaboration with Cadence and NVIDIA has been very exciting,” said Kara. “We are collectively working on the development of next generation electronic design automation technologies, with unique insights into the challenging problems.”

Media contact:
Lisa Kulick, lkulick@andrew.cmu.edu