Abstract:
Physical systems — including optical and microwave systems, be they classical or quantum — can be engineered to be programmable and then have their functionality set not by conventional human design but rather learned using gradient descent. Optimization trains the programmable degrees of freedom to achieve some desired behavior, defined either by an analytic description (inverse design) or training data (physical neural networks). Applications of trained programmable physical systems span computing, sensing, and communications. Fully translating the remarkable effectiveness of differentiable programming and gradient-based optimization in software to the physical world requires advances across the full stack from devices upwards, but has potential advantages in one or more of design effort, area, power, speed, accuracy, variation tolerance, scalability, and flexibility. In some cases, physical programmability can even enable new functionality that is otherwise unachievable. In this talk, McMahon will describe examples of his group’s work in two areas: (i) trainable photonic waveguides, a novel class of photonic device that we can use for computing, sensing, and communications; (ii) trainable superconducting systems for quantum computational sensing of microwave signals, combining quantum sensors with small-scale quantum processors. McMahon will explain how their group engineers devices for programmability and will introduce their procedure for backpropagation-based training of physical systems when there are only imperfect simulation models of them.
Biography:
Peter McMahon is an associate professor of applied and engineering physics at Cornell University. He received his doctoral degree in electrical engineering from Stanford University in 2014, where he also performed his postdoctoral training in applied physics until starting as a faculty member at Cornell in 2019. He is the recipient of the Packard and the Sloan Fellowships, an Office of Naval Research Young Investigator Program award, and the Lomb Medal from Optica.
