As the Fourth Industrial Revolution is approaching, large-scale computing is becoming more demanding and popular than ever. However, the performance of conventional electronic microprocessors has almost reached their limits for device speed, on-chip density and power consumption and will not be able to continue sustaining the upcoming data explosion. Optical computation can be extremely fast and with low-power requirements compared to electronics, for its intrinsic high speed, large bandwidth, and unlimited parallelism, which are critical to ease the data traffic associated with applications where artificial intelligence decisions need to be made in real time. Novel approaches towards programmable computations are required for data-driven training of modern artificial intelligence. In this project, the investigators will leverage the state-of-the-art integrated photonics technology to develop an innovative programmable photonic computation accelerators (PPCA), accelerating the computation speed and reducing the cost and energy consumption to sustain long term performance requirements for machine learning. This research is closely integrated with the existing educational activities, providing both undergraduate and graduate students with the opportunity to participate in cutting-edge science and technology in an innovative way. The investigators also provide educational outreach activities in integrated photonic devices, machine learning, and computer algorithms to promote the interests and participations of K-12 students and broaden the participations from underrepresented groups.
With funding from the Electrical, Communications and Cyber Systems (ECCS) Division, the investigators from the University of Pennsylvania and University of California, San Diego are developing a disruptive system-level integrated nanophotonic circuits ? Programmable Photonic Computation Accelerators (PPCA) ? through active control via strategic engineering of quantum symmetry, to perform real-time programmable mathematical operations and implement machine learning algorithms. Unique symmetry-driven geometries will be explored to deliver novel topological photonic components required for matrix multiplication, which can be dynamically programmed by flexible control of spatial-variant optical modulation. On the developed programmable photonic computation accelerator platform, different iconic machine learning algorithms will be performed to demonstrate optical machine learning for the first time and test its corresponding speed and fidelity. The investigators have highly complementary expertise on active photonic circuits, integrated devices, systems, and packaging, as well as computation and machine learning, which will be actively synergized, enabling a paradigmatic shift towards system-level integration of large-scale photonic computation accelerators. If successful, the innovative programmable photonic computation accelerators could be applied in the domains which demand extreme speed, energy efficiency, parallelism, significant complexity, and high scalability on an ultra-compact footprint, and full programmability.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
