Qian Yu
University of California-Santa Barbara, CA·Engineering
Partial differential equations (PDEs) describe how quantities change across space and time. They underpin aircraft design, semiconductor chip design, medical imaging, autonomous vehicles and weather prediction.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
physics-informed neural networks
partial differential equations
PDE solvers
- Grants
- 1 known grant, through Sep 2027
What they work on
- Develop a highly compressed training method for physics-informed neural networks (PINNs) that uses only forward passes and no backpropagation, so it can run on photonic chips without photonic memory.
- Build a theoretical understanding of the forward-only, compressed training method that provides performance guarantees when training realistic PINNs of about 1,000 neurons.
- Pair the training algorithms with Hewlett Packard Labs' resonator-based photonic tensor cores and charge-trap flash memory. These process many wavelengths of light in parallel and compute inside memory.
- Demonstrate end-to-end, real-time PINN training on a 2.5-dimensional photonic chip that co-packages electronic and photonic integrated circuits.
More about the research
Traditional solvers chop space and time into many grid points, which costs huge amounts of memory and computing. Physics-informed neural networks (PINNs) are neural networks trained to obey known physical laws. They outperform grid-based solvers on high-dimensional PDEs and on control problems constrained by PDEs. But training them is slow even on powerful graphics processing units. That keeps PINNs out of settings with strict limits on a computer's size, weight and power. This project is a collaboration between UC Santa Barbara and Hewlett Packard Labs. It aims to build the first real-time, realistic-size PINN training accelerator on photonic chips, which compute with light.
Why it matters
Fast, low-energy PDE solving could help in areas such as aircraft and chip design, medical imaging, autonomous vehicles and weather prediction. It would matter most where computers must be small, light and power-efficient. A photonic PINN training accelerator could let such problems be solved in real time at very low energy cost. The partnership with Hewlett Packard Labs aims to transfer the technology to industry.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Omiya Hassan
Boise State University, ID·Engineering
Co-design an electro-optical application-specific integrated circuit that pairs photonic (light-based) and electronic processing units, assigning AI training and inference to the best-suited unit.
Jiaqi Gu
Arizona State University Campus Immersion, AZ·Engineering·Early-career
Develop a learned electromagnetic solver, a machine-learning model of how light moves through photonic devices, and a circuit-level co-simulator for fast simulation of mixed electronic-photonic chips.
Sarah Swisher
University of Minnesota, MN·Engineering·Early-career
Create a vertical ferroelectric memcapacitor, a capacitor that stores memory states, that can be fabricated in the back end of the CMOS process for monolithic 3D chip integration.
The school
University of California-Santa Barbara: 17 labs on PI Pages, $303M a year on research, 403 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.