My research covers various topics. See all my research projects for more details.

Research Interests

My primary research interests are novel computing paradigms and systems for artificial intelligence, particularly with high performance and energy efficiency on edge devices and autonomous systems. My philosophy on next-generation computing of the post-CMOS era is "let the physics do the computing," namely employing dynamical systems for computing. Such an idea was derived by Allan Turing, a pioneer and a farseer of connectionism and brain-inspired computing. He also mentioned that nonlinear dynamical systems could be the key to the ultimate AI. 

During my PhD study, I spent most of my time on computing systems based on coupled oscillators, which are the essential elements of dynamical systems. Neurons, signals, and quantum spins can be interpreted as oscillators. I collaborated with various material scientists who are working on emerging nano-device technologies (e.g., spin-torque, vanadium oxide device, memristor) and modeled different types of oscillators, including soft hydrogels. We demonstrate how these nano-oscillator systems can perform pattern recognition, image processing, and accelerate machine learning. Back then the idea was not entirely popular until recent attention on the field of quantum computing. After I became a postdoc researcher, my job mostly focused on neuromorphic computing systems. I explored neuron dynamics imitated by ferroelectric devices and merged the paradigms of swarm intelligence and spiking neural networks for solving optimization problems. I lead a neuromorphic robot team, which explores end-to-end brain-inspired event computing systems that learn robotic locomotion in real time. If you are interested in these topics, you are welcome to visit my project page.

There is much more to be explored about "computing with dynamical systems". 

• Energy-Efficient Edge Intelligence. Energy-efficient neuromorphic computing for autonomous systems and biomedical devices. Recently our team is trying to incorporate Dynamic Vision Sensors with a spiking neural network to create an end-to-end event-based computing system. 

• Neuro-symbolic AI. 

• Neuromorphic Swarms. 

• Hybrid smart materials that compute for Soft Robot and Wearable Medical Devices. (See below)

Here I briefly introduce my research with soft materials, which is the essential part of my PhD desertation, but somewhat deviate from electrical and computing engineering. 

In the later stage of my Ph.D., I began to work on "Materials that Compute", an interdisciplinary field that merges research in computer science and material science. The main goal is to design hybrid soft material systems that incorporate multiple functions of sensing, actuating and computing.

Recent scientists and engineers have begun to explore the design and fabrication of soft-bodied robots composed of compliant materials. Inspired by the biological systems, these machines have deformable structures and multi-functions like sensing, actuation, communication. Compared to traditional rigid robots, their adaptation, sensitivity, and agility enable friendly interaction with the natural and biological environment, providing a better opportunity to bridge the gap between machines and people. However, one of the main challenges for creating these soft intelligent machines is to design adaptive, flexible and stretchable materials with hybrid complex functions. 

We believe “materials that compute” is a promising method to address this challenge. We want to design systems where the material and the computers are one and the same entity. For sensing and actuation, these systems should be able to generate mechanical force in a relatively self-sustained autonomous manner and be sensitive to the external environment. Meanwhile, in terms of communication and computation, the desired material system could process gathered information in a de-centralized and unconventional way, which harness the intrinsic properties of the materials.

The current design of hybrid material is capable of storing and retrieving patterns while performing complex tasks in learning and vision still remain difficult. I am looking at incorporating more features and inspirations from neural systems and biological tissue.