Research

Memristor + X

The memristor, experimentally demonstrated in 2008, is a programmable passive electronic device that naturally stores and processes information through its programmable resistance. As conventional CMOS circuits continue to scale, active components increasingly suffer from limitations in power consumption, heat dissipation, and area overhead. Memristors offer a compact and energy-efficient alternative by performing memory and computation within the same physical device (=in-memory computing). 

We integrate memristors with sensors, circuits, algorithms, and optoelectronic systems to build next-generation neuromorphic hardware. Rather than treating the memristor as a standalone device, we investigate how it can unlock entirely new analog computing architectures. 

Physical nonlinear dynamics

Nonlinear systems exhibit diverse behaviors from only tiny differences in their initial conditions, like the well-known double pendulum. Although these dynamics appear irregular, they are not random; they evolve within bounded trajectories known as chaotic attractors. Interestingly, biological neural networks also operate in a local activity, referred to as edge of chaos, enabling efficient information processing and adaptability. We investigate nonlinear dynamics, chaos, and bifurcation in electronic devices to understand and harness these physics for neuromorphic computing. 

AI for nanotechnology

Developing high-performance memristors still relies heavily on manual fabrication and characterization, making the process both time-consuming and inefficient. We envision physical AI learning directly from experimental data to accelerate device discovery and optimization. By closing the loop between fabrication, characterization, and AI, we aim to identify optimal material structures, improve memristive robustness, and establish high-precision programming protocols.