Quantum Materials in the Age of AI: From 2D Heterostructures to Agentic Discovery

Understanding and controlling quantum phases of matter is one of the central challenges of condensed matter physics. In this talk, I will describe how our group uses this challenge as a unifying thread to connect experiment, theory, and artificial intelligence, with topological van der Waals materials as our primary arena.

I will begin with the experimental foundation. Two-dimensional van der Waals heterostructures, assembled by stacking crystals without the constraints of conventional epitaxy, are an ideal arena for isolating and tuning quantum phenomena by design. I will highlight isotope-controlled topological switching in HfTe5, where substituting isotopes alters the quantum zero-point lattice fluctuations sufficiently to drive the material across a topological phase boundary. This work, in collaboration with Luis Jauregui, Joel Moore, and a broad team of theorists and experimentalists, demonstrates that isotopic substitution can serve as a clean, disorder-free knob for exploring topological phase diagrams, and it underscores how tightly coupled experiment and theory must be to interpret such subtle quantum effects.

This brings me to the question motivating the second part of the talk: given how difficult it is to find and characterize materials near topological phase boundaries, how do we do it more systematically? I will introduce LEAP (Low-Energy, AI-Informed Phase Transitions), a University of California AI Science at Scale initiative that coordinates theory, experiment, and simulation in an AI-accelerated loop targeting topological materials for energy-efficient electronics. Within this framework, I will present Struct-R1, an agentic system that navigates the materials structure space by coordinating measurements of varying fidelity within physics-grounded reasoning loops, and LEAP-Bench (Chen, Liu, Jeong et al., ICLR 2027, under review), a benchmark that evaluates whether AI agents can reason from raw experimental data to defensible scientific conclusions, a capability essential for deploying AI co-scientists in real laboratory settings.

Speaker: Dr. Archana Raja is a Staff Scientist at the Molecular Foundry, and a faculty member at the Kavli Energy NanoScience Institute, Lawrence Berkeley National Laboratory.