Automating Reactive Force Field Fitting to Model Bi₂Se₃/Nb Heterostructure Synthesis Conditions

Topological materials with time-reversal symmetry-protected surface states hold promise for applications in spintronics, quantum computing (qubits), and thermoelectrics. When interfaced with a niobium (Nb) superconductor, the van der Waals-layered topological insulator bismuth selenide (Bi₂Se₃) can form a topological superconductor, which is of great interest for fault-tolerant quantum computing due to its ability to host Majorana particles. However, achieving a high-quality, defect-free Bi₂Se₃/Nb heterojunction using molecular beam epitaxy (MBE) requires fine-tuning the synthesis conditions (e.g., temperature, pressure, and the chemical potentials of Bi, Se, and Nb), which span a vast and complex parameter space.

This talk details progress towards building autonomous workflows that generate DFT data and fit reactive force fields to guide 2D heterostructure synthesis. Our approach uses ensembles of Jax-ReaxFF fitted force fields (FFs) and uncertainty quantification (UQ) to efficiently converge FF parameters; FFs derived from our approach recapture the relevant Bi-Se and Nb-Se experimental phases during melt/quench recrystallization runs and can therefore help select the conditions for 2D heterojunction synthesis. We anticipate that the flexibility of these autonomous approaches will allow them to generalize to other compositional systems with low-dimensional topological materials.