Digital twins for synthesis: Frameworks to computationally accelerate and control pathways of growth for 2D materials

Accelerating conventional synthesis campaigns in van der Waals (vdW) heterostructures, is an important aspect to discover novel pathways which can harness unprecedented functionalities in 2D materials. Recent surge in efforts towards prototyping autonomous synthesis and characterization frameworks through AI/ML is expediting such discoveries. However, narrowing down the optimal thermodynamic and kinetic parameter space solely through experiments is a nontrivial automation challenge, especially when gaining finer control over tuning an emerging functionality on demand is the ultimate goal.

In this talk, we introduce a simulation-driven approach empowered by extremely scalable real-time Bayesian sampling workflows that couple automated high-throughput ab initio and large-scale classical atomistic simulations integrated with on-the-fly simulated diffraction pattern analysis via a recently developed extreme-scale autonomous simulation platform MatEnsemble. By means of autonomously tracking and controlling nanoscale dynamics dictated by on-demand synthesis signatures either in real (i.e. correlation functions) or reciprocal spaces (e.g.  electron/X-ray diffraction), I will show how we can computationally discover and optimally control phase transformation pathways for crystallization of amorphous transition metal dichalcogenides, in synergy with parallel experimental synthesis and characterization campaigns.