Atomistic Growth Simulation and AI‑Driven Structure Generation of Amorphous Carbon

Amorphous monolayer carbon offers a unique platform to study disordered systems, whose growth pathways remain elusive. Starting from it, we present a three-part program linking atomistic growth, machine learning force fields, and inverse design across dimensions. (1) Using HyPASE—a bottom-up framework that couples a hybrid potential with fine-grid based Monte Carlo—we simulate carbon atomic manufacturing on Cu surfaces and this method could be applicable to other materials. Additionally, we uncover a chaining-branching-networking-filling (CBNF) mechanism, a unique growth process that might be overlooked in predefined commensurate lattice model. (2) We distill high value configurations and trajectories from HyPASE+DFT to fine-tune a MACE-based machine learning force field (MLFF) for carbon–metal systems, reproducing key energetics and kinetic barriers associated with CBNF while substantially accelerating sampling. (3) Guided by the CBNF priors and the MLFF, we launch a flow-based generative model over carbon networks conditioned on target structural statistic properties, yielding candidate amorphous–crystalline hybrid frameworks and testable routes for disorder modulation. Together these thrusts bridge atomic interactions at interfaces with bulk disorder architectures, provide hypotheses for amorphous–crystalline transitions and outline a path to physically grounded inverse design of disordered materials.