ALD/ALE 2026 Session AM2-WeA: Digital Twins for ALD

The core challenge in ALD is that growth-per-cycle (GPC) emerges from an interplay between surface chemistry and operating conditions. On one hand, we look at ligand-exchange pathways, parasitic decomposition, nucleation behavior, and temperature-dependent kinetics. On the other, we need reactor geometry, substrate identity, temperature, dosing time and purge time. A high-fidelity digital twin for thermal atomic layer deposition (ALD) would thus not be limited to chemistry, but also incorporate process modelling. It would convert routine metrology and operating logs into predictive capability for process analysis and control, enabling faster recipe development, reduced precursor waste, and improved tool-to-tool reproducibility.

Entalpic is developing such AI-enabled digital twin of thermal ALD reactors, predicting macro-scale deposition outcomes while remaining anchored to mechanistic understanding. The model ingests precursor and co-reactant identity together with key process parameters (substrate, substrate temperature, and timing variables) and outputs GPC with calibrated uncertainty. To improve transferability and interpretability, we standardize experimental context across datasets, incorporate structured representations of reactor delivery behavior, and integrate mechanistically informed descriptors derived from reaction pathways and energetics.

The resulting digital twin supports two complementary capabilities. First, it enables recipe optimization within chemically consistent constraints, identifying conditions that achieve target growth while minimizing excess dosing, exposure time, and process margins. Second, as the dataset expands in chemical and operational diversity, it can generalize toward new precursors by connecting molecular-scale energetics and mechanistic signatures to reactor-scale GPC trends, while signaling when predictions fall outside the model’s reliable domain. Models are trained on curated literature deposition reports together with experimental datasets from partners, providing a continuously improving, metrology-aware foundation that links ALD process control with precursor discovery and down-selection, and that can be directly integrated with our internal AI pipeline for precursor discovery to prioritize candidates most likely to deliver robust growth and manufacturable process windows.