Current statusThe model runs synthetic deposition cycles and compares well-mixed and spatial predictions. Numerical checks are implemented. Experimental validation is still pending, and DEZ transport inputs are placeholders.
Model goal
I want to understand how much precursor reaches different parts of an ALD reactor, how surface reactions consume it, and how long the gas takes to clear after a pulse.
The well-mixed model treats the reactor as uniform. I compare it with a spatial model to check what that assumption misses along the channel.
My contribution
I built the Python model, coupled delivery and transport to surface-reaction kinetics, and added repeatable simulations and run comparisons. Desktop, command-line, and Python interfaces use the same solver. Saved runs retain inputs, results, and source hashes.
What is implemented
- Precursor delivery: a synthetic A-pulse, purge, B-pulse, purge sequence with explicit inlet doses.
- Transport: a well-mixed description and a spatial channel model, with gas balances and inlet/outlet accounting.
- Surface reactions: an effective two-event model that follows site termination and surface turnover across repeated cycles.
- Run comparison: precursor consumption and escape, surface completion, purge residuals, and variation along the channel.
The current examples are synthetic. The fictional A/B case reports surface turnover without assigning a film thickness or a QCM signal.
Well-mixed versus spatial predictions
In the documented synthetic example, the well-mixed model predicts 97.8% A completion, while the spatial model gives 91.1% at the channel outlet. The reactor average misses the lower completion at the outlet.
These are simulation outputs under stated assumptions, not measured deposition results or a validated operating recipe.
Numerical checks and current milestone
The workflow checks cycle recurrence, successive spatial grids, and tighter time-integration tolerances. It also checks precursor ledgers, event balances, and physical bounds. Numerical status and recipe acceptance are reported separately.
The public documentation reports a completed synthetic study of 63 recipe/scenario pairs. All calculations passed their numerical checks, but no recipe satisfied every scenario. Nine earlier high-Peclet cases and one capacity-profile point remain unverified.
These checks test the calculation, not its agreement with a physical reactor.
Open questions and next work
DEZ transport is the main unfinished input. Separate molecular-property work has not yet produced an accepted interaction energy or diffusivity for this model. I need sourced values and must recheck the affected calculations before replacing those placeholders.
Next steps are to resolve those inputs, complete physical-model checks, and compare predictions with published ZnO growth data using separate fitting and test conditions. TMA/H₂O remains the reference chemistry; ZnO comparisons would not validate TMA kinetics.
Experimental comparison, including QCM, is still planned. So far, the model is for offline simulation; it has not been experimentally validated or used to establish an optimized process.
Artifacts and documentation
The repository provides the code and example workflows. The input reference documents units, model assumptions, acceptance criteria, and the distinction between synthetic and physical inputs.
This working note reflects the public project documentation reviewed September 9, 2026.