aldenv

A collection of simple models and environments of atomic layer deposition designed to test the ability of machine learning and AI algorithms to autonomously optimize ALD processes.

aldenv in a nutshell

aldenv considers two different types of objects:

• Models are physics-based simulations of self-limited processes

• Environments incorporate some non-idealities found in real systems, such as noise or delays.

Motivation

aldenv was created to develop and benchmark machine learning algorithms and AI agents based on LLMs meant to be integrated with experimental tools.

The environments in aldenv can be viewed as virtual tools that provide realistic simulations of both the processes and tool interfaces. These virtual tools can also be used to generate benchmarks that can be used to evaluate models and agents and identify shortcomings of existing models.

For more information on how we use simulations in the context of autonomous materials synthesis you can check our works:

• Design and performance of AI agents interfacing with an atomic layer deposition tool
• Performance of AI agents based on reasoning language models on ALD process optimization tasks

Install

The easiest way is to use pip:

pip install aldenv

Funding acknowledgement

The work conducive to aldenv was funded as part of Argonne National Laboratory's Laboratory Directed Research and Development microelectronics portfolio.

Copyright and license

Copyright© 2026, UChicago Argonne, LLC

aldenv is distributed under the terms of BSD License.

Argonne Patent & Intellectual Property File Number: SF-26-055