Optimizing Inkless Printed Electronics with Machine Learning Models

Inkless dry additive nanomanufacturing enables ink-free printing of electronics across diverse feedstock and substrates, but the high versatility and coupled parameter space can make tuning of print parameters a time-intensive, manual process. We demonstrate a fully autonomous printing capability that closes the loop between printing, measurement, and model-based decision making to drive the process toward a user-specified target resistance with minimal operator intervention. An active learning algorithm trade study was used to select modeling and optimization choices (surrogate model, acquisition strategy, and batch sizes) that efficiently captures our printing process while balancing convergence speed with print constraints. Using the chosen active learning model, the autonomous system iteratively proposes new parameter batches, executes prints, makes measurements, and updates its process model until the target is achieved. In validation on both silver and copper, the automated workflow has proven itself capable of mapping a four-dimensional print parameter space while achieving a targeted resistance in less than 30 printed lines and yielding a transferable parameter space map to utilize for future targets. Overall, this work establishes self-driving inkless printed electronics where resistance targeting and process learning occur in situ.