Bayesian Optimization of a Scalable Coating Process Using a Self-Driving Laboratory

The commercialization of energy materials can take decades, in part due to the challenge of scaling-up laboratory synthesis techniques for manufacturing. For example, perovskite solar cells were discovered in 2009, but are not yet commercially available. Scaling-up the solution-based fabrication process is widely cited as the major reason for this. To address this challenge, we have designed and built a self-driving laboratory, “Ada”, for autonomously optimizing ultrasonic spray-coating, a scalable coating process. We used this self-driving laboratory to maximize the conductivity of a spray-combustion synthesized Pd coating by optimizing seven experimental variables under the control of a Bayesian optimization algorithm. This optimization yielded coatings with conductivities twice the previous state-of-the-art for spray-combustion synthesis. The best coatings deposited using our vacuum-free process exhibit conductivities comparable to vacuum-sputtered coatings and are directly scalable to larger area substrates with no loss in conductivity. This work shows how self-driving laboratories can contribute to accelerated commercialization of energy materials by rapidly optimizing scalable coating processes.