Towards an Autonomous Combinatorial Co-Sputtering Reactor

Computational databases can predict novel and promising materials, but those that can be synthesized and subjected to characterization face the challenge of reproducibility. For example, tin-based oxynitride ferroelectric perovskites with band gap in visible, or zinc-based oxynitrides wurtzites with perfect short-range order, have been computationally proposed. However, the oxygen to nitrogen ratio required for these promising properties, such as long-term stability and semiconductor-like charge transport, is difficult to achieve and reproduce.<br/>To answer the question of how to reproducibly synthesize promising computationally predicted materials, a custom-designed co-sputtering reactor “combi-9” was recently designed and built at NREL. Equipped with four cathodes, this ultra-high vacuum instrument allows the exploration of a wide substrate temperature range, from cryogenic temperatures up to 1000 °C, besides RF and DC substrate biasing. Additional capabilities include real-time deposition data logging of sputtering parameters (such as power, voltage, pressure, gas flow), control of gas distribution to individual targets, time-sequenced shutters, and turbo gate position, all of which enable the user to execute complex programmable synthesis recipes.<br/>However, this automated instrument still requires an expensive and time-consuming trial-and-error approach for synthesis and optimization of novel materials, since the researcher oversees all the sputtering parameters of an experimental campaign to obtain a specific material composition or properties. Therefore, to allow reproducible synthesis of promising materials with minimal human intervention, we are transforming this <i>automated</i> instrument into an <i>autonomous</i> one, in which a Bayesian algorithm has full control of all sputtering variables, (e.g., plasma powers, gas pressure, substrate temperature, etc.). This algorithm is informed about the process environment by in-situ spectral measurement tools, with the objective to learn how to control the chemical composition of the film, especially it’s mixed-anion content, during the deposition process.<br/>This presentation will describe this state-of-the-art automated sputtering instrument and discuss our progress towards turning it into an autonomous sputtering reactor, such as estimating film composition from in-situ spectral data, integration among many software platforms, data logging and integration with databases.