Paper TF-TuA1
Accelerated Searching of Potential Precursors for Silicon Carbide-atomic Layer Deposition from Ab-initio Machine Learning Methods

Tuesday, October 31, 2017, 2:20 pm, Room 20

Due to the superior thermophysical properties of silicon carbide at high temperature, silicon carbide (SiC) coatings have the potential to offer excellent resistance to both oxidation and hydriding of zircaloy-based nuclear fuel cladding used in light water reactors. Unfortunately, the current deposition technique for SiC using chemical vapor deposition (CVD) can be only achieved at relatively high substrate temperatures, which can severely degrade the performance of zircaloy cladding. As a comparison, atomic layer deposition (ALD) enables the growth of pinhole free films on large and convoluted substrates with nanometer precision that operates at low temperatures. Developing a new ALD process requires better understanding of how the film growth process takes place, which depends on the chemicals being used. The major obstacle to low-temperature ALD of SiC is to determine the right precursors. To this end, we use high-throughput ab initio calculations and state-of-the-art machine learning (ML) techniques to accelerate the precursor screening. The method involves two different types of prediction: the forward and backward predictions. The objective of the forward prediction is to train a set of machine learning models for the free energies of silicon and carbon-containing molecules from the high-throughput ab initio database. Inverting the trained forward models through Bayes’ law, we quantitatively predict the free energies of all the possible silicon and carbon-containing molecules from the PubChem compound database. By calculating the Gibbs energy of reaction using the ML predicted energy, several potential silicon and carbon precursors are predicted to be promising for ALD of SiC at low temperature. We believe the present method will be helpful to develop novel ALD precursors for other applications.