Application of Bayes optimization and machine-learning regressions to seek fabrication condition of a-In-Ga-Zn-O for gas sensor applications

Machine learning approaches were applied to efficient exploration of the relationship between deposition conditions and properties of a-In-Ga-Zn-O thin films. Using Bayes optimization based on Gaussian process regression, we found the optimum deposition condition to obtain the lowest film density only with approximately one-fifth of the total data. The detailed relationship between the deposition conditions and film density was modelled by the non-parametric Gaussian process and multilayer perceptron regression. In addition, the relationship between deposition condition and thin-film transistor characteristics was also investigated.