Materials Informatics for Thermistor Properties of Mn–Co–Ni Oxides

Because the electrical properties, sensitivity (B constant), and resistance of thermistors must be fine-tuned according to the environment in which they are used, complex multicomponent transition metal oxides are often used to ensure the degree of freedom of the B constants and resistance parameters. However, the precise control of electrical properties is generally difficult, owing to the complex changes in crystalline phases with composition. In this study, we focused on quaternary Mn–Co–Ni oxides, performed exhaustive sintered body preparation by dividing the entire composition space into 50 parts, and evaluated the crystal phase, bulk density, and thermistor properties (B constant and resistance) of the sintered bodies. Furthermore, machine learning regression analysis was performed on the composition and electrical property data were obtained. However, even for the model with the lowest root-mean-square error, the prediction error of the B constants averaged ∼300 K and that of the common logarithm of resistance (LogR) averaged 0.3 log kΩ mm, indicating the difficulty in controlling the desired electrical properties from the composition at a practical level. In contrast, compositions with arbitrary B constants and LogR could be efficiently determined by Bayesian optimization using the composition ratio as a descriptor.