Researchers at NTT have successfully utilized artificial intelligence (AI) to enhance the scientific process, particularly in the production of thin films from a semiconductor material known as beta-gallium oxide (β-Ga₂O₃). This material is being explored for its potential use in future power semiconductors, which are critical for the control and conversion of electrical power in electronic devices.
In a recent experiment, NTT directed its efforts toward producing a single-crystal β-Ga₂O₃ film using a method called sputtering. Sputtering involves coating a surface with an extremely thin layer of material by knocking atoms from a source material within a vacuum chamber. However, until this experiment, no one had achieved the production of single-crystal β-Ga₂O₃ films through sputtering.
Successful film production depends on several variables, including temperature, sputtering power, and the flow of argon and oxygen gases into the chamber. Changes to any single variable can impact the quality of the resulting film, making finding the optimal conditions a complex and repetitive challenge. NTT’s approach involved automating the experimental process with AI.
The AI system facilitated a cycle of deposition and evaluation, allowing the research team to produce and assess films at a speed that outpaced traditional methods. After 56 experiments, the AI identified conditions that led to the creation of high-quality single-crystal β-Ga₂O₃ films, marking a significant development in the field.
Despite the success, challenges remain with the black box nature of AI, which can identify effective settings without providing insight into why they work. To address this, NTT employed another machine-learning technique, known as a “random forest,” to analyze data generated during the AI-led experiments. This analysis revealed how the experimental parameters influenced film quality and highlighted critical relationships, particularly between temperature and oxygen flow.
By applying insights gained from the AI’s findings, NTT researchers were able to further refine their process, leading to the production of even higher-quality films. This collaborative approach recognizes the importance of human oversight in scientific inquiries, particularly in understanding complex relationships among variables.
As modern technology continues to rely on advancements in materials science, NTT aims to expand its AI for Science initiative to include additional materials and experimental techniques. The goal is to create a more efficient framework for experimentation that enhances scientists’ ability to innovate and improve products across various technological domains.
This initiative represents a promising avenue for integrating AI into scientific research, offering a way to streamline experimental procedures and deepen the understanding of material characteristics through automated analysis.
