Machine learning approach leads to discovery of high-performance infrared functional materials

Infrared optoelectronic functional materials are essential for applications in lasers, photodetectors, and infrared imaging, forming the technological backbone of modern optoelectronics. Traditionally, the development of new infrared materials has relied heavily on trial-and-error experimental methods. However, these approaches can be inefficient within the extensive chemical landscape, as only a limited number of compounds can achieve a balance of several critical properties simultaneously.

To tackle this challenge, researchers from the Xinjiang Technical Institute of Physics and Chemistry of the Chinese Academy of Sciences have made significant strides in the machine learning (ML)-assisted discovery of infrared functional materials (IRFMs). The research team has developed a cohesive framework that integrates interpretable ML techniques to facilitate the targeted synthesis of these materials.

The paper is published in the journal Advanced Science.

Through in-depth interpretable analysis of their model, the team was able to extract key domain knowledge pertinent to the chalcogenide system. Utilizing this knowledge, they employed an IRFM predictor to effectively guide the experimental synthesis of new materials.

This led them to the discovery of a new family of selenoborate halides: ABa₃(BSe₃)₂X, where A represents Rb or Cs and X stands for Cl, Br, or I. These compounds demonstrate a well-balanced set of properties, including wide bandgaps, strong second harmonic generation response, moderate birefringence, and high laser-induced damage thresholds, indicating great potential as high-performance IRFMs.

Furthermore, an analysis of the structure–property relationships revealed that the [BSe₃] unit significantly contributes to the outstanding optical properties, suggesting its potential as an active building block for future exploration of high-performance IRFMs.

This study overcomes the limitations of conventional trial-and-error methods, paving the way for AI-driven design of functional crystal materials.