Designing of near-IR organic semiconductors for photodetectors: Machine learning and data mining assisted efficient pipeline

Abstract

Near-infrared organic semiconductors are attractive candidates for photodetector applications due to their inherent characteristics such as room temperature operating conditions, flexible substrates compatibility, ease of processing, and tailorable optoelectronic properties. However, it is challenging to select the appropriate organic semiconductor material for photodetector applications due to their ubiquitous existence. The development, designing, prediction, and discovery of high-performance materials for photodetectors can be accelerated by combined use of computer science and artificial intelligence with traditional synthetic methods. Before carrying out laboratory synthetic methods (trial-error-methods), it is essential to predict the properties of potential candidates in a rapid and computationally cost-effective manner. The characteristics of the molecules can be theoretically established via computational investigations. To develop machine-learning (ML) models, data from previous scientific and theoretical studies are gathered. Additionally, advanced screening technology is used for the evaluation of photodetector materials. Data visualization analysis is used to examine the trends in the database. Cook's distance is employed in machine learning algorithms to identify outliers. Importantly, only two models i.e., extra trees regressor and random forest regressor out of the 15 examined models had the better predictive ability. To improve their capacity to predict, the best algorithms were further adjusted. By searching for new building units in the near-IR database, the validity of our suggested method was further confirmed. The findings showed that a good link between experimental findings and model predictions was attained. To further test the applicability of this novel approach, extra trees regression is utilized, and more than 1000 new organic semiconductors are built using readily synthesizable organic building blocks. This suggests that machine-learning (ML) is an effective method for predicting the characteristics of photodetectors and can speed up their spread across a variety of applications.