In the field of materials science, aluminum nitride (AlN) materials have great application potential in key fields, such as optoelectronic devices and high-frequency electronic devices, due to their unique optoelectronic properties. However, it is difficult to obtain the parameters related to its optoelectronic properties, and traditional measurement methods have low precision and poor efficiency. This study aims to solve this problem by constructing an attention-based deep neural network model to perform parameter inversion with high precision and efficiency. An experiment was conducted by employing a dataset comprising the optoelectronic properties of AlN materials, covering different preparation processes and crystal structure states, and the results were compared with an improved physical model (Model 1), a model based on statistical learning (Model 2), and a model based on traditional neural networks (Model 3). The results show that the deep learning model with attention has a significant advantage in terms of the measurement error rate, and the error rate of the light absorption coefficient is only 3.2%, which is much lower than that of Model 1 (12.5%), Model 2 (10.8%), and Model 3 (8.6%). In terms of the measurement efficiency, when the light absorption coefficient is measured as an example, the number of effective measurements per unit time can reach 50, which is far greater than that of the other models. This study provides a new way to measure the optoelectronic parameters of AlN materials and is expected to promote the development of related industries.

抄録全体を表示