Estimation of the mobility of low temperature polycrystalline silicon thin film transistors through deep learning

The crystallization of a-Si leads to alterations in the morphology of Si film such as surface color and surface roughness as a result of excimer laser annealing (ELA). These surface changes correlate with the characteristics of polysilicon films. The quality of crystallized poly Si has been evaluated by Non-destructive optical inspection methods. This study aims to use deep learning to estimate the quantitative relationship between the microscope images of a low-temperature polycrystalline silicon (LTPS) film and the mobility of an LTPS thin film transistor (TFT). This method would make it possible to measure the mobility from the images captured after annealing and improve the crystallization by in situ feedback. An a-Si substrate with a film thickness of 100 nm was polycrystallized by employing a KrF (wavelength of 248 nm) excimer laser, after which an optical microscope image of the substrate was captured. By changing the laser fluence and the number of shots (44 conditions N=10), LTPS films of various surface morphology were fabricated. We fabricated 440 transistors using these LTPS channels (channel size L = 20 μm, W = 30 μm) and measured their mobilities. Then, we performed deep learning with these sets of annealed optical microscope images and the corresponding mobilities. The mobility was estimated with an accuracy of ±12.8 cm²V^-1s^-1. Further improvement of the prediction accuracy (<±5 %) is needed for in-situ feedback. We plan to increase the number of images and use transfer learning to improve prediction accuracy.