在LCM Chip-on-Glass (COG) 封裝製程，因玻璃基板和驅動IC之間熱膨脹係數的差異，當驅動IC藉由異方性導電薄膜(ACF)與玻璃基板黏合後，將會使面板產生應力進而導致漏光Mura發生。隨著市場產品的需求，對應的面板有越來越薄的趨勢，而使用更薄的玻璃基板將使得漏光Mura缺陷更加嚴重。 本研究首先以光測力學方法中的光彈法，量測玻璃基板封裝矽晶片周圍區域的相位差(Retardation)，並經由應力-光學定律求得玻璃基板應力值。但玻璃是一種低雙折射性材料，意即玻璃在受到應力時所產生的雙折射效應非常微弱，在量測薄化玻璃基板十分不易。 因此，利用類神經網路模擬出COG封裝產生之應力的預測模型，作為薄化玻璃基板應力值預測，並且藉由模型預測了解改善對策對於降低應力值之效益，以節省大量材料、人力與時間成本。 ;In the LCM Chip-on-Glass (COG) packaging process, there are differences in the thermal expansion coefficient between the glass substrate and the driver IC, when the driver IC uses anisotropic conductive film (ACF) to bond to the glass substrate, the panel produces stress which leads to the occurrence of light leakage through Mura defects. The market demand trend is for increasingly thinner production panels, but the use of thinner glass substrates increases the likelihood of more serious light leakage Mura defects. The first phase of the research was to measure the light mechanics by the photoelasticity method through measuring retardation in the area surrounding the glass substrate packaged silicon chip, and adding stress to obtain the optical glass substrate stress values. However, as glass is a material with low birefringence, the birefringence effect generated when glass is subjected to stress is very weak, and so the measurement of thin glass substrate is extremely difficult. Therefore, a neural network model is used to simulate COG package production stress prediction. Thin glass substrate stress values were predicted and used by the model to predict and understand the improvement efficiency measures in reducing stress values. This could save a lot of materials, manpower and time costs.