P‐117: AI Analysis of HOP Circuit Failure and Improvement

AMOLED's HOP(hybrid of Oxide semiconductor and Polysilicon TFT) technology was born to meet the market's demand for low power consumption. Also, the IGZO oxide process was introduced with new driving technology to support the low driving frequency operation. However, this new GC circuit of oxide TFT driving caused new and randomly appearing horizontal defects which do not exist in LTPS circuits. To solve this issue, we studied to find the design parameters using pipelined deep learning models composed of CGAN(Conditional Generative Adversarial Network), VGM(Variational Gaussian Mixture), and XAI(eXplainable Artificial intelligence) technologies to figure out what causes these defects. We collected HOP panel design information to achieve the given goal and confirmed the degree of defect per HOP model. Since the number of HOP models is limited, XAI analysis might also be negligible. We solve this issue by generating data using CGAN and VGM in sequence. After, ALE(Accumulated Local Effects), a popular XAI technique, is applied to find affected design parameters and change‐related circuit designs to help verify the proposed algorithm's performance. As a result of this design change, we identified that the related defect rate decreases from 15000ppm to less than 500 ppm, in which XAI plays a significant role in solving a newly introduced defect issue. That was believed for a long time a challenging problem to analyze using the legacy SPICE tool in the OLED circuit field. Our result also suggests that even predicting the defect rate of future HOP models becomes feasible with our proposed deep learning architecture.