22‐2: Low power consumption optimization using conformalized quantile regression

Abstract

AMOLED IT develops low‐frequency technology to meet the market's low power consumption needs. If the frequency is lowered to 48hz, the logic power consumption compared to 120hz can be reduced by about 30%. However, the flicker phenomenon during low‐frequency operation is a significant defect that customers can immediately recognize. The cause of the flicker phenomenon is known to have various effects, such as process, design, and device characteristics. However, there needed to be a way to calculate it, considering all parameters accurately. To solve this issue, we developed an AI that can predict by collecting information on the process, panel design, device characteristics, and flicker data mentioned above and using XAI to analyze the AI. As a result, we could determine the cause of the flicker under what circumstances. We applied synthetic generation AI to improve insufficient experimental data and improved predictive consistency through optimization. We additionally discovered a cause that was not well known before, and the importance of each parameter was identified. As a result, using this technology to predict the flicker phenomenon in advance when changing the process, prevent failures, and predict the risk of new IT models. Moreover, the prediction reliability of the model was verified by conformalized quantile regression to minimize the risk.