Practicality Study on AI-Assisted AOI in Inkjet-Printed FPC Manufacturing for Predictive Electromagnetic Responses

Abstract

An artificial intelligence (AI) model based on Gaussian process regression with covariance function of exponential was established with inkjet-printed flexible printed circuits (FPCs). The AI model was proposed for antenna characteristic prediction through line edge roughness (LER) of the fabricated patterns as a part of the automated optical inspection (AOI). FPCs of antennas in spiral shapes operating at 2.45 GHz were designed with three pattern distortions to imitate practical situations in inkjet printing; and 144 and 36 samples were fabricated as the training and testing groups for AI modeling, respectively. The capacitance value, operating frequency, and coupling efficiency in terms of intensity loss were characterized through simulation, measurement, and prediction in this work. Results indicated that along the deterioration of the LER, all three characteristics degraded as expected. Quantified statistics proved that the predicted capacitance value, operating frequency, and intensity loss exhibited remarkable correctness: root mean square error was 0.0499 pF, 0.0004 GHz, and 0.0430 dB, respectively; with linear coefficient of determination of 0.9862, 0.9828, and 0.9946 from their measured counterparts, respectively. These results outperformed those demonstrated with an AI model established with printed circuit board, which failed to reflect genuine wetting behaviors usually appear in printed electronics. The difference between the measured and predicted capacitance value, operating frequency, and coupling efficiency was merely 0.1583%, 0.0159%, and 0.0564%, respectively; indicating the effectiveness and practicality of introducing the AOI and AI for off-line characteristic prediction from in-line pattern integrity of microelectronic component in a production line.