Comparison of Intelligent Modeling Techniques Applied to Microvia Formation using Excimer Laser Ablation

摘要

Excimer laser ablation is one of the key methods for microvia generation in a multilayer microelectronics packaging substrate. A primary challenge in modeling laser ablation is process nonlinearity, particularly when laser-to-material interactions are involved. In this study, three artificial intelligence (AI) techniques for modeling the formation of microvias in dielectric polymers are compared based on the percentage error achieved in predicting microvia responses. Vias with diameter range from 40 - 50 μm are ablated in DuPont Kapton E polyimide using an Anvik HexScan 2150 SXE excimer laser operating at 308 nm. Central composite circumscribed (CCC) design is utilized to determine the significance of laser fluence, shot frequency, number of pulses, and helium pressure flow affecting the via diameter and via wall angle. Statistical analysis shows that all input parameters are found to be significant (p-value < 0.05) in affecting at least one response. The ablation process is subsequently modeled using the experimental data with these techniques: (1) neural networks trained using error back-propagation; (2) Mamdani's fuzzy logic; and (3) Sugeno's neuro-fuzzy networks trained using a hybrid algorithm. The Mamdani fuzzy logic shows the worst average prediction error (< 10%) for all responses, whereas, both neural and neuro-fuzzy networks show improved prediction error.