Modeling the Laser Ablation Process Using Neural Networks

Laser ablation has become an increasingly important process in the fabrication of microelectronic packaging substrates. In this research, laser ablation has been applied to drilling via holes during the sequential build-up process of a multilayer substrate. Since the quality of the process is influenced considerably by the process set points, studies that characterize the relationship between these conditions and the characteristics of the vias formed are necessary. This paper examines vias with diameters of 30, 40, and 50 μm ablated in DuPont Kapton^® E polyimide using an Anvik HexScan™ 2150 SXE excimer laser. A statistical experiment using a 2⁵⁻¹ fractional factorial design was conducted to determine the significance of laser energy, shot frequency, number of pulses, and vertical and horizontal positions of the debris removal system in the laser tool affecting the top via diameter, via wall angle, and the ablated thickness of the dielectric. Several input factors and two-factor interactions were found to be statistically significant (p-value < 0.05). Following the collection of the experimental data, neural networks (NNs) were trained using the error back-propagation algorithm to model the average values of the responses. The prediction error for all the NN models was less than 5.5%.