Neural network modeling of the resistance of metallized vias formed by laser ablation in polymer dielectrics

Abstract

Laser ablation, a material removal process that uses localized thermal energy caused by stimulated radiation, has become an important process in the fabrication of microelectronic packaging substrates, particularly in the fabrication of vias. During laser ablation, debris in the form of carbon residue is generated as a by-product. In this paper, resistance measurements on metal deposited in ablated vias are conducted to characterize the degree to which debris remaining inside the vias affects their quality. Vias with diameters of 10-50 /spl mu/m are ablated in DuPont Kapton/spl reg/ E polyimide using an Anvik HexScan/spl trade/ 2150 SXE excimer laser. A statistical experiment using a 2/sup 5-1/ fractional factorial design is conducted to characterize five process conditions, namely: laser energy, shot frequency, number of pulses, and the vertical and horizontal positions of the debris removal system in the laser tool. Measurements indicate that 10, 20, and 30 /spl mu/m vias are not opened by any combination of the five process conditions. As for the 40 and 50 /spl mu/m vias, both number of pulses and the horizontal position of the debris removal system, as well as their two-term interaction, are found to be statistically significant (p-value<0.05). Following the collection of the experimental data, neural networks are trained and subsequently tested to model the measured resistance through the metallized 40 and 50 /spl mu/m vias. Results indicate that the prediction error of these models is less than 15%.