Simulation study and parameter optimization of laser TSV using artificial neural networks

Abstract

Through-Silicon Vias (TSV) play an important role in the field of semiconductor 3D packaging by providing interconnections during layer-by-layer stacking. Laser machining of TSVs is advantageous over expensive photolithography steps in the DRIE process. It is environmentally friendly, reduces production costs, and increases throughput. In this study, a nanosecond laser (100 ns, 1064 nm) was used to produce TSVs on a 500 μm thick silicon wafer, and the effects on parameters such as drilling depth, aspect ratio, taper angle, and heat-affected zone (HAZ) were investigated in air. Furthermore, an optimized circle-packing design between pulse energy (E) and the number of pulses (N) was constructed. The developed computational model was employed to run simulations to predict the drilling depth, aspect ratio, taper angle, and HAZ. Subsequently, the obtained results were used to train surrogate models for all individual parameters in the given range. A final processing map was created by combining all the individual surrogate models and filtered with four criteria to obtain the optimized region with proper drilling depth, less taper angle, high aspect ratio, and comparatively less HAZ TSVs. Lastly, points were chosen from each region, and experiments were performed to validate the optimized zone, and full agreement was found with the simulation results. With this optimized zone, TSVs with a higher aspect ratio (11.72), less taper angle (1.37°), and comparatively less HAZ thickness (10.48 μm) were achieved for the combination of pulse energy at 0.3614 mJ and pulse number at 835.

Keywords

• TSV
• Nanosecond laser
• Parameter optimization
• HAZ
• Aspect ratio