FINE-PITCH STENCIL PRINTING PROCESS MODELING AND OPTIMIZATION

In this paper we present a statistical-neural network modeling approac h to process optimization of fine pitch stencil printing for solder pa ste deposition on pads of printed circuit boards (PCB). The over all o bjective was to determine the optimum settings of the design parameter s that would result in minimum solder paste height variation for the n ew board designs with 20-mil, 25-mil, and 50-mil pitch pad patterns. A s a first step, a Taguchi orthogonal array, L27, was designed to captu re the main effects of the six important printing machinery parameters and the PCBs pad conditions. Some of their interactions were also inc luded Fifty-Sour experimental runs (two per setting) were conducted. T hese data were then used to construct neural network models relating t he desired quality characteristics to the input design parameters. Our modular approach was used to select the appropriate architecture for these models. These models in conjunction with the gradient descent al gorithm enabled us to determine the optimum settings for minimum solde r paste height variation. Confirming experiments on the production lin e validated the optimum settings predicted by the model. In addition t o the combination of all the three pad patterns, i.e., 20, 25, and 50 mil pitch pads, we also built neural network models for individual and dual combinations of the three pad patterns. The simulations indicate different optimum settings for different pad pattern combinations.