Modeling and optimization of integral capacitor fabrication using neural networks

Integral capacitors for MCM-L/D technology are an important component for next-generation electronic packaging applications. Thin film integral capacitors using polymer-ceramic composites are developed to achieve this need. To achieve a higher dielectric constant, bimodal ceramic particle distributions, along with particles modified by a surfactant and mixed ultrasonically with the polymer are explored. We present a statistically designed experiment for systematic characterization of dielectric constant and loss tangent of integral capacitors formed in this manner by using barium titanate particles in an epoxy polymer dielectric. We determine these quantities as a function of the particle size of the ceramic, the volume fraction of ceramic in the polymer matrix, the polymer cure time, the polymer cure temperature, the percentage of surfactant, the ultrasonic mixing time, and the ball milling time for ceramic surface modification. These factors are initially examined by means of a partial factorial screening experiment requiring 32 runs. Result indicate that these six factors are statistically significant for modeling dielectric and loss. Further experimentation is performed to generate sufficient data for process modeling. To develop such models, we train neural networks to model the variation as a function of input variables using the experimental data