Neural network modeling to predict quality and reliability for BGA solder joints

Quality is major competitive advantages in today's business environment. Engineering tasks encompasses the assurance of quality and reliability. Therefore, one goal is the prediction and modeling of quality and later on reliability of systems, subsystems and components. An approach of quality and reliability assurance uses failure prevention and process control, which by itself is based on quality data and technological understanding. The bases for quality and reliability prediction are information about used materials, design parameters and process parameters as well as the underlying relationships. Analyzing these data for underlying relationships between control parameters (materials and process setups), monitoring parameters (such as humidity) and target variables is one approach to assure quality output. Within this paper neural networks for analyzing relationships are investigated. Two types of neural networks are investigated which are namely back propagation networks (BPNN) and secondly radial basis function networks (RBFNN). The test objects are BGA solder joints which are manufactured using different process setups and materials. As quality measure the ratio of voids in a solder joint is used. The criterion for good prediction quality is the ability of generalization of the depicted models when applying new data to it.