Analysis Of Impact Factors On The Bonding Quality Of Gold Ball Bonding Process

As the oldest and most widely used inner connecting mode, wire bonding technology changes constantly, in order to meet the demands and challenges of new crafts and materials. There are so many process parameters affecting bond quality in wire bonding and these parameters are directly related to the reliability of semiconductor device. It is hard to express the true relationships between them using mathematics, because of the nonlinear relationships and interactions between these parameters and the final bond quality. Reasonable parameters settings are very difficult to determine. They depend on the experience of the engineer, but still the optimization and stabilization can't be guaranteed. Therefore this dissertation carries out a thorough study on the factors which influence the ball bonding quality in 1mil gold wire bonding, and demonstrates a new process to predict the bond strength and uses ANFIS to optimize those parameters. The research work is focused on the following aspects:1. Conventional methods of prediction, such as time sequence, regression analysis have problems of response instability, factors coupling explicitly mentioned, resulting in low precision. The modern prediction techniques use Artificial Neural Networks (ANN), but this has weaknesses of low training speed caused by random initialization, unclear physical meaning of inner nodes and over-fitting. An alternative method of predicting complex relationships is Fuzzy theory, this depends upon the experience and knowledge of experts to build the rule base. The ANFIS (Adaptive Neuro-Fuzzy Inference System) which combines ANN learning and Fuzzy logic is adopted by this study. It not only can accommodate the approach degree of nonlinear model, but also get the inner law of real model, namely knowledge base. This is important for adjusting the model built by ANFIS to suit different production conditions or die materials.2. There are many factors affecting bond quality in wire bonding process. The parameters investigated in this research are the main process parameters power, force, USG (ultrasonic generator) frequency etc., which are described in detail. They provide the basis for the subsequent tests. The responses of shear force and MBD (Mashed Ball Diameter) are chose for the process output variables. For the input variables of ANFIS, the influence extents of nine parameters against the two responses are compared using ANOVA of Taguchi design. Accordingly, six parameters are confirmed as inputs. Sixty and fifteen groups of Stat. data are measured and used as training samples and cross-validation samples respectively. It is shown that the average errors of shear force and MBD are 3.16% and 1.24%, the maximum errors of the responses are 7.34% and 2.06% respectively. It is also shown that ANFIS models the real process of wire bonding very well.3. Finally, the rules of each input parameter affecting bonding strength and quality are analyzed based on ANFIS model; optimizations of three cases are calculated.The ANFIS model of wire bonding and Fuzzy rule base created automatically provide a foundation of research on fast model adjusting under different bonder or die materials, and real-time monitoring technique.