Physical and Fuzzy Logic Modeling of a Flip-Chip Thermocompression Bonding Process

Abstract

Flip-Chip connections using gold-to-gold, gold-to-aluminum, or gold-to-solder bondings or contacts enhanced by epoxy are low-cost alternatives to soldering. To assist their technology advancements, we have developed yield models for a representative assembly process with flip-chip, thermocompression bondings. Based on bonding mechanics, a physical yield model has been developed to characterize the process. Then, a fuzzy logic model has been established to improve the modeling’s accuracy by including experimental data. The physical yield model can predict the assembly yield as a function of forces and planarities of the end effector, bump height variations, bump geometries, mechanical properties corresponding to different materials and temperatures, and distribution patterns of bumps. Consistent with our experimental experience, the calculated force level for a high-yield process was around 3000 gmf for a 30-gold-bump chip with a bump diameter of 60 μm and a height of 50 μm. The fuzzy logic model can be trained and adjusted by the results of physical models and experiments. It correlates very well to the nonlinear relationships between the yield and the assembly parameters, and has a self-learning capability to update itself with new data. Such capabilities have been demonstrated by studying the bonding on a substrate with or without a compliant layer.

Geo-environmental impact assessment and management information system for the mining area, Northeast China

“…TSFNN, proposed by Takagi and Sugeno (1985), can achieve a better performance in terms of mathematical function approximation in modeling highly nonlinear systems (Kang et al 1993;Tzafestas and Zikidis 2001). Compared with the aforementioned methods, one of the main advantages of TSFNN is that its operation is comparatively simple and easy to understand owing to a relatively small number of free parameters that can be tuned without calculating weight (Clerc and Kennedy 2002).…”

Modelling and optimization of fluid dispensing for electronic packaging using neural fuzzy networks and genetic algorithms

“…It has the capability to transform a nonlinear mathematical model into a simplified input-output structure (Ying 1998). Kang et al (1993) have proved that the TSK fuzzy system approach outperforms statistical regression and polynomial models in both correlation and prediction in modelling of highly nonlinear systems. Compared with conventional approaches of fuzzy logic, recent research has shown that neural fuzzy systems can achieve better performance, at least in mathematical function approximation, compared with the conventional approaches with the same number of fuzzy sets used in input variables (Fiordaliso 2001).…”

“…However, previous studies also found that the performance of a developed neural network is quite dependent on the pre-defined neural network architectural design as well as on the setting of the neural network parameters. A fuzzy logic modelling technique has been applied successfully in developing models of various manufacturing processes such as the Flip-Chip bonding process (Kang et al 1993), vapor phase soldering (Xie et al 1995) and the waterjet depainting process (Babets & Geskin 2000). In this approach, the basic elements of a fuzzy logic model are the internal functions, the membership functions and the outputs.…”

“…Recent research shows that the approach of neural fuzzy networks has been used to develop process models for various manufacturing processes which are both nonlinear and complex (Tzafestas & Zikidis 2001, Kang et al 1993, Brinksmeier et al 1998, Petri et al 1998, Ling et al 2003. It has the capability to transform a nonlinear mathematical model into a simplified input-output structure (Ying 1998).…”

The impacts of mining exploitation on the environment in the Changchun–Jilin–Tumen economic area, Northeast China

“…Takagi-Sugeno fuzzy neural network (TSFNN) which was proposed by Takagi and Sugeno (1985) has attracted much attention (Yoon et al 1994;Kuo and Xue 1998;Zelezniknow and Nolan 2001). When compared with the conventional approaches of fuzzy logic, TSFNN can achieve better performance in mathematical function approximation in modeling highly nonlinear systems (Kang et al 1993;Tzafestas and Zikidis 2001). When compared with the aforementioned methods, one of the main advantages of TSFNN is that it is comparatively simple in operation and easier to understand owing to a smaller number of free tunable parameters without calculating weight (Clerc and Kennedy 2002).…”