Integration of analytical techniques in stochastic optimization of microsystem reliability

This paper presents an integrative application of several numerical analytical techniques and associated analysis tools for design optimization and damage prediction in electronics packages and microsystems. This design-for-reliability approach is based on four different types of numerical techniques that allow (1) high-fidelity modelling, (2) reduced order modelling, (3) numerical optimization and (4) uncertainty analysis. The capabilities and the characteristics of the methods that underpin these four types of modelling and analysis tools are firstly investigated. The integration of the methods and tools is then examined and a methodology for coupling the tools in an optimization process is proposed. This numerical methodology involves the following steps: (1) Define sampling points for the design of interest by design of experiments (DOE) and calculate the design response at each DOE point using high-fidelity analysis; (2) construct reduced order models (ROM) for fast analysis using the obtained response values at the DOE points; (3) Undertake deterministic optimization in the defined design space by ROM; and (4) Probabilistic optimization by including variation and uncertainty of the design in the optimization task. This approach is suitable to address design-for-reliability requirements at early design stages in a wide range of application areas. The application of this approach is demonstrated in a case for minimizing the thermal fatigue damage of flip-chip solder interconnects. Design modifications show that this approach can provide improved reliability of the package and in the same time satisfy a number of design requirements.