Monte Carlo Simulation of BGA Failure Distributions for Virtual Qualification

Any approach to qualification of advanced technologies during product development must include an assessment of variation expected in product life over the life cycle. However, testing product design options in development, to approach an optimal design, is costly and time consuming. Hence, simulation of product life distributions for virtual qualification, can be a valuable tool to evaluate and qualify design options. This paper presents a physics of failure based approach to virtual qualification of advanced area array assemblies, against solder fatigue failure. The approach applies Monte Carlo Simulation to evaluate solder joint fatigue life distributions, given material property variations and manufacturing capabilities. Preliminary results using the simple Engelmaier Model as the basis of simulations are presented. Simulation results are compared to data accumulated from two test environments and two BGA product types. The results reveal some of the limitations of the Engelmaier Model as a basis for simulation. They also show the potential of this approach to virtual qualification for design and manufacturing capability assessment in development. INTRODUCTION Product qualification is intended to assure that a new design will meet the lifetime required for the application with minimal risk of failure. In addition, qualification is used to assure manufacturing processes produce minimal risk of early failure due to defects or inadequate process capability. We are forced to recognize in applying qualification, that materials properties and various geometric variables subject to manufacturing processes are random variables. These variations give rise to uncertainty in product life, as shown in Figure 1. Material Properties Manufacturing Capability Product Life + pdfm pdfg pdflife Figure 1 Materials property variation Qualification must assure the design and manufacturing capability produce adequate product life. The method of qualification must also recognize variations in materials properties and manufacturing capability produce uncertainty in product life. In conventional qualification, we therefore select a sample size and test the sample products under stress to failure or until some specified time period. We then examine the confidence we have in our design and manufacturing processes. Yet, many options in architecture may be selected to create a hierarchy of packaging for an electronic product. There are many materials options and potential manufacturing processes. In addition, new developments provide expanding options. Virtual qualification is therefore desirable as an alternative to testing all potential options. Virtual qualification implies that we evaluate a model of the product under stress. We can therefore evaluate and qualify many options in architecture for a product in rapid time, at lower cost. However, we must still fulfill the requirement of understanding the impact of variation as illustrated above, in Figure 1. Hence, Monte Carlo simulation becomes a valuable approach to developing a virtual qualification scheme. The following sections present an approach to simulation of life based upon input variation of materials properties and manufacturing capability. To illustrate this approach to virtual qualification, we apply the process to Ball Grid Array architectures and compare the results to qualification test data on actual product. Two test conditions and architectures are evaluated. As summarized in Table 1. Table 1. Qualification Conditions Test Facility BGA Package Style Actual Test Sample Size Cyclic Conditions Jet Propulsion Laboratory 313 PBGA 35 mm X 35mm 1.27 mm pitch Full Array 15x15mm die n=13 -30 to +100C tD @100 C = 20min tD @-30 C = 10 min Ramp time = 20 min Motorola Semiconductor Products 119 PBGA 14 mm X 22 mm 1.27 mm pitch Full Array 9 x 16 mm die n = 27 0 to +100C tD = 5 min Ramp time = 10 min VIRTUAL QUALIFICATION BY SIMULATION A method of virtual qualification is summarized in Figure 2. In this case, time to failure models representing the dominant failure mechanisms are embedded in a Monte Carlo simulation. Input variations representing manufacturing capability and material properties are modeled as triangular distributions. These estimates must be extracted from materials testing and manufacturing history. The process is exercised as follows: • The test or application conditions are determined. • A failure model is selected. • The input distributions are then sampled using a random number generator. • The life is calculated from the failure model. • The result is stored. • The input distributions are sampled again and calculation is repeated for a preset number of samples. • The results of the stored are analyzed by fitting the data to a distribution, which represents the life distribution of the failure mechanism. In the case of the BGA application, we initially have selected the well-known Engelmaier model to represent the failure mechanism of solder fatigue for these initial studies. Fig. 2. Virtual Qualification Process by Monte Carlo Simulation (Evans and Evans 1999). Select Failure Mechanism Model t f=f(g,m,p,e,d) Calculate tf from Failure Mechanism Model and Save Result P r o b a b i l i t y