Measurements of Anisotropic Thermal Conductivity of a Multilayer Ceramic Substrate Over a Uniform Spatial Distribution

The thermal conductivity of multilayer packages and printed wiring boards is often modeled as a volumetric weighting of thermal conductivities of the component materials. We demonstrate that this approach overestimates effective thermal conductivity, especially normal to the plane of the package. The reason is that the interfacial heat flux between metal and dielectric layers is significantly less than that within a metal layer. As electronic packaging becomes more intricate and complex, thermal performance analyses and predictions must be more accurate than ever for the industry to progress. In comparison to the ceramic packages we subsequently analyze, printed circuit board layups typically include conductive and dielectric materials that have a higher differential of thermal conductivity, more materials, thinner layers, vias, or embedded components. Each aspect further complicates the analysis and increases the likelihood of error in overestimating the resultant thermal conductivity. We developed an experimental procedure that confirmed our prediction of overestimation by measuring the thermal conductivity in x, y and z directions on 5 millimeter square segments cut from a 31 millimeter square multilayer ceramic package. The data we collected was used to refine a thermal model of our package, and to validate finite element thermal models of the 5 millimeter square substrate pieces. We are proceeding to use parametric finite element models to compute the thermal performance of a large number of hypothetical package segments. The computed thermal conductivity vectors from these models will be combined with attributes of each segment, such as metal fraction, number of layers, conductor line length, etc. to create a data set that can be used to train an inference engine to predict the thermal performance of an advanced package design from its artwork data.