Thermal simulations of a system-in-package for space applications

Abstract

Frontgrade Gaisler AB is investigating a System-in-Package (SIP) based on their radiation hardened, fault-tolerant, octa-core, SPARC LEON and RISC-V microprocessor (GR765) together with DDR4 memories. The SIP aims to provide reliable, high performance processing in a small package form factor for space applications. The packaging of processor and memories together increases the demands on the package for proper thermal management. This task is further complicated by the harsh space environment, which needs special considerations during the design process. This work supports the developments of the GR765 SIP through thermal simulations using computational fluid dynamics (CFD). Example models of the SIP were built and simulated using ANSYS Icepak to determine where in the package hot spots occurred, how heat transferred through different parts of the package, and how different packaging methods and materials impacted the results. The simulations of the thesis featured six different models. The first three models centered around finding an appropriate level of detail, which should give accurate results while not becoming too computationally intensive. These models also showed the impact which simplifications like ideal contact resistances and homogeneous heat generation had on temperature calculations. Model I, featuring both simplifications, had a peak temperature of 43◦C lower than Model III which featured contact resistances and uneven heat generation. The following two models focused on lowering the package temperatures to fit a thermal target of 125◦C at an ambient temperature of 55◦C. The results of these simulations showed that swapping a plastic overmold for a metal lid could increase the peak temperatures of the package. With a dedicated heat sink the peak temperatures were within the thermal target at 99◦C with 55◦C as the ambient temperature. The last model explored memory stacking using film over wire (FOW). In these simulations, the stacked dies struggled with thermal management resulting in a peak temperature of 151◦C in the memories at an ambient temperature of only 20◦C. The combined results suggest that simplifications can have a large impact on the results, a dedicated heat sink is needed for the design, and additional thermal management methods are needed if die stacking is to be used in this kind of SIP.