Generation_of_Ultra_packed_Thermal_Greases_and_Evaluation_of_their_Effective_Properties.pdf (2.28 MB)

Generation of Ultra-Packed Thermal Greases and Evaluation of their Effective Properties

Download (2.28 MB)
posted on 16.01.2019, 15:11 by Sukshitha Achar Puttur Lakshminarayana
Thermal Greases are gap-filling interface materials that are used in semiconductor packages to efficiently transfer heat from the component to the heat sink or spreader. Thermal greases are typically particle filled composite materials comprising of highly conducting fillers in a poorly conducting, but mechanically soft, silicone or epoxy base matrix. Generally, the effective conductivity of the greases increases with increasing volume fractions of fillers. However, the fillers also have high elastic modulus that induces undesirable thermal stresses on the brittle silicon device. Therefore, as device power density increases, there is a need to increase particle volume loading, which in turn necessitates optimally balancing the material’s thermal and mechanical characteristics.

In this thesis, procedures are developed to simulate packed microstructures of particles so as to identify the optimal trade-off between thermal and mechanical behavior. Experimental and numerical simulations of microstructures that have been generated as reported in the literature were found to have volume fractions of around 60%. However, as commercially available thermal greases have volume fractions in the range of 60 − 80%, there is a need to develop an efficient algorithm to generate microstructures numerically. The particle packing is initially posed as a nonlinear programming problem and rigorous optimization search algorithms are systematically applied to generate particle systems that are compactly packed, but without particle overlap. Since the packing problem is computationally expensive, the algorithms are systematically evaluated to improve computational efficiency as measured by the number of particles in the system, as well as the time to generate the microstructure. The evaluated algorithms include the inefficient penalty function methods, best-in-class sequential programming method, matrix-less conjugate gradient method as well as the augmented Lagrangian method. In addition, heuristic algorithms are also evaluated to achieve computationally efficient packing. The evaluated heuristic algorithms are mainly based on the Drop-Fall-Shake method, but modified to more effectively simulate the mixing process in commercial planetary mixers. With the developed procedures, Representative Volume Elements (RVE) with volume fraction as high as 74% were achieved.

After the microstructurs were generated, the effective thermal conductivity and effective elastic modulus were estimated using a ‘Random Network Model (RNM)’ that was previously developed. The RNM solves the near-percolation heat conduction problem with hundreds of thousands of particles in minutes compared to hours or days that a full-field simulation requires. The approximations inherent in the RNM are valid if the particulate composite has widely different matrix and particle properties, which is true in the case of thermal greases. In the present thesis, the previously developed RNM was modified to account for the fact that the generated RVEs contain sides with cut particles.


Degree Type

Master of Science


Mechanical Engineering

Campus location

West Lafayette

Advisor/Supervisor/Committee Chair

Ganesh Subbarayan

Additional Committee Member 2

Amy Marconnet

Additional Committee Member 3

Justin Weibel