The Scaling and Microstructure Effects on the Thermal Stress and Reliability of Through-Silicon Vias in 3D Integrated Circuits
Supervised by: Dr. Paul S. Ho
Through-silicon vias (TSVs) enable full three-dimensional integration by providing high-density vertical interconnections, improving device bandwidth and power consumption. However, TSVs pose reliability risks due to the thermal stresses induced by the thermal expansion coefficient mismatch between silicon and copper, which causes thermal stress buildup and TSV extrusion to degrade device reliability and performance. It has been proposed that optimal post-plating annealing or downscaling TSVs could mitigate deleterious effects for TSVs. While results show some reductions in the average extrusion, the worst cases and statistical spread are not improved. This work investigates the scaling and microstructure effects on stress and extrusion statistics of TSVs to assess reliability risks for future 3D technology with continued TSV downscaling. The basic mechanisms of the extrusion phenomenon and its correlation to the copper microstructural characteristics are examined in order to trace the root cause of the high statistical spreads in the extrusion results.
Experimental results first establish the characterization of the TSV samples and demonstrate that neither annealing nor downscaling can fully resolve the reliability threats, as further annealing perpetuates abnormal grain growth to increase the TSV extrusion heights and statistical spreads. Extrusion is shown to be statistical in nature, depending on the microstructure and elastic anisotropy of copper, and is not improved by scaling. Synchrotron x-ray microdiffraction is used to measure the local plasticity of individual copper grains and the results were consistent with the extrusion characteristics, confirming the plasticity mechanism of extrusion. This directly correlates the extrusion profile to the local plasticity in the copper grains, where the statistical spread increases with downscaling. Thermomechanical models validate the non-uniform effect of the grain structure on the stress and extrusion behavior. Additionally, a microstructure simulation was carried out, which accounted for the orientation-dependence of the surface, grain boundary, and strain energies. The results, which confirm the statistical scatter observed in the microstructure and extrusion, indicate that elastic anisotropy drives microstructure evolution, causing the bimodal grain distribution upon annealing. The simulation, having been experimentally validated, is further used to assess the scaling effect on copper TSV reliability and the use of alternative materials to improve reliability.