The Woodruff School

SUMMARY

Packaging for high-performance computing requires multiple logic and memory dies assembled on a single substrate. Such a 2.5D package demands a large (35x35mm or larger) and ultra-thin (100um or less) substrate with asymmetric build-up, high density wiring, and ultra-fine pitch interconnects (35mm or less). While glass is an ideal substrate material for such packages, there are challenges, such as glass cracking due to dicing-induced defects and RDL stresses as well as debonding of copper redistribution layers (RDL) from the smooth glass surface. To address these challenges, there is a need to understand plasticity effects on thin film adhesion, the role of dicing defects on glass cracking, and process-induced stresses due to RDL. The objectives of this work are to understand the fundamental factors that contribute toward the cracking of glass and debonding of RDL, to design and demonstrate thermo-mechanically reliable 2.5D glass packages, and to develop design and process guidelines for such reliable glass packages. This work studies how RDL stresses propagate dicing-induced defects into cohesive cracks as well as interfacial delamination, how geometry and process modifications could mitigate such failures, demonstrates prototypes that are reliable through processing and thermal cycling, and develops design guidelines for current and future glass packages. Stresses in glass caused by RDL are correlated to models through birefringence and shadow moire warpage measurements for model validation. Dicing methods and the associated defects are comprehensively quantified. Test vehicles are designed and their reliability is demonstrated through 1000 thermal cycles. Three alternative solutions to glass cracking, edge coating, two-step dicing, and laser dicing, are proposed, analyzed, and demonstrated. A method to determine the critical energy release rate for peeling of a copper thin film from a glass substrate is developed, and employed to enhance adhesion of copper wiring. In addition, general design and process guidelines for mechanical reliability, which are applicable to other packaging applications, are developed.

SUBJECT:	Ph.D. Dissertation Defense

BY:	Scott McCann

TIME:	Monday, March 20, 2017, 1:00 p.m.

PLACE:	MARC Building, 201

TITLE:	Design for Mechanical Reliability of Redistribution Layers for Ultra-thin 2.5D Glass Packages

COMMITTEE:	Dr. Suresh K. Sitaraman, Chair (ME) Dr. I. Charles Ume (ME) Dr. Shuman Xia (ME) Dr. Rao R. Tummala (MSE) Dr. Venkatesh Sundaram (ECE)

SUMMARY Packaging for high-performance computing requires multiple logic and memory dies assembled on a single substrate. Such a 2.5D package demands a large (35x35mm or larger) and ultra-thin (100um or less) substrate with asymmetric build-up, high density wiring, and ultra-fine pitch interconnects (35mm or less). While glass is an ideal substrate material for such packages, there are challenges, such as glass cracking due to dicing-induced defects and RDL stresses as well as debonding of copper redistribution layers (RDL) from the smooth glass surface. To address these challenges, there is a need to understand plasticity effects on thin film adhesion, the role of dicing defects on glass cracking, and process-induced stresses due to RDL. The objectives of this work are to understand the fundamental factors that contribute toward the cracking of glass and debonding of RDL, to design and demonstrate thermo-mechanically reliable 2.5D glass packages, and to develop design and process guidelines for such reliable glass packages. This work studies how RDL stresses propagate dicing-induced defects into cohesive cracks as well as interfacial delamination, how geometry and process modifications could mitigate such failures, demonstrates prototypes that are reliable through processing and thermal cycling, and develops design guidelines for current and future glass packages. Stresses in glass caused by RDL are correlated to models through birefringence and shadow moire warpage measurements for model validation. Dicing methods and the associated defects are comprehensively quantified. Test vehicles are designed and their reliability is demonstrated through 1000 thermal cycles. Three alternative solutions to glass cracking, edge coating, two-step dicing, and laser dicing, are proposed, analyzed, and demonstrated. A method to determine the critical energy release rate for peeling of a copper thin film from a glass substrate is developed, and employed to enhance adhesion of copper wiring. In addition, general design and process guidelines for mechanical reliability, which are applicable to other packaging applications, are developed.