Design and Demonstration of Glass Panel Embedding for 3D System Packages for Heterogeneous Integration Applications

Abstract

Abstract This article demonstrates a next-generation high-performance 3D packaging technology with smaller form factor, excellent electrical performance, and reliability for heterogeneous integration. High-density logic-memory integration, today, is built predominantly using interposers which are fundamentally limited in assembly pitch and interconnect lengths, and they also are expensive as the package sizes increase. On the other hand, high-frequency applications continue to use laminates which are also limited by package size and ability to integrate many components. Wafer-level fan-out (WLFO) packaging promises better performance and form factor at lower costs, but current WLFO packages are mold-based and hence are limited to small packages. This article presents a 3D packaging technology using glass panel embedding (GPE) for high-performance with potential for large body size heterogeneous integration applications. The tailorable coefficient of thermal expansion of glass allows a reliable direct board attach of large GPE packages that not only benefits the form factor and signal speed but also provides radical benefits to power delivery. Unlike interposers and silicon bridges, GPE packages are not bump-limited and can support I/O densities comparable with backend-of-line with silicon-like redistribution wiring at much lower costs. The fundamental limitations such as die shift and poor dimensional stability of current organic WLFO packages are addressed by parametric process improvements to reduce die shift to <2 μm while also improving the RDL surface planarity for high-yielding fine-line structures and integrating through glass via (TGV) in the fan-out region for 3D packaging. This article describes the fabrication process for 3D GPE, leading to demonstration of a technology using embedding of chips with all-Cu interconnections at 40-μm I/O pitch with TGVs at 300-μm pitch, thus enabling double-side RDL and assembly of chips to achieve three levels of device integration.