Abstract:Advanced packaging technologies such as 2.5D Si TSV interposer, Fanout RDL interposer and 3D hybrid bonding packaging have been developed for chiplets and system heterogeneous integrations to fulfil the continuous pursuit of higher performance, higher bandwidth, lower power consumption, higher capacity and lower cost. However, the new heterogeneous integration solutions especially for HPC, AI and edge computing applications have also urged the demands for the laminated substrates with higher density interconnects and larger footprint body sizes. Conventional laminated substrate such as flip-chip ball grid array (FCBGA) package with multi-layer (typically over 10 layers) will suffer higher cost from the yield loss with the increasing substrate layer counts and body size.In this paper, the new packaging solutions of combining high density fanout with laminated substrate were developed. Two different high density fanout (FO) solutions named chip first and chip last for a single ASIC chip on standard ABF laminated substrates with different body sizes and layer counts were introduced. This new packaging technology was called Fanout BGA (FOBGA). The passive components were also embedded on redistribution layer (RDL) for FO chip last hybrid substrate. The test vehicles and process flows for both FOBGA chip first and chip last solutions were also elaborated. The warpage of different TVs was also demonstrated. Furthermore, the package signal integrity (SI) and power integrity (PI) performance for FO on substrate were conducted by electrical stimulations. The impacts of RDL routing and RDL line and spacing (L/S) on electrical performance of FOBGA package were also discussed. Finally, the design guidance of FO on substrate to improve the performance by reducing substrate layer counts were proposed. The high density FO on substrate has provided the higher density interconnect than that of conventional laminated substrate, it also has demonstrated the flexibility to reuse substrate and good potential to leverage standard laminate substrate without increasing body size and layer counts for high performance applications.

AMD InstinctTM MI250X Accelerator enabled by Elevated Fanout Bridge Advanced Packaging Architecture

Advanced Packaging: Enabling Moore’s Law’s Next Frontier Through Heterogeneous Integration

DaDianNao: A Machine-Learning Supercomputer

Porting HPC Applications to AMD Instinct$^\text{TM}$ MI300A Using Unified Memory and OpenMP

Advanced Fanout Packaging Technology for Hybrid Substrate Integration

A Scalable Network-on-Chip Microprocessor With 2.5D Integrated Memory and Accelerator.

Modular High-Performance Computing Using Chiplets

FPGA-based Multi-Chip Module for High-Performance Computing

DeACT: Architecture-Aware Virtual Memory Support for Fabric Attached Memory Systems

A 10TFLOPS Datacenter-Oriented GPU with 4-Corner Stacked 64GB Memory by the Means of 2.5D Packaging Technology

(Invited, Digital Presentation) Overview of Current and Future 3D Heterogeneous Integration Architectures

EMIB and advanced substrate packaging technologies to enable heterogeneous integration (HI) applications

OpenACC offloading of the MFC compressible multiphase flow solver on AMD and NVIDIA GPUs

Execution‐Cache‐Memory modeling and performance tuning of sparse matrix‐vector multiplication and Lattice quantum chromodynamics on A64FX

A Digital-Optical System-on-package Module Architecture for High Performance Multiprocessors

A Scalable Multi-Chiplet Deep Learning Accelerator with Hub-Side 2.5D Heterogeneous Integration.

Sapphire Rapids: The Next-Generation Intel Xeon Scalable Processor

Arvon: A Heterogeneous System-in-Package Integrating FPGA and DSP Chiplets for Versatile Workload Acceleration

Performance Portable Graphics Processing Unit Acceleration of a High-Order Finite Element Multiphysics Application

Understanding Data Movement in AMD Multi-GPU Systems with Infinity Fabric

The evolution of heterogeneous integration and packaging for the age of chiplets