US-Based Broadcom’s 3.5D Chips: The Future of AI!
Broadcom has introduced its groundbreaking 3.5D eXtreme Dimension System in Package (XDSiP) platform, designed to support the growing demand for generative AI (GenAI) chips. This new technology integrates up to 6000 mm² of silicon and 12 high-bandwidth memory (HBM) stacks into a single package, enhancing efficiency and reducing power consumption for large-scale AI applications. The rapid evolution of large language models (LLMs) and the increasing computational power needed for their training have made advanced system integration essential, and Broadcom’s 3.5D platform takes a major leap beyond traditional methods.
Unlike conventional Face-to-Back (F2B) stacking, Broadcom’s 3.5D approach uses a Face-to-Face (F2F) method, which connects the top metal layers of the dies. This innovation leads to a sevenfold increase in signal density and a tenfold reduction in power consumption at the die-to-die interfaces. The result is improved mechanical strength and minimal electrical interference, making it an ideal solution for the next generation of AI accelerators. The smaller interposer and package sizes also contribute to cost savings and enhanced stability, offering a more compact and efficient platform.
The 3.5D XDSiP platform integrates four compute dies, one I/O die, and six HBM modules, leveraging advanced processes from Taiwan Semiconductor Manufacturing Company (TSMC). Broadcom’s proprietary design flow, based on industry-standard tools, ensures that the platform is highly reliable and performs seamlessly, despite the complex integration of these cutting-edge technologies. This platform will significantly boost the performance of AI accelerator clusters, helping to meet the ever-growing demand for GenAI processing power.
Broadcom’s 3.5D XDSiP platform has already gained attention from major industry players. Fujitsu, for example, plans to integrate the technology into its next-generation 2-nanometer Arm-based processor, FUJITSU-MONAKA, to deliver higher performance with lower power consumption. With over five products in development and production shipments set to begin in February 2026, Broadcom is well-positioned to lead the future of AI accelerator technologies, responding to the soaring need for more powerful, efficient computing solutions in artificial intelligence.
