Broadcom Revolutionizes AI Processing with 3.5D Chiplet Technology

2024-12-10 industry

None specified, Tuesday, 10 December 2024.
Broadcom’s new 3.5D XDSiP platform enhances AI processing with up to 6000 mm² silicon integration, offering 7x signal density and 10x power efficiency improvements. Products are set for 2026 release.

Groundbreaking Architecture

Announced on December 5, 2024, Broadcom’s 3.5D eXtreme Dimension System in Package (XDSiP™) platform represents a significant leap in chip packaging technology [1][2]. The revolutionary design integrates over 6,000 mm² of silicon and accommodates up to 12 high-bandwidth memory (HBM) stacks [1][3]. At its core, the platform combines traditional 2.5D chiplet structure with advanced 3D stacked die and face-to-face (F2F) configuration, optimizing both performance and thermal management [1].

Technical Advantages

The face-to-face configuration delivers remarkable improvements over conventional face-to-back (F2B) approaches. Key performance metrics include a sevenfold increase in interconnect signal density and a 90% reduction in die-to-die power consumption through hybrid copper bonding (HCB) technology [2][3]. The architecture integrates four compute dies, one I/O die, and six HBM modules, utilizing TSMC’s cutting-edge process nodes and CoWoS® packaging technologies [2]. As Frank Ostojic, Senior Vice President at Broadcom, explains, ‘Advanced packaging is critical for next generation XPU clusters as we hit the limits of Moore’s Law’ [5].

Industry Collaboration and Validation

The development represents a strong collaboration between industry leaders. Dr. Kevin Zhang, Senior Vice President at TSMC, affirmed their partnership: ‘TSMC and Broadcom have collaborated closely over the past several years to bring together TSMC’s most advanced logic processes and 3D chip stacking technologies with Broadcom’s design expertise’ [6]. The platform has already gained traction with major players, including Fujitsu, whose Senior Vice President Naoki Shinjo highlighted the technology’s potential for their next-generation 2-nanometer Arm-based processor [2].

Future Impact and Deployment

The timing of this innovation is crucial, as generative AI applications demand unprecedented processing power, requiring clusters of 100,000 to over one million processing units [2][6]. Broadcom currently has more than five 3.5D products in development [5], with production shipments scheduled to begin in February 2026 [2][6]. This technology platform is positioned to address the growing demands of AI data centers while offering significant improvements in performance, efficiency, and cost [1][2].

Bronnen

semiconductors chiplets