Broadcom Ships First 2nm 3.5D Compute SoC for AI Clusters
Event summary
- Broadcom began shipping its first 2nm custom compute SoC built on the 3.5D eXtreme Dimension System in Package (XDSiP) platform.
- The technology combines 2.5D techniques with 3D-IC integration using Face-to-Face (F2F) technology for high-density, low-power computing.
- Broadcom's XDSiP platform enables independent scaling of compute, memory, and network I/O in a compact form factor.
- The SoC is designed to meet the computational demands of gigawatt-scale AI clusters with superior power efficiency and low latency.
The big picture
Broadcom's shipment of its first 2nm 3.5D compute SoC underscores the growing demand for advanced semiconductor solutions tailored to AI workloads. The technology addresses critical industry needs for high-density, low-power computing, positioning Broadcom as a key player in the next generation of AI infrastructure. This development comes amid broader trends toward modular and multi-dimensional chip designs aimed at enhancing performance and efficiency in data-intensive applications.
What we're watching
- Technology Leadership
- How Broadcom's 3.5D XDSiP technology will influence the competitive landscape for AI infrastructure solutions.
- Customer Adoption
- The pace at which other customers beyond Fujitsu adopt Broadcom’s 3.5D platform for XPU development.
- Scalability Challenges
- Whether Broadcom can sustain the high-complexity manufacturing and integration required for gigawatt-scale AI clusters.
Our editorial coverage:
Related topics