Skip to main content

Broadcom Launches Jericho 4: A Breakthrough 3nm Ethernet Chip

·327 words·2 mins
Broadcom 3nm Jericho 4 Distributed AI
Table of Contents

On August 4, 2025, U.S. semiconductor giant Broadcom officially introduced its latest Ethernet routing chip, Jericho 4. Built using TSMC’s cutting-edge 3nm process, Jericho 4 is engineered to support massive, cross-data-center AI infrastructure, boasting 200G PAM4 signaling and an innovative 3.2Tbps HyperPort for ultra-high-speed connectivity.


🔧 Key Features & Innovations
#

⚙️ Advanced 3nm Process
#

Leveraging TSMC’s 3nm manufacturing, Jericho 4 achieves superior integration and efficiency. The result is increased bandwidth and reduced power consumption for next-gen network systems.

🔗 Massive Interconnect Capacity
#

Each Jericho 4-based system can connect over 1 million XPUs, enabling large-scale, collaborative computing across multiple data centers—multiplying the capabilities of previous solutions.

🚀 Industry-Leading Bandwidth
#

With 51.2 Tbps total throughput, the chip introduces 3.2T HyperPort technology, merging four 800GE links into a single logical port. This eliminates traditional load balancing inefficiencies and boosts link utilization by up to 70%.

Broadcom 3nm Jericho 4

🌐 Long-Distance Lossless Transmission
#

Through deep buffering and smart congestion management, Jericho 4 enables lossless RoCE (RDMA over Converged Ethernet) over distances up to 100 km—crucial for distributed AI deployments that span across geographies.

Broadcom 3nm Jericho 4

🔒 End-to-End Encryption at Wire Speed
#

Each port is equipped with MACsec encryption, ensuring secure data transfers between data centers—without compromising throughput even under full load.

✅ UEC Standard Compliance
#

Jericho 4 is fully aligned with the Ultra Ethernet Consortium (UEC) standard, ensuring seamless compatibility with UEC-certified NICs, switches, and software environments.

Broadcom 3nm Jericho 4


🧠 The Complete Ethernet AI Platform
#

With Jericho 4, Broadcom completes its AI-optimized Ethernet solution stack:

  • Jericho 4: High-speed interconnect across data centers.
  • Tomahawk 6: 100Tbps backbone for intra-data center clusters.
  • Tomahawk Ultra: Sub-400ns ultra-low latency for inter-XPU communication.

Broadcom 3nm Jericho 4


📦 Market Availability & Outlook
#

Sampling of Jericho 4 began in early August, with broad commercial deployment expected within the next nine months. As AI workloads scale, Jericho 4 is positioned to become a cornerstone of secure, high-performance, distributed AI networks, marking a shift toward more distributed, user-proximate data center architectures.

Broadcom 3nm Jericho 4

Related

博通Emulex Secure HBA:后量子加密与零信任架构的存储网络革命
·140 words·1 min
Broadcom HBA Quantum
Broadcom Introduces 3 5d Xdsip Chip Packaging for Data Center Processors
·19 words·1 min
Broadcom 3.5D XDSip
OpenAI to Launch Its First Custom AI Chip with Broadcom and TSMC
·465 words·3 mins
OpenAI Broadcom TSMC AI Chip Custom AI Hardware Inference