Broadcom’s silicon division has released its next-generation Jericho networking chip, designed to link data centres more than 96 kilometres apart and improve the speed of AI processing.
The new chip, called Jericho4, introduces upgrades aimed at moving more traffic in large-scale networks that connect facilities in and between data centres. With AI models becoming increasingly complex, building and training them now requires linking thousands of graphics processing units (GPUs). Cloud providers like Microsoft and Amazon need faster, more capable networking chips to keep data flowing efficiently.
Transferring data between data centres also raises security concerns, as information is more vulnerable when it travels beyond a facility’s physical boundaries. Broadcom says Jericho4 addresses this by supporting encryption on every port without slowing performance, even under heavy traffic.
According to Ram Velaga, senior vice president and general manager of Broadcom’s Core Switching Group, the chips are designed for massive-scale deployment, with a single system using around 4,500 Jericho4 chips. To help handle congestion, the chip includes high-bandwidth memory (HBM) – the same type used in AI processors from Nvidia and AMD – allowing it to store traffic temporarily until network conditions improve.
“The switch is actually holding that traffic (in memory) till the congestion frees up,” Velaga said. “It means you need to have a lot of memory on the chip.” Longer distances between facilities also require more onboard memory.
Jericho4 is built using Taiwan Semiconductor Manufacturing Company’s 3-nanometre process. A single system can scale to 36,000 HyperPorts, each operating at 3.2 terabits per second. It supports deep buffering, line-rate MACsec encryption, and RDMA over Converged Ethernet (RoCE) transport over distances beyond 100 kilometres.
The chip’s HyperPort technology combines four 800-gigabit Ethernet links into one logical port, removing inefficiencies from load balancing and improving use by as much as 70%. This also simplifies how traffic is managed in large-scale networks. With buffering and congestion control, Jericho4 can support lossless RoCE over extended distances, making it possible to build distributed AI systems without being limited to the power or space of a single location.
Security is a central feature, with MACsec encryption running at full speed on every port. Broadcom says this ensures data moving between data centres remains protected without a performance trade-off.
The chip also features Broadcom’s 200G PAM4 SerDes technology, offering long reach without requiring extra components like retimers, which helps cut power use, lowers costs, and improves reliability. It meets Ultra Ethernet Consortium (UEC) specifications, ensuring it works with other UEC-compliant network interface cards, switches, and software.
Jericho4 joins Broadcom’s Tomahawk 6 and Tomahawk Ultra in the company’s networking lineup.
“As AI models grow in size and complexity, the infrastructure requirements exceed the power and physical limits of a single data centre,” the company said. “Distributing XPUs (diverse processor types) in multiple facilities – each provisioned with tens to hundreds of megawatts of power – requires a new class of router, optimised for very high-bandwidth, secure and lossless transport in regional distances.”
On the day of the launch, Broadcom’s stock rose 3.2% to close at 297.72.
(Photo by Compare Fibre)
See also: Broadcom’s latest Ethernet switch tackles AI demands on networks
Looking to revamp your digital transformation strategy? Learn more about Digital Transformation Week taking place in Amsterdam, California, and London. The comprehensive event is co-located with IoT Tech Expo, AI & Big Data Expo, Cyber Security & Cloud Expo, and other leading events.
Explore other upcoming enterprise technology events and webinars powered by TechForge here.
