Jericho2 switch-router silicon was presented as a way to combine high-capacity forwarding, flexible packet processing and carrier-grade traffic management in a merchant silicon platform. Announced in 2018, the architecture targeted service provider and data center networks that needed to scale bandwidth while supporting varied protocols, traffic profiles and deployment models. Its design offers a useful example of how programmable packet pipelines, deep buffering and modular systems can work together in network infrastructure.
Jericho2 switch-router silicon and the move to higher capacity
Broadcom described Jericho2 as a 10 Tb/s switching device, with 208 50G PAM-4 SerDes and support for network interfaces ranging from 10GE through 400GE. The presentation also described a chipset approach that could scale beyond 200 Tb/s in a single modular system, using Jericho2 together with the Ramon switch fabric. These figures describe the announced architecture and its design targets, rather than a guarantee of performance in every system configuration.
The broader change was the growing role of merchant silicon in carrier networks. Rather than relying exclusively on custom application-specific integrated circuits (ASICs) designed for individual platforms, equipment makers could use commercial switching and routing silicon as a foundation for their systems. The stated aim was to give original equipment manufacturers (OEMs) and service providers more options for building high-capacity routers, including whitebox and disaggregated designs.
Elastic Pipe: adapting packet processing to network needs
A central feature of the Jericho2 packet processor was Elastic Pipe. The DNX packet-processing architecture uses programmable stages to handle packets as they move through the device. Jericho2 was designed to let a system extend that pipeline by drawing on a pool of additional general-purpose stages. A centralized database could also be flexibly bound to different stages, rather than being fixed to one place in the pipeline.
This approach addresses a practical challenge for network equipment designers: service provider, aggregation, VPN and data center applications do not all require the same forwarding-table scale or processing sequence. Jericho2’s database sizing could be adjusted through profiles, while external lookup expansion using a knowledge-based processor (KBP) was presented as another way to increase scale.
Broadcom also positioned programmability as a means of accommodating changing protocols and requirements. The presentation described a C++ development environment for qualified customers, with new datapath programs integrated into the Broadcom software development kit and regression-tested. In principle, this gives equipment developers a way to adapt packet processing without treating every protocol change as a reason to redesign the full silicon pipeline. The actual development and validation work remains an important part of deploying such changes.
Buffering and traffic management for congestion
Buffer behavior is a critical consideration when traffic arrives faster than an output can transmit it. Jericho2’s distributed buffering design combined on-chip buffer memory (OCB) with high-bandwidth memory (HBM) in-package expansion. The stated operating model used on-chip memory for normal traffic conditions and assigned congested flows additional capacity in HBM. The presentation described an 8 GB HBM configuration and a transition between OCB and HBM based on congestion.
Queues are organized as virtual output queues (VOQs), allowing traffic to be held by its intended egress destination rather than requiring all packets to wait in a shared output queue. The architecture’s traffic manager was described as hierarchical, with programmable scheduling and service meters. Together, queueing, buffer assignment and scheduling are intended to help manage bursts and congestion across different flows. The source presentation characterized the design as preventing packet drops; in real deployments, buffer outcomes depend on traffic conditions and system configuration, so that statement should be understood as the design objective rather than an unconditional result.
Why the fabric architecture matters
The material contrasts an Ethernet fabric approach, in which traffic can converge on an egress buffer, with a DNX fabric design that stores ingress traffic in VOQs and lets the egress port pull packets as capacity becomes available. In the latter model, buffering is distributed across ingress devices. This can increase the aggregate buffering available to handle incast, where multiple sources send traffic toward one destination at the same time. The comparison highlights how switching fabric and queue placement can affect congestion behavior—not only the nominal size of an individual buffer.
System design, interfaces and deployment choices
Jericho2 was presented as part of a modular system rather than as an isolated chip. The announced configuration included a fabric interface and a network-facing interface, with support for multiple Ethernet rates up to 400GE. The wider StrataDNX system was described as supporting up to 9.6 Tb/s per device and scaling in a modular chassis, with backward and forward compatibility across generations cited as a design objective.
The platform also included functions relevant to carrier operations, including operations, administration and maintenance (OAM), service activation testing (SAT), instrumentation, telemetry and flexible counters. Routing and packet-processing support listed in the presentation covered technologies such as IPv4 and IPv6, MPLS, VPLS, L2VPN, L3VPN and VXLAN. These capabilities matter to system integrators because switching capacity alone does not determine whether a platform fits a network: protocol coverage, diagnostics, traffic handling and integration with the fabric all shape the implementation.
Broadcom also described Ramon as a fabric option intended to support chassis systems, with credit-based VOQs and cell-based switching identified as mechanisms for addressing incast and bandwidth transitions. The presentation proposed this architecture for service provider core and aggregation networks as well as data center roles, including spine-leaf, data center interconnect and core applications. Those are target use cases, not evidence that one configuration is appropriate for every topology.
Practical implications for network decision-makers
Jericho2’s significance lies in how its components address related design problems: Elastic Pipe provides a programmable processing structure; modular databases address differing table-scale needs; OCB and HBM provide a tiered buffering model; and VOQ-based fabric behavior supports distributed traffic handling. Together, these ideas aim to give equipment developers options for building systems across multiple network segments.
For engineers and technical decision-makers evaluating merchant silicon, the practical lesson is to assess the complete system architecture. Consider the required interface rates, forwarding scale, congestion patterns, buffer behavior, protocol support, fabric design and software integration—not just peak switching capacity. Jericho2’s 2018 design illustrates how those factors can be brought together to support scalable carrier and data center platforms, while leaving system selection and validation dependent on the needs of the deployment.