company blog about Supply Broadcom StrataDNX™ Switch Solutions:Jericho4,Jericho3,Jericho3-AI,Jericho2x,Jericho2c+

Supply Broadcom StrataDNX™ Switch Solutions:Jericho4,Jericho3,Jericho3-AI,Jericho2x,Jericho2c+

Shenzhen Mingjiada Electronics Co., Ltd., as an authorised independent distributor of world-renowned electronic components, leverages its many years of industry experience and stable supply chain system to provide customers with comprehensive electronic component solutions.

Supply Advantages

All products are genuine OEM items, backed by comprehensive quality assurance and traceability services.

Our inventory comprises over 2 million SKUs, covering the full range of electronic components.

Through large-scale procurement and optimised operational costs, we achieve industry-leading price competitiveness.

An efficient logistics system ensures on-time delivery.

Comprehensive after-sales service addresses any customer concerns.

I. Solution Overview: StrataDNX Jericho Product Line Positioning Framework

The Broadcom StrataDNX™ Jericho series is a high-end integrated routing and switching ASIC platform designed for carrier backbones, cloud data centres, intelligent computing (AI) clusters, multi-data centre interconnect (DCI), 5G transport, and large-scale campus and metropolitan networks, Its key differentiating features include ultra-deep buffers, carrier-grade hierarchical traffic scheduling, lossless long-distance transmission, modular horizontal scaling of the entire system, and a unified Elastic Pipe programmable packet processing architecture.

The product line forms a complete performance gradient: Jericho2c+ (entry-level high-end metropolitan transport) → Jericho2x (mid-range aggregation / edge DCI) → Jericho3 (native hyperscale cloud core) → Jericho3-AI (dedicated switching and routing for AI training clusters) → Jericho4 (ultimate platform for cross-region distributed AI interconnection). When paired with the Ramon series of switching matrix chips, these enable the construction of a global Ethernet architecture capable of delivering hundreds of Tbps per chassis and interconnecting millions of XPUs across the entire network, covering all network scenarios from metropolitan edges and local data centres to inter-city intelligent computing clusters.

II. Technical Specifications, Features and Applicable Scenarios of Each Core Chip Model

1. Jericho2c+ (BCM888xx, 7nm process, 14.4 Tbps)

Key Parameters

Single-chip system throughput of 14.4 Tbps; supports up to 36 × 400GE ports per chip; integrates line-rate MACsec encryption; supports FlexE, OTN direct connection and 5G network slicing; HBM high-capacity packet buffer, multi-protocol routing tables supporting tens of millions of entries; a single chip can realise a high-density 400G line card, whilst two chips support a full 36-port 400GE board, significantly reducing the number of hardware components, power consumption and rack space required for the entire system.

Key Technical Advantages

7nm advanced process technology, industry-leading power consumption per Tbps, suitable for compact 2U/5U box-type routers and modular edge line cards;

Comprehensive carrier-grade features: high-precision 1588v2/PTP synchronisation clock, SyncE, Layer 3 VPN, EVPN, Segment Routing, and high-throughput VoQ virtual output queues;

Compatible with 10G/25G/100G/400G full-speed Ethernet, designed for mobile backhaul, government and enterprise dedicated lines, campus core networks, and aggregation in small and medium-sized cloud data centres.

Typical Applications

5G mid-to-backhaul transport routers, carrier metropolitan area aggregation, core switches for large enterprise campuses, DCI internet gateways for small and medium-sized cloud providers, and broadcast and television video distribution backbones.

2. Jericho2x (BCM88830, 3.2 Tbps)

Key Specifications

Single-chip processing bandwidth of 3.2 Tbps; native SerDes support for 10G–400GE; integrated FlexE and Interlaken interfaces enabling direct connection to optical transmission equipment and baseband processors; features a hierarchical traffic manager with HBM deep caching to absorb micro-bursts, ensuring zero hardware-induced packet loss; Utilises a unified, programmable ‘Elastic Pipe’ data plane, supporting in-field protocol upgrades and enabling the iteration of business functions without the need for hardware replacement.

Key Technical Advantages

A lightweight mid-to-high-end DNX solution that balances bandwidth, cost and power consumption, designed primarily for edge access and small-to-medium-scale transport;

Comprehensive multi-tenant scheduling, supporting bandwidth throttling, traffic shaping and QoS tiering for millions of operator subscribers;

Can be horizontally stacked with Ramon switch chips to build metropolitan transport clusters at the 100 Tbps level.

Typical Applications

Operator edge PE routers, government and enterprise dedicated line access, 5G base station backhaul, campus aggregation switches, and low-cost interconnection between small remote data centres.

3. Jericho3 (5nm process, 51.2 Tbps)

Key Specifications

5nm process technology, 51.2 Tbps full-duplex throughput per chip, 144 x 106G PAM4 SerDes, flexible port configuration: 18×800GE / 36×400GE / 72×200GE; 160 Fabric interconnect channels interfacing with the Ramon3 matrix, enabling a single cluster to scale to tens of thousands of ports; Massive HBM packet buffer, lossless RoCEv2 hardware offload, and hardware-level network telemetry and traffic visualisation acceleration engine.

Core Technical Advantages

Native 800GE high-speed ports, designed for next-generation cloud data centre spine cores and high-performance computing (HPC) clusters;

Non-blocking multi-level switching architecture, providing seamless support for spine-leaf flat networks;

Hardware-accelerated congestion control, lossless Ethernet and high-traffic load balancing, eliminating micro-burst packet loss in AI and big data scenarios.

Typical Applications

Hyper-scale public cloud core Spine switches; backbone for on-premises single-campus clusters with tens of thousands of GPUs; high-performance computing (HPC); lossless interconnection of large storage pools; and the core of national-level scientific research and education networks.

4. Jericho3-AI (BCM88890, AI-dedicated chip, 28.8Tbps)

Key Specifications

A customised variant of the Jericho3 series for AI, offering 28.8 Tbps throughput, 144 channels of 106G PAM4 SerDes, and support for up to 18×800GE; optimised specifically for large-model training with a dedicated AI traffic scheduling unit, paired with the Ramon3 switching matrix, enabling a single network architecture to stably support parallel training across 32,000 GPUs; Integrated lossless RoCE, zero-outage lossless failover, and hardware-accelerated perfect ECMP traffic balancing.

Key Technical Advantages

Dedicated AI traffic load balancing: Addresses the challenges of mixed transmission of massive numbers of small flows and extremely long flows during large-model training, significantly reducing job completion time (JCT);

Deeply optimised congestion control algorithms, boosting link utilisation to over 90% under high loads without queue overflow or packet loss;

Built-in AI network visualisation telemetry, which collects real-time data on stream latency, jitter and congestion status, enabling rapid identification of network bottlenecks in computing clusters.

Typical Applications

Hyper-scale large language model / multimodal AI training clusters, spine/leaf nodes in intelligent computing centre networks, autonomous driving simulation computing pools, and distributed machine learning inference backbones.

5. Jericho4 (BCM99450, 3nm process, 51.2Tbps, flagship for cross-region distributed AI)

Key Specifications

The industry’s first 3nm StrataDNX chip, delivering 51.2 Tbps full-duplex bandwidth, featuring an exclusive HyperPort 3.2 Tbps aggregation interface (four 800GE ports bonded into a single logical port), with support for up to 36,000 HyperPorts per rack; Native support for 100GE to 1600GE full-speed ports, with full-line-rate hardware MACsec/IPSec encryption; lossless RoCE transmission distance exceeds 100 kilometres, enabling lossless computing interconnection between data centres across cities and provinces.

Core Technical Advantages

Patented HyperPort technology: Compared to traditional multi-link ECMP load balancing, link utilisation is increased by 70 per cent, significantly reducing load imbalance issues in multi-data centre interconnections;

The cornerstone of a cross-region distributed AI architecture: A single cluster can centrally orchestrate over 1 million GPUs/TPUs/XPUs, breaking through the power, space and computing capacity limitations of a single data centre;

3nm low-power process technology and hardware optimisation for long-distance lossless transmission, delivering comprehensive improvements in latency, packet loss and power consumption compared to previous generations in wide-area DCI scenarios;

Unified security acceleration, with line-rate encryption across all ports, ensuring trusted isolation throughout data transmission between cross-region intelligent computing clusters.

Typical Applications

Multi-region distributed AI super-clusters; interconnection of national-level intelligent computing hubs; backbone DCI for large cloud providers’ multi-active data centres; national-level backbone core routers for telecoms operators; and Ethernet transport for cross-border computing power orchestration.