Broadcom Jericho 4 Architecture
The Broadcom Jericho 4 is the fourth generation of Broadcom's StrataDNX deep-buffer carrier routing ASIC family. It targets next-generation IP core, provider edge, and AI/ML back-end fabric deployments at 400G–1.6T per port, extending the two-tier OCB + HBM buffer model of the Jericho line into the 1+ Tbps per NPU era.
Data quality note: As of mid-2025, Jericho 4 had been referenced in Broadcom roadmap communications and some vendor announcements, but detailed silicon specifications (exact OCB size, HBM generation and capacity, exact per-NPU capacity) had not been confirmed in a public, citable Broadcom datasheet. The figures below reflect what has been disclosed in public Broadcom and vendor communications; figures marked as inferred are based on the generational scaling pattern of the Jericho family.
Overview
| Field | Value |
|---|---|
| ASIC | TBD (Jericho 4 / BCM8xxxx) |
| Target Capacity | ~6.4 Tbps full-duplex per NPU (inferred) |
| On-Chip Buffer (OCB) | Not publicly confirmed |
| Off-Chip Buffer | HBM3e or HBM4 (generation not confirmed); capacity not confirmed |
| Buffer Model | VOQ (Virtual Output Queueing) |
| Process Node | TSMC 3 nm or N3E (not confirmed) |
| Interface Support | 800G, 1.6T (targeted) |
Architecture Continuity
Jericho 4 is expected to continue the defining architectural characteristics of the StrataDNX line:
Two-Tier Buffer (OCB + HBM)
Every Jericho generation has shipped with:
- A small, fast on-chip SRAM buffer (OCB) handling non-congested traffic at full line rate
- A large off-chip HBM pool absorbing sustained congestion bursts
The pattern across generations:
| Generation | OCB | HBM | Capacity |
|---|---|---|---|
| Jericho (BCM88670) | ~16 MB | GDDR5 (~4–8 GB) | 0.8 Tbps FD |
| Jericho 2 (BCM88690) | ~32 MB | 8 GB HBM2/2e | 1.6 Tbps FD |
| Jericho 3 (BCM88830) | ~48 MB | Up to 32 GB HBM3 | 3.2 Tbps FD |
| Jericho 4 | TBD | TBD (HBM3e/HBM4) | ~6.4 Tbps FD (inferred) |
If the OCB growth tracks roughly with the doubling of capacity per generation, a Jericho 4 OCB in the range of 64–96 MB would be consistent with the pattern, though this is speculative until Broadcom publishes confirmed figures.
Virtual Output Queueing (VOQ)
VOQ scheduling is a defining characteristic of the StrataDNX family and is expected to be retained in Jericho 4. See the Jericho architecture overview for VOQ mechanics.
Why Jericho Matters Alongside Tomahawk
The Tomahawk and Jericho families address fundamentally different use cases:
| Tomahawk 5 (BCM78900) | Jericho 3 (BCM88830) | |
|---|---|---|
| Target | Hyperscale spine, AI fabric | Carrier IP core, provider edge |
| Buffer type | On-chip SRAM only | OCB + HBM (two tiers) |
| Max buffer | 165.2 MB | ~32 GB (HBM) |
| Buffer model | Shared pool | VOQ (per-egress queues at ingress) |
| HBM | None | Yes |
| Routing | Limited L3 | Full carrier-grade IP/MPLS |
| Use case | Lossless AI fabric | Large-scale IP core transit |
Jericho 4 extends the lead in total buffer capacity, sustained congestion absorption, and carrier-grade routing depth, at the cost of additional latency and complexity compared to Tomahawk's simpler shared-memory model.
References
- Broadcom StrataDNX Family Overview
- Broadcom Jericho Architecture — Jericho and Jericho 2 details
- Broadcom Jericho 3 Architecture — confirmed J3 specifications
See Also
- Broadcom Jericho 3 Architecture — predecessor with confirmed figures
- Back to the main buffer table