Huawei Demonstrates 7.2T Near-Packaged Optics

Read original on Pandaily

#optical-networking#data-centers#ai-clustershuawei-7.2t-optical-modulehuawei

AI clusters need faster links, and Huawei is showing capacity beyond the emerging standard.

30-Second TL;DR

What Changed

Huawei claims the module is the industry’s first 7.2T near-packaged optics design.

Why It Matters

Higher-capacity optical interconnects could help data centers address bandwidth demands from AI clusters. However, interoperability and standards alignment will determine whether Huawei’s demonstrated capacity becomes broadly deployable.

What To Do Next

Ask your AI infrastructure vendor for a 200Gbps-per-lane roadmap and verify compatibility with emerging 6.4T optical standards.

Who should care:Enterprise & Security Teams

Key Points

•Huawei claims the module is the industry’s first 7.2T near-packaged optics design.
•The demonstration used 200Gbps per lane across 36 channels.
•The international standard under development is targeting 6.4T.
•The technology is aimed at high-bandwidth networking infrastructure.

Deep Insight

Background and context from public sources — not the original article. 8 sources cited.

Enhanced Key Takeaways

•The module eliminates the standalone optical Digital Signal Processor (oDSP) and integrates optical functions directly onto silicon photonics chips, cutting latency from 100ns to 10ns and power consumption by roughly 60%.
•Physical placement mounts the optical engine directly onto the switching ASIC's package substrate roughly 5 cm from the processor, substantially closer than standard 20 cm front-panel pluggables.
•The 7.2T NPO architecture serves as Huawei's 3rd-generation optical interconnect, following 1st-generation optics in the Atlas 910C SuperPoD and 2nd-generation Linear Drive Pluggable Optics (LPO) in the Atlas 950 SuperPoD.
•Huawei plans an open commercial deployment model, explicitly stating it will provide the 7.2T NPO modules to third-party cloud service providers and global supply chain partners rather than keeping it proprietary.
•Huawei and Chinese standards consortia (such as IPEC and CAICT) are prioritizing NPO over Co-Packaged Optics (CPO) favored by Western chipmakers like NVIDIA and Broadcom to ensure optical modules remain separable and field-replaceable.

Competitor Analysis

Packaging Architecture

Huawei 7.2T NPO: Near-Packaged Optics (substrate edge, ~5 cm)
Western CPO (e.g., Broadcom / NVIDIA): Co-Packaged Optics (direct co-substrate integration)
Standard Pluggable Optics (OIF / 6.4T): Front-panel pluggable (~20 cm from ASIC)

Aggregate Bandwidth

Huawei 7.2T NPO: 7.2 Tbps (36 × 200 Gbps)
Western CPO (e.g., Broadcom / NVIDIA): 3.2 Tbps - 6.4 Tbps (typical current targets)
Standard Pluggable Optics (OIF / 6.4T): 6.4 Tbps (OIF industry standard)

Field Serviceability

Huawei 7.2T NPO: High (modules remain separable and replaceable)
Western CPO (e.g., Broadcom / NVIDIA): Low (ASIC replacement risk upon optical engine failure)
Standard Pluggable Optics (OIF / 6.4T): High (hot-swappable transceivers)

DSP & Latency Profile

Huawei 7.2T NPO: ~10 ns latency (oDSP eliminated via silicon photonics)
Western CPO (e.g., Broadcom / NVIDIA): Ultra-low latency (direct optical engine drive)
Standard Pluggable Optics (OIF / 6.4T): ~100 ns latency (requires standalone oDSP)

Ecosystem Alignment

Huawei 7.2T NPO: IPEC / Open commercial offering
Western CPO (e.g., Broadcom / NVIDIA): Proprietary / Emerging CPO standards
Standard Pluggable Optics (OIF / 6.4T): Optical Internetworking Forum (OIF) / Tencent

Feature / Metric	Huawei 7.2T NPO	Western CPO (e.g., Broadcom / NVIDIA)	Standard Pluggable Optics (OIF / 6.4T)
Packaging Architecture	Near-Packaged Optics (substrate edge, ~5 cm)	Co-Packaged Optics (direct co-substrate integration)	Front-panel pluggable (~20 cm from ASIC)
Aggregate Bandwidth	7.2 Tbps (36 × 200 Gbps)	3.2 Tbps - 6.4 Tbps (typical current targets)	6.4 Tbps (OIF industry standard)
Field Serviceability	High (modules remain separable and replaceable)	Low (ASIC replacement risk upon optical engine failure)	High (hot-swappable transceivers)
DSP & Latency Profile	~10 ns latency (oDSP eliminated via silicon photonics)	Ultra-low latency (direct optical engine drive)	~100 ns latency (requires standalone oDSP)
Ecosystem Alignment	IPEC / Open commercial offering	Proprietary / Emerging CPO standards	Optical Internetworking Forum (OIF) / Tencent

Technical Deep Dive

Channel Architecture: Delivers 7.2 Tbps aggregate bandwidth across 36 physical optical channels operating at 200 Gbps per lane.
Substrate Proximity: Positioned approximately 5 cm from the switching ASIC on the edge of the package substrate, sharply reducing electrical trace lengths compared to traditional ~20 cm front-panel reaches.
Photonic Integration: Integrates optical transmission components onto silicon photonics dies, eliminating the standalone optical Digital Signal Processor (oDSP).
Latency Performance: Reduces signal transmission latency from 100 ns to 10 ns.
Thermal & Energy Efficiency: Lowers overall transceiver power consumption by roughly 60% compared to legacy DSP-driven pluggable optics.
Serviceability Design: Implements Near-Packaged Optics (NPO) packaging, preserving module modularity and field replaceability without requiring the disposal of host ASICs during optical engine failures.

Future ImplicationsAI analysis grounded in cited sources

Huawei's 7.2T architecture will accelerate divergence from Western 6.4T networking standards.

By pushing 7.2T silicon photonics into commercial channels, Huawei incentivizes regional data centers to bypass 6.4T OIF standards in favor of higher-density channel configurations.

Near-term hyperscale AI clusters will favor separable NPO over monolithic CPO.

NPO offers substantial power and latency reductions while avoiding the prohibitive yield loss and unserviceability associated with co-packaged optics.