NPO (Near-Packaged Optics)
Near-Packaged Optics (NPO) is an optical interconnect architecture in which optical engines are placed in close physical proximity to a switch ASIC, AI accelerator, CPU, GPU, or networking processor within the same system or package environment. The objective of NPO is to reduce the electrical trace length between high-speed silicon and optical transceivers, thereby improving bandwidth density, signal integrity, and power efficiency.
NPO is considered an intermediate architectural step between traditional pluggable optical transceivers and fully integrated Co-Packaged Optics (CPO) systems.
The technology is increasingly used in:
- AI infrastructure,
- hyperscale data centers,
- high-performance computing (HPC),
- Ethernet switching,
- and next-generation scale-out networking.
Background
As networking bandwidth has increased from:
- 100G,
- to 400G,
- 800G,
- 1.6T,
- and beyond,
traditional electrical interconnects based on copper traces and pluggable optics have encountered growing limitations in:
- signal loss,
- power consumption,
- PCB complexity,
- retimer requirements,
- and thermal management.
Modern AI clusters require massive interconnect bandwidth between:
- GPUs,
- AI accelerators,
- switches,
- and memory systems.
Electrical SerDes operating at 112G PAM4 and future 224G signaling increasingly suffer from insertion loss over long PCB traces. This has accelerated industry interest in bringing optics physically closer to the compute silicon.
Near-Packaged Optics emerged as a compromise architecture that provides many of the benefits of Co-Packaged Optics while avoiding some of the packaging and serviceability challenges of full optical co-packaging.
Architecture
General Concept
In an NPO system:
- the switch ASIC or accelerator remains packaged conventionally,
- while optical engines are mounted nearby on the same board, substrate, or interposer region.
Electrical connections between the ASIC and optics are significantly shortened compared to traditional pluggable modules.
Main Components
Switch ASIC or Accelerator
The primary compute or networking silicon device.
Examples:
- Ethernet switch ASICs,
- AI accelerators,
- GPUs,
- XPUs,
- HPC processors.
Optical Engines
Compact optical modules containing:
- photonic integrated circuits (PICs),
- modulators,
- photodetectors,
- laser interfaces,
- driver ICs,
- and transimpedance amplifiers (TIAs).
These engines often rely on:
- silicon photonics (SiPh),
- indium phosphide (InP),
- or hybrid photonic technologies.
Electrical Interfaces
High-speed electrical SerDes links connect the ASIC to the optical engine.
Typical signaling technologies include:
- 112G PAM4,
- 224G PAM4,
- future ultra-high-speed SerDes standards.
Fiber Connectivity
Optical fibers connect the optical engines to external network infrastructure.
Fiber attachment methods include:
- MPO connectors,
- fiber arrays,
- edge coupling,
- and grating couplers.
Related News
- NewPhotonics® Introduces Industry-First Serviceable Near-Packaged Optics (NPO) Chiplet Solution with Interoperability for 1.6T
- Crealights’ Self-Developed Silicon Photonics Chip “Greyhound” Makes Its Appearance at CIIE, Showcasing a “Chinese Solution for the Token Economy”
- Tower Semiconductor and Innolight Advance High-Performance Near-Package Optics for AI Infrastructure
- Eliyan Introduces NuLight™: World’s First 2D microVCSEL Interconnect Solution to the Copper Wall
- Coherent Launches PhotonLink™ Integrated Optics Platform for AI Infrastructure
Featured Content
- Starscream: Reimagining LLM Sharding on AMD Instinct GPUs with GPU Partitioning
- JEDEC® Announces Creation of JC-14.8 Subcommittee on System Level Reliability Testing (SLRT); Invites Industry Participation
- UCIe Consortium Advances from Chiplet Connectivity to System-Level Integration with UCIe 4.0 Specification
- Netherlands and Taiwan Launch Four Joint Photonics R&D Projects
- ASRA and imec release joint architectural requirements for interoperable automotive chiplets
- Crealights’ Self-Developed Silicon Photonics Chip “Greyhound” Makes Its Appearance at CIIE, Showcasing a “Chinese Solution for the Token Economy”
- Tri-Stack-PIM: A Heterogeneous Multi-Die 3D-Integrated Processing-in-Memory Architecture for Efficient LLM Inference
- CU-3D: A Scalable Legality-Aware RTL-to-GDS Flow for 3D-IC Physical Design
- Staera Semiconductor Emerges from Stealth with US$4.6 Million to Build Light into the Chip and Break AI's Data Bottleneck
- The chip beneath the chips: Why silicon interposers matter more than ever
- Vertical Compute expands partnership with imec to accelerate commercial manufacturing of Vertically Integrated Memory (VIM)
- Imec advances Europe's silicon interposer technology capabilities through NanoIC pilot line, paving the way for open chiplet design
- When Every Chiplet Passes, Who Fixes the Package?
- Beyond Host-Side Control: HBM Controller Partitioning Across Host and HBM Base Dies
- The European Chip Ecosystem