Laser Chip Brings Multiplexing to AI Data Centers

2026-03-16 · Source: IEEE Spectrum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

Tower Semiconductor and Scintil Photonics have announced the production of the world's first single-chip Dense Wavelength Division Multiplexing (DWDM) light engine designed for AI infrastructure. This innovation addresses the critical need for optical scale-up networking in AI data centers, where high bandwidth and low latency are paramount for connecting dozens of GPUs and memory as a single entity. Unlike traditional co-packaged optics (CPO) that often integrate discrete optical components, this new solution integrates the laser itself onto a silicon wafer. Scintil's "SHIP" (Scintil Heterogeneous Integrated Photonics) technology uses a 300-millimeter silicon photonics wafer and precisely bonds tiny InP/III-V semiconductor dies to create eight distributed feedback lasers, enabling multiple wavelengths per fiber. This allows for a "slow and wide" architecture, transmitting 50 Gb/s over 8 channels to achieve up to 1.6 terabit per second data speeds per fiber, significantly improving power efficiency and reducing latency in scale-up AI networks.

Key takeaway

For CTOs and VPs of Engineering evaluating next-generation AI infrastructure, the advent of single-chip DWDM light engines from Tower Semiconductor and Scintil Photonics signals a critical shift. This technology directly addresses the escalating power and latency demands of scale-up AI clusters, enabling significantly higher GPU utilization and data throughput. You should consider integrating these advanced co-packaged optics solutions into your future data center designs to optimize performance and energy efficiency, with initial units expected by late 2026.

Key insights

Integrating DWDM lasers directly onto silicon wafers revolutionizes AI data center networking by enhancing bandwidth and reducing latency.

Principles

• Optical networking is crucial for AI data center scale-up.
• DWDM improves data capacity and power efficiency.
• Low latency is critical for GPU utilization in AI clusters.

Method

Scintil's SHIP technology integrates lasers, photodiodes, and modulators onto a 300-millimeter silicon photonics wafer by bonding InP/III-V semiconductor dies and using photolithography to etch diffraction gratings for distributed feedback lasers.

In practice

• Deploy DWDM for scale-up AI networking.
• Utilize multi-wavelength lasers for higher data capacity.
• Prioritize low-latency interconnects for GPU clusters.

Topics

• AI Data Centers
• Optical Networking
• DWDM
• Silicon Photonics
• Co-Packaged Optics

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, AI Architect, AI Hardware Engineer

Related on AIssential

• This Will Power Everything — Optical interconnects are critical for scaling AI data centers beyond copper's physical and thermal limitations.
• Connectivity Revolution or Evolution Inside Data Centers? — AI workloads necessitate a revolutionary shift to optical, scale-out intra-data center networks, integrating optics closer to compute.
• Scaling Power-Efficient AI Factories with NVIDIA Spectrum-X Ethernet Photonics — Co-packaged optics in Ethernet switches significantly enhance AI factory scalability, power efficiency, and network resilience.
• Photonics: A Foundational Scaling Layer for AI-Era Computing — Photonics is becoming the foundational layer for scaling AI computing by addressing system-level data movement and memory bottlenecks.
• All-optical signal processor developed to break AI data center transmission bottleneck — An all-optical signal processor addresses AI data center transmission bottlenecks by eliminating O/E/O conversions.
• Coherent Breaks Ground on Expanded Texas Facility, Scaling AI’s Optical Backbone — Optical interconnects, leveraging 6-inch indium phosphide wafers, are fundamental for scaling high-performance AI infrastructure efficiently.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.