AI Photonics – creating the optical interconnects of tomorrow

ams OSRAM wide-and-slow architecture for scale-up, scale-out and future chip-to-chip use cases in AI data centers.

Micro-array technology for higher-density AI interconnects

Can your datacom architecture keep pace with rising AI bandwidth demands without adding power, heat and complexity? As copper and fast-and-narrow optical links have their practical limitations, another approach is needed.
ams OSRAM is developing wide-and-slow optical interconnects based on 2D micro-emitter arrays. Instead of pushing more data through a small number of complex high-speed channels, this architecture distributes data across a larger number of parallel, lower-speed channels. The result: a new optical link architecture that simultaneously provides reach and low latency, while maintaining cost and power efficiency.
How do we achieve this? Our vision of AI Photonics is set to combine four building blocks to provide the complete and optimized photonics engine:

Emit: micro-emitter arrays emitting light signals with high bandwidth density
Detect: micro-photo-diode arrays, converting light back into electric data and support signal integrity
Process: mixed-signal CMOS ICs (like AFE) processing, conditioning, and converting data to digital.
Integrate: advanced packaging brings the optical and electronic elements together.

Are you developing HPC or AI data center interconnects? Contact our team to explore how ams OSRAM’s micro-emitter, mixed-signal IC and photo-diode expertise could support your next architecture.

Contact the optical interconnect team

Build more efficient AI interconnects with integrated photonics

How can you increase bandwidth density without increasing power, heat and system complexity? ams OSRAM helps you explore a wide-and-slow approach for AI and HPC data center connectivity. Our upcoming micro-emitter arrays, micro-photodiode arrays and mixed-signal CMOS ICs, are designed to work together as integrated optical building blocks. This gives you a foundation for developing efficient, reliable and scalable links for scale-up, scale-out and chip-to-chip use cases. Whether you design optical transceiver modules, interconnect architectures or data center systems, you can work with ams OSRAM to assess and model how these technologies fit your requirements.

Solving the bandwidth density bottleneck with optical interconnects

Traditional copper-based links are hitting their bandwidth density and reach limit just as bigger AI scale-up domains require exactly that. Our 2D micro-emitter arrays enable highly parallel optical links, delivering significantly higher bandwidth density while enabling compact form factors.

Improving power efficiency and system-level performance

Fast‑and‑narrow optical links based on high‑speed emitters require higher signal-to-noise requirements and increased energy overhead.. By using many lower‑speed emitters in parallel, our wide-and-slow architecture relaxes signal integrity requirements, it simplifies electronics and enables lower energy per bit.

Low latency and latency jitter

A key benefit of the wide-and-slow architecture is low latency and minimal latency jitter – both critical for tightly synchronized AI training systems. While fast-and-narrow links use complex Digital Signal Processing (DSP) that introduces variable processing delays and jitter, wide-and-slow significantly reduces processing stages and retiming overhead cost, ensuring predictable and stable packet delivery times.

Enabling higher reliability through redundancy and robustness

Single-emitter or low-count architectures are vulnerable to failures that can disrupt entire links. The inherent redundancy of 2D micro-emitter arrays allows continued operation despite individual emitter degradation or failure. The material systems used for our wide-and-slow links have proven reliability in automotive and industrial applications for decades, supporting robust operation in demanding environments.

Enabling greater reach and flexibility vs. copper

Future AI data centers will be based on larger scale-up up fabrics stretching across multiple rows. This drives the need for greater reach to give AI infrastructure architects more flexibility. As copper-based interconnects hit their limit, ams OSRAM’s approach increases reach by up to 10x.

Attractive TCO compared to alternative solutions

The wide-and-slow architecture offers an attractive value proposition compared to solutions like copper or fast-and-narrow. The price per bandwidth (EUR/Gbps) is closer to that of copper while providing major benefits in reach and bandwidth. Compared to fast-and-narrow the price per bandwidth is expected to be lower.

Proven high-yield, scalable production

Manufacturing and packaging partners require high yield, scalable production with clear process windows and test strategies. ams OSRAM is pioneering the development and high-volume manufacturing of micro-emitters. , One proof point is the volume shipments of EVIYOS products to the automotive industry since 2023.

A forward-looking platform for next-generation architectures

The industry is approaching the physical and economic limits of speed only scaling. Our approach shifts the scaling paradigm from ‘faster’ to ‘more parallel,’ creating a future proof optical interconnect platform that aligns with emerging chiplet and AI system architectures. The platform scales also with array density rather than only with per lane performance, offering a longer and more sustainable roadmap.

Why choose ams OSRAM?

Pioneering wide-and-slow architecture for next gen optical interconnects

Bigger AI scale-up domains need interconnects designed for power, latency, bandwidth density and reliability together. ams OSRAM’s wide-and-slow architecture distributes data across dense, parallel optical channels to provide the reach and predictable low latency that larger AI fabrics require, while maintaining cost and power efficiency.

All photonics engine building blocks under one roof

ams OSRAM brings together the building blocks of a photonics engine: micro-emitter arrays, micro-photodiode arrays, mixed-signal CMOS ICs, optics and advanced packaging. This integrated approach supports system-level optimization of emission, detection, processing and integration.

As a global manufacturer with established semiconductor, emitter and sensor production expertise, ams OSRAM is positioned to address the ultra-high-volume requirements of AI scale-up interconnects. Our quality approach builds on proven manufacturing methods, automotive-grade integration experience and a focus on reliability, manufacturability and supply continuity.

Are you working on the next generation of AI data center connectivity? Contact the ams OSRAM optical interconnect team to discuss your requirements and explore a collaboration to jointly develop the optical interconnects of tomorrow.

Contact our team

Optical interconnect support request

表单加载

Please fill out the form and enter your data.

I agree to the processing of my data for registration purposes, as described in privacy policy.*

I would like ams OSRAM to use my e-mail address for the purposes of personalised and generalized advertising and marketing communication related to the products of ams OSRAM.

Your data is used only for internal purposes related to the registration process.

All fields marked with an asterisk(*) must be filled out.

MicroLED data transmitters in AI servers

Simulation model sheds new light on potential performance gains.

microLEDs: from headlamps to the data center

How adaptive beam headlight technology is inspiring next‑generation optical interconnects for AI data centers.

ams OSRAM unveils breakthrough Thin-Film VCSEL for AI Scale-Up, showcasing connectorized multi-core fiber solution and expanding ecosystem network

Forthcoming events

September 28, 2026, Hardware.AD, Munich (Germany), Clemens Hofmann: ”Wide & Slow Interconnects: Photonic Micro Arrays Enabling High-Bandwidth, Low Latency, Optical Data Transmission”