ams OSRAM Opens a New Chapter for microLEDs

At SpectroNet, we are always excited when cluster members take proven photonics expertise into entirely new application fields. A strong current example comes from ams OSRAM, which announced on 18 March 2026 that it is advancing its established EVIYOS™ microLED technology toward slow-and-wide AI optical interconnects for data links within and between AI server racks.

What makes this especially interesting for our network is the direction of travel: a technology platform that was industrialized for high-resolution automotive forward lighting is now being adapted for one of the most pressing infrastructure challenges in AI computing. As AI systems scale, the bottleneck is no longer only compute performance itself, but increasingly the power, heat, and complexity associated with moving data. That is exactly where ams OSRAM sees an opportunity for Digital Photonics to deliver real impact.

The new prototype shows why this development is attracting attention. According to ams OSRAM, the microLED array reaches a cut-off frequency above 1 GHz, targets energy consumption below 2 pJ/bit, and is designed for massively parallel short-reach optical I/O paths. The company further reports that individual emitters can support at least 3.0 Gbit/s per lane across a 10-meter link while meeting a bit-error-rate target below 10⁻¹⁵. These are highly relevant figures in AI infrastructure, where interconnect efficiency is becoming a decisive system-level factor.

For many SpectroNet partners, the most compelling aspect may be the transfer of manufacturing know-how. ams OSRAM’s microLED leadership is rooted in EVIYOS™, described by the company as the first high-volume automotive micro-emitter array. In that platform, 25,600 individually addressable pixels are integrated with a CMOS driver chip in a compact package, and the underlying microLED-CMOS integration concept was recognized with the German Future Prize under the theme of “Digital Light.” This background matters because it means the move into datacom is not based on a lab-only concept, but on technology that has already demonstrated industrial scalability and robustness in series vehicle production.

The adaptation to AI interconnects is technically elegant. While EVIYOS™ in automotive applications relies on dense monolithic arrays, the data-center approach uses singulated microLEDs that are separated from the wafer and mounted so that each emitter can couple into its own optical channel. This supports a slow-and-wide architecture: instead of pushing one ultra-fast optical lane to the limit, the system distributes data across many parallel channels. The result is a concept that promises higher bandwidth density, lower switching frequency per emitter, and a fundamentally more parallel data path.

That architectural shift comes with three advantages that should resonate well beyond hyperscale computing. First, reliability: parallel systems can include redundant channels and degrade gracefully if one emitter fails. Second, efficiency: lower-frequency emitters consume less power than ultra-high-frequency laser-based solutions and also reduce waste heat. Third, simplicity: a slow-and-wide approach can reduce the serialization and deserialization overhead that is typical of fast-and-narrow interconnect architectures. In other words, this is not only about raw speed, but about building smarter optical systems for real operating environments.

From a SpectroNet perspective, this news is a remarkable example of cross-market innovation in photonics. It shows how competencies built in one demanding sector, in this case automotive lighting, can become relevant in another fast-growing field, namely AI infrastructure. It also underlines a broader message for our network: the future of photonics will be shaped not just by better components, but by the ability to translate proven platform technologies into new, system-critical applications.

We will be following ams OSRAM’s progress with great interest. For other SpectroNet partners, this development is a strong reminder that innovation often happens where established manufacturing excellence meets emerging market needs. In ams OSRAM’s case, the path from adaptive headlamps to AI data centers may look surprising at first glance, but it is a powerful demonstration of what photonics can achieve when industrial maturity and technological vision come together.