IN Brief:
- CEA-Leti and NcodiN are moving nanolaser-enabled optical interposers onto a 300 mm silicon photonics process.
- The work targets one of AI hardware’s biggest constraints: bandwidth and energy limits in data movement between advanced chips.
- A successful 300 mm transition would move optical interconnects closer to scalable, wafer-level semiconductor manufacturing.
CEA-Leti and NcodiN have launched a collaboration to industrialise nanolaser-enabled optical interconnect technology on a 300 mm silicon photonics process, a move aimed squarely at one of the semiconductor industry’s most stubborn bottlenecks: getting far more data in and out of advanced processors without allowing energy consumption and packaging complexity to spiral.
NcodiN’s NConnect platform is built around a laser-on-silicon architecture designed for dense optical interconnect integration inside advanced computing packages. The company says its nanolaser approach is 500 times smaller than conventional devices, with the potential to support integration densities above 5,000 nanolasers per square millimetre and energy operation around 0.1 pJ/bit. The significance of the new partnership is that those claims are now being pushed beyond proof-of-concept territory and into a 300 mm, CMOS-compatible manufacturing path.
That transition matters because optical interconnects have long promised to relieve the limits of copper inside high-performance systems, yet packaging, manufacturability, and cost have repeatedly held them back. AI accelerators and bandwidth-heavy compute architectures are changing that equation. As data movement becomes a larger share of total system power and a more visible constraint on scaling, the commercial case for shorter-reach, in-package optical links has become markedly stronger.
CEA-Leti’s role gives the project industrial weight. The institute has spent years building out silicon photonics, heterogeneous integration, and advanced semiconductor process capabilities, and the focus here is not simply on demonstrating another lab-scale photonic component. It is about proving that NcodiN’s interposer architecture can migrate onto a wafer-level flow that is credible for volume production, where yield, repeatability, and integration with existing semiconductor manufacturing matter more than isolated device performance.
Francesco Manegatti, co-founder and CEO of NcodiN, said the collaboration is intended to demonstrate NConnect’s compatibility with 300 mm wafers, which he described as essential for commercial-scale production and cost-effective use in AI-centric processors and high-bandwidth computing systems. Sébastien Dauvé, CEO of CEA-Leti, said moving photonics onto a 300 mm CMOS-compatible process marks a turning point for optical interconnects that need to be produced at the scale, cost, and reliability demanded by the AI market.
The harder part comes next. Optical packaging has generated no shortage of promising demonstrations over the years, but industrial relevance depends on process control, thermal behaviour, integration density, and the ability to fit photonics into supply chains already built around mature semiconductor manufacturing logic. That is precisely why 300 mm matters: it shifts the conversation from elegant device physics to whether optical interconnects can begin to behave like a scalable manufacturing technology.
If the collaboration delivers on that transition, it will not simply add another photonics milestone to the European research ledger. It would strengthen the argument that next-generation AI hardware will need optics not at the rack edge alone, but much closer to the silicon itself.
