With generative AI booming, demand for ever-faster and more efficient hardware is exploding. But traditional chip designs-built around copper interconnects-are hitting hard limits. That’s where NcodiN comes in: this deep-tech startup has just raised €16 million to advance a breakthrough-the world’s smallest silicon-based laser, aimed at transforming how data moves inside AI chips.
What is NcodiN doing?
NcodiN is developing a photonic interposer platform called NConnect. Instead of moving data electrically-through copper wires-NConnect uses light-tiny laser signals and optical links-to carry data between chip components. The real star: a nanolaser embedded on silicon, small enough and efficient enough to fit directly on a chip.
Their proof-of-concept lasers use less than 0.1 picojoule per bit (<0.1 pJ/bit), and support extremely dense integration, far beyond what copper wires can handle. This could unlock dramatically higher memory bandwidth and much lower energy use.
With the new funding, NcodiN plans to move from lab prototypes to industrial-scale production, including work in a CMOS pilot line-on 300mm wafers-scaling the supply chain, hiring engineers, and preparing for collaborations with leading chip manufacturers.
Why It Matters for AI Hardware
Interconnects, in fact, are among the biggest bottlenecks for AI chips today, be they GPUs, accelerators, or neural-network processors. As chips grow powerful and utilize many “chiplets” working together, the moving of huge amounts of data between chiplets becomes slow, hot, and energy-hungry. In this regard, copper wiring struggles.
NcodiN’s optical technique offers key advantages:
- Massive data bandwidth: light can carry far more information compared to electrical wires over the same distance.
- Lower energy consumption: Photonic links waste much less energy, which reduces heat and power draw at scale.
- Better scalability: As AI chips are getting larger and more complex, photonic interconnects scale more gracefully compared to copper-based wiring.
- Compatibility with existing designs: NcodiN says its tech can integrate without dramatic redesigns of current chip architectures.
It could, if successful, push the next generation of “AI factories” into a new era of power and efficiency: data centers, big models, cloud GPUs.
What the Funding Does
The €16 M seed round was led by MIG Capital, joined by Maverick Silicon, PhotonVentures, and Verve Ventures, in addition to previous investors such as Elaia, Earlybird, and OVNI. This round was oversubscribed – a clear sign that investors believe in the roadmap.
With that money, NcodiN will:
- Expand its R&D and clean-room capabilities
- Begin industrial-scale production efforts
- Establish the operation in Silicon Valley in order to create closer relationships with leading chipmakers.
- Start pilot production of its NConnect interposers on 300 mm wafers
- Build supply chains and partnerships required for large-scale manufacturing
Difficulties Ahead — And What’s Being Done
This is cutting-edge work: integrating lasers on silicon is complicated because lasers traditionally rely on materials that don’t mix well with standard CMOS processes. NcodiN claims to have solved it with advanced nanotechnology, but moving from prototype to mass production remains a big challenge.
There’s another hurdle: heat, interference, and precision. Optical signals are finicky; chips have to be carefully designed to prevent crosstalk, signal loss, and thermal problems. NcodiN is building a clean-room and working with industry advisors-including some photonics veterans-to address those challenges.
Finally, chip manufacturers need to adapt, albeit minimally, to accept photonic interposers. That means testing, validation, and a shift in how chips are packaged. The seed funding will help NcodiN push through those early hurdles.
What This Means for You and the Tech Industry
- For AI researchers and data-center operators, this might mean faster, cheaper, and greener computing infrastructure.
- For everyday consumers, this could mean that more efficient AI hardware might, over time, drive more AI-powered services onto mainstream devices without massive energy costs.
- For the technology industry, photonic interconnects could become commonplace in high-end chips, initiating a transition away from copper wiring, used for decades, to light-powered data highways.
NcodiN’s success could shape the next wave of hardware design, especially as AI models keep growing in size and complexity.
What to Watch Next
- That might well be the case if NcodiN’s industrial pilot is successful: successful production on 300 mm wafers could make the company’s lasers part of actual AI-chip products.
- Tie-ups with major chipmakers or cloud providers-such deals, when forged, could accelerate the pace.
- Performance benchmarks and energy-efficiency reports: a comparison between photonic interposers and traditional interconnects at scale.
- Integration into commercial AI systems: once proven, this technology may show up in next-generation GPUs or AI accelerators within a few years.
