LIGHTSPRING raises €3.1 million to tackle photonic chips’ weakest link

Besançon-based deeptech company Lightspring has raised €3.1 million in a seed round led by Atlantic and OVNI Capital. Built on research conducted at the FEMTO-ST laboratory, it is developing a process for fabricating three-dimensional optical interfaces directly on chips. Its objective is to address a major obstacle to the industrialisation of photonic circuits: connecting them to fibres, lasers and other components. That problem is set to become more pressing as AI infrastructure makes greater use of optical interconnects.

Photonic chips can carry and process information using light. First, however, that light has to enter the chip and find its way out again.

This apparently secondary operation is where Lightspring has chosen to work. The manufacture of photonic integrated circuits, or PICs, has become considerably more structured in recent years. Industrial platforms can now produce circuits based on silicon photonics, silicon nitride or indium phosphide. Connecting those circuits to optical fibres, lasers and other chips remains far less standardised.

The components must be positioned with micrometre, sometimes sub-micrometre, precision. Engineers then measure the signal, adjust the position and secure the assembly. Such active alignment is expensive, relatively slow and difficult to scale to volumes comparable with those of the semiconductor industry.

According to Lightspring, packaging can account for around half the cost of a photonic chip, compared with roughly 20% for fabrication at a foundry. Those proportions vary substantially by architecture, but the underlying concern extends across the industry: as PIC manufacturing advances, assembly becomes a major obstacle to mass production. Nanoscribe, for example, also identifies assembly complexity and alignment requirements as central challenges for industrial photonic packaging.

Printing the interface directly onto the chip

Lightspring proposes a different way of addressing that obstacle. It uses two-photon polymerisation, or 2PP, a lithography technique capable of producing highly precise three-dimensional structures only a few micrometres across. Instead of trying to align a fibre and a waveguide perfectly, it prints an optical structure directly onto the chip to fit their respective geometries.

Microlenses, three-dimensional waveguides and beam-shaping elements can then act as an intermediary between the components.

The technology is intended to support connections between a fibre and a chip, a laser and a photonic circuit, or even multiple chiplets. Lightspring is targeting losses below 0.5 dB per interface and aims to remove the need for active alignment in some configurations.

The idea is to manufacture an interface that adapts to the components’ positions rather than force every component into an almost exact meeting point.

“The whole industry is working on making photonic chips faster,” says Grégoire Bonnat, Lightspring’s co-founder and CEO. Yet light still loses energy as it enters or leaves a component, he argues. Lightspring’s answer is to fabricate that optical interface much as the rest of the chip is fabricated, “in one reproducible process step.”

Turning packaging into a programmable process

Lightspring’s main ambition extends beyond 3D printing. The company is developing what it describes as a process design kit, or PDK, for 3D optical coupling. That design layer could ultimately prove more valuable than any individual printed structure.

In the semiconductor industry, a PDK connects chip design with a factory’s physical capabilities. It specifies the available components, their properties, design rules and manufacturing constraints. Designers can use it to create circuits that the factory can then reproduce.

Lightspring wants to apply this logic to optical interfaces. In time, a PIC designer could select a coupling architecture, incorporate it into the chip design and generate both the 3D structures and the parameters required to manufacture them. Packaging would cease to be treated solely as a bespoke operation at the end of the production line and could instead be planned from the design stage.

This may be the company’s most consequential proposition: turning a complex mechanical operation into a programmable manufacturing process.

Lightspring CTO Adrià Grabulosa and scientific director Daniel Brunner, who is also a CNRS research director, have worked for several years on additive three-dimensional photonic integration at FEMTO-ST. Their research has included CMOS-compatible processes and the integration of hybrid photonic platforms.

AI is bringing optics closer to the processor

The packaging problem takes on a different scale as AI infrastructure grows.

In compute clusters, a rising share of energy is spent moving data between processors, memory and networking equipment, alongside the energy used for computation itself. As accelerator counts and data rates increase, electrical interconnects face growing constraints involving power consumption, distance and heat dissipation.

The industry is consequently moving optical interfaces closer to compute. Beyond the optical modules mounted on the front of network switches, some companies are pursuing co-packaged optics: placing photonic components inside the package, or very close to network ASICs and potentially other accelerators. The approach promises better interconnect performance, but it also moves one of the hardest assembly problems into the package itself.

ASICs, photonic components, lasers and fibres must then be connected within extremely tight tolerances, using processes suitable for high-volume manufacturing. As FW.MEDIA noted in its analysis of OLIX, FRACTILE and ETCHED, bringing photonics into AI hardware does not make the integration, alignment and testing problem disappear.

This is the market Lightspring hopes to address. The startup plans co-development programmes with industrial partners focused on co-packaged optics and AI infrastructure.

The technology already has competitors

Lightspring is entering a field in which other companies have made significant progress. Germany has two particularly advanced players. Vanguard Automation, now part of Sweden’s Mycronic group, develops Photonic Wire Bonds: three-dimensional waveguides printed between pre-positioned components. Its offering combines machines, materials and software and, according to the company, spans prototyping and automated production. Its BrightWire 3D software detects the positions of waveguides and fibres before generating the structure that connects them.

Nanoscribe occupies another important position. It sells 3D microfabrication systems capable of printing optical elements directly on fibres, wafers or photonic circuits. Its Quantum X align platform claims alignment-detection accuracy down to 100 nanometres and coupling losses of 1 dB or less in some configurations. Work using its equipment has already demonstrated fibre-to-PIC connections with sub-decibel losses.

The case for being independent of the machine

Lightspring emphasises two characteristics of its proposed approach: it aims to work across different materials and manufacturing equipment.

The first addresses a peculiarity of photonics. Unlike electronics, which is largely built around silicon, photonic applications use a range of materials. Silicon photonics, silicon nitride, indium phosphide and lithium niobate each come with their own benefits and constraints.

A system capable of connecting these platforms without requiring packaging to be redesigned from scratch for each one could occupy a valuable position in the supply chain.

Equipment independence may prove even more significant. Nanoscribe and Vanguard sell ecosystems that combine equipment, processes and software. If Lightspring can make its designs and manufacturing recipes work across several microfabrication platforms, the process itself could become its product, rather than the machine that executes it.

That would bring its business model closer to that of a semiconductor intellectual-property and process supplier.

€3.1 million to move from the laboratory to the production line

Lightspring has raised capital to test that industrial proposition. Its €3.1 million seed round was led by Atlantic and OVNI Capital, with participation from Concept Ventures, Ixcore, Plug and Play Ventures and industry business angels. The company has also received a €350,000 i-Lab award and says it has secured more than €3.5 million in equity and non-dilutive financing combined.

Legally established in Besançon in 2026 by Bonnat, Grabulosa and Brunner, Lightspring builds on several years of earlier scientific work. It now plans to recruit in microfabrication, optical characterisation and photonic circuit design, with the aim of reaching a team of around 12. Its website already presents the technology as a platform for making photonic packaging a programmable manufacturing step.

The funding will also be used to complete the PDK, industrialise the design tools and qualify the process under conditions relevant to industrial customers.

Manufacturing yield will be the real test

A high-performing optical connection in a laboratory is not yet a semiconductor product. Lightspring must demonstrate that the same interface can be manufactured hundreds, then thousands, of times with sufficiently consistent performance. Industrial customers will examine fabrication time, production yield, component-to-component variation and long-term stability as closely as optical loss.

The use of polymer materials will be particularly important to assess. Temperature, humidity, vibration, optical power and thermal cycling will require lengthy qualification before adoption in critical applications. The question is whether a connection made through 3D microfabrication will continue to perform across different environments and years of use, at industrial volumes.

From component to industrial standard

Several business models remain open to Lightspring. It could initially carry out packaging itself for customers, particularly through co-development programmes. That would allow it to retain control of the process and gather the data needed for qualification, while eventually requiring substantial industrial investment.

Another route would be to transfer its processes to foundries, packaging specialists or outsourced semiconductor assembly and test providers, known as OSATs. Lightspring would then supply the PDK, designs, recipes and intellectual property needed for production.

That model would be easier to scale. It would make the startup a technology supplier embedded in chip design and production, rather than a manufacturer of optical components. As FW.MEDIA’s examination of Europe’s semiconductor value chain makes clear, control over a critical layer need not depend on owning the factory.

The question is whether Lightspring can establish its layer early enough for several manufacturing platforms to adopt it as a standard.

The market will not wait. Vanguard is already automating photonic wire bonding, while Nanoscribe is advancing aligned lithography systems for photonic production. Lightspring must move quickly through the qualification stages of an industry in which an innovation becomes strategically important only when it can be reproduced at scale.

The €3.1 million announced today will finance that transition. The next milestone is to show that Lightspring’s interface can become a reproducible, transferable industrial building block designed into the next generation of photonic chips from the outset.