At Holst Centre, the real breakthrough is after the breakthrough

A better electrolyser coating. Photonic chips that work with their electronic counterparts. Laser-assisted chip assembly at extreme precision. Artificial intelligence that not only runs on the edge, but helps design the systems that bring it there.

At first glance, the four pitches from researchers at Holst Centre covered very different worlds. Hydrogen, photonics, semiconductor packaging and edge AI do not naturally belong in one story. Yet during the presentations, a clear common thread emerged: Europe’s technological future will not be decided by a single breakthrough in a lab. It will be decided by whether those breakthroughs can survive the journey into a real, reliable and scalable system.

That journey is where Holst Centre wants to make a difference.

Cassia Santana, Holst Centre, © Bram Saeys

For Cassia Santana, the challenge lies deep inside the hardware of green hydrogen production. Proton exchange membrane water electrolysers are among the key technologies for producing hydrogen from water. But their components have to operate under highly acidic conditions and high electrical potentials. Titanium, a commonly used material in porous transport layers, quickly forms an insulating oxide layer. That is bad news for performance.

The obvious solution is to apply a protective coating, such as platinum. The problem is that porous transport layers are anything but simple surfaces. They are complex, three-dimensional structures with countless tiny spaces that conventional coating methods struggle to reach evenly.

Santana’s team is working on what it calls surface-terminated electroplating. The approach combines a controlled etching step with the formation of a noble-metal seed layer, followed by an electrodeposition process whose growth naturally stops at the required point. That gives researchers far more control over coating thickness, uniformity and conformality - even inside a highly porous structure.

The result, Santana said, is not merely a better coating. It is a more precise way of engineering surfaces in places where conventional methods fail. In tests at 2 volts, the coated porous transport layer delivered a current density 30% higher than a benchmark-coated alternative. The next challenge is no longer laboratory proof. It is building a manufacturing tool that can automate the process and make it ready for market deployment.

Chip assembly

That shift from component innovation to manufacturable systems also defines the work of Riccardo Ollearo. His focus is advanced chip assembly, a field becoming increasingly important as AI drives demand for more compute, higher bandwidth and faster movement of data.

Riccardo Ollearo, Holst Centre, © Bram Saeys

For decades, semiconductor progress depended largely on shrinking transistors. But that route alone is no longer enough. The industry is now moving towards heterogeneous integration: combining specialised chips in a single platform, shortening interconnects and increasingly replacing electrical signals with optical ones where possible.

That sounds efficient in theory. In practice, it creates a painful new bottleneck: assembly.

The more tightly chips and photonic components are integrated, the more precisely they need to be placed. Ollearo is talking about alignment accuracies of around 200 nanometres or less, even though the components themselves may be millimetres in size. At the same time, the process has to work at high volume. Accuracy and throughput are usually uncomfortable partners.

His team’s answer is laser-induced forward transfer, or LIFT. In simple terms, laser light is used to transfer a component from one surface to another without physical contact. The method is designed to avoid damage, place only the desired components and achieve throughput levels that Ollearo said could be at least 100 times higher than existing approaches.

For photonics, cleanliness is just as important as positioning. Light travelling between integrated chips can easily be lost through scattering or poor alignment. Holst Centre’s approach combines sub-micron placement with adhesive-free bonding, keeping the optical interface as clean as possible. It also opens the door to more unusual architectures: stacking components vertically, attaching them to the sidewalls of chips and building three-dimensional systems that use chip real estate previously unavailable for integration.

The technology is still moving towards market application, but Ollearo hinted that discussions are underway with partners in Brabant around a startup, spin-off or joint venture. That is exactly the point where a technical demonstration begins to become an industrial proposition.

Copper interconnects

Jerom Baas picked up the same story from the perspective of photonic chip design. Photonic integrated circuits can place entire optical systems on a chip, enabling applications such as lidar, optical transceivers and quantum sensors. But photonics alone does not make a working product. The system also needs electronic drivers, readouts and digital control.

Jerom Baas, Holst Centre, © Bram Saeys

Traditionally, photonic and electronic chips can be connected through a printed circuit board. In high-speed data applications, however, those copper interconnects increasingly become the bottleneck. That is why the industry is moving towards co-packaged optics, in which electronics and photonics are placed together in a single package. The next step is even more radical: monolithic integration, where both functions are combined on one chip.

The more tightly the technologies are brought together, though, the more complicated the design becomes. Thermal effects, routing constraints, interfaces and layout choices all start influencing one another. There is no universal solution. Each use case requires its own integration strategy.

For Baas, that makes electronic-photonic co-design as much a human challenge as a technical one. Photonic designers and electronic designers have to learn to “speak each other’s language”, he said. Holst Centre’s strength, in his view, is precisely that it brings those disciplines together in one place and turns cross-domain collaboration into a practical design process.

Future of AI

Marzieh Hashemipour-Nazari took the conversation one step further, towards the devices that will eventually use all this computing power. The future of AI, she argued, cannot remain in vast cloud data centres alone. It also needs to move into small, low-power devices at the edge: systems that react instantly, preserve privacy and operate with severe energy constraints.

Marzieh Hashemipour-Nazari, Holst Centre, © Bram Saeys

Smart glasses offer a useful example. Such a device must be context-aware, responsive and always available. But building it is not simply a case of adding an AI accelerator to a chip. It requires a full-system approach, bringing together low power consumption, low latency, connectivity, reliability and verification.

That creates an immense design space. And with growing pressure to shorten time to market, traditional design flows are no longer enough. Hashemipour-Nazari’s argument was that AI should become part of the design process itself. The designer’s role will shift from manually carrying out every step towards guiding, reviewing and accelerating systems that can help explore, implement and optimise designs.

Her warning was equally important: the sector should use AI “smartly, not blindly”. AI may become a design partner, but it cannot replace engineering judgment. The future edge device will therefore be shaped by two forms of intelligence: the AI inside the product, and the human-AI collaboration used to create it.

Awkward, demanding steps

Taken together, the four pitches showed what Holst Centre is really working on. Not isolated inventions, but the awkward, demanding steps that determine whether an invention can become infrastructure, a product or a new industry.

A hydrogen system fails if its coating cannot reach the right surfaces. A photonic chip disappoints if light and electronics cannot work together. Advanced packaging goes nowhere if accuracy cannot be combined with speed. Edge AI remains a promise if the entire system cannot be designed, verified and manufactured in time.

The breakthroughs may begin in the lab. But at Holst Centre, the harder work is making sure they can leave it.