Lightium: Shaping the Future of Photonic Chips Beyond Silicon

With AI workloads advancing rapidly, it’s no longer compute that holds us back, but the bandwidth required to transfer data at high speed.

Compute performance has improved 1,000 times more than bandwidth over the past two decades. As workloads scale, traditional electrical interconnects can’t keep up. An increasingly adopted architectural approach integrates optical interconnects at or near the chip level, using light instead of electricity. One promising material for this is Thin-Film Lithium Niobate (TFLN), enabling high-speed, low-energy communication in next-gen systems.

Lightium was founded in 2023 by Amir Ghadimi (CEO), Frédéric Loizeau (CRO), and Dirk Englund (advisor and first business angel), who also co-founded Lightmatter. Based in Zurich, the team is developing next-generation photonic chips using TFLN, a material that enables optical data transmission up to four times faster and four times more energy-efficient than conventional electrical links. 

What sets them apart: a capital-efficient “virtual foundry” model that scales industrial-grade photonics without the need for a $200M facility and years of development. In September 2024, Lightium raised $7M in a seed round led by Vsquared Ventures and Lakestar.

Learn more about the future of photonic chips beyond silicon from our interview with Lightium co-founder and CRO, Frédéric Loizeau:

What Inspired You to Start Lightium?

I’m an engineer by training and have spent over 15 years in the semiconductor industry. Early in my career, I devoted a significant amount of time to research. I completed a PhD, followed by a postdoc in the U.S., and spent nearly seven years working in labs, often alone in basement environments without windows. I enjoyed the science and hands-on experimentation, but over time, I realized I wanted more connection with people and the broader industry.

About eight years ago, I moved into more business-focused roles, concentrating on customer acquisition, relationship building, and understanding market needs. That shift became essential when we later founded Lightium.

Unlike many founders, I never dreamed of becoming an entrepreneur. The idea for Lightium emerged from deep frustration with being unable to serve large customers and meet their high-volume demands.

Around 2018 or 2019, my co-founder Amir and I were working at a Swiss research institute, similar to the Fraunhofer Institutes in Germany. We had developed promising photonic technology and were exploring it only from a scientific rather than a commercial perspective. Specifically, we focused on TFLN technology in the context of quantum computing, which was still a relatively niche field at the time (and in terms of volume, largely remains so).

But things changed. In 2022, following the release of ChatGPT, we began receiving inbound interest from major companies, including tier-1 transceiver manufacturers, telecom providers, and hyperscalers. Companies began testing our chips, validated the performance, and requested larger volumes. As a research lab, we lacked the capacity to support production at scale. That gap between demand and our ability to supply it was the real catalyst for founding Lightium. That shift from curiosity to necessity confirmed that we were working on something critical to the future of computing.

We founded Lightium to close that gap and bring this technology out of the lab and into real-world deployment.

How Is Lightium Solving the Bandwidth Bottleneck?

Over the past two decades, processing power has advanced roughly 1,000 times faster than bandwidth. While CPUs and GPUs have made enormous strides, the ability to transfer data between them has lagged behind. The bottleneck has shifted from computation to communication bandwidth.

Editor’s note: Processing power (also called computing power) refers to a system’s ability to perform calculations and execute tasks. It determines how fast a computer can process data.

Editor’s note: Bandwidth, in contrast, is the amount of data that can be transferred between systems per second. It affects how efficiently data moves across components—a key factor in data-intensive applications like AI.

Photons are the fastest known medium for communication, which is why today’s internet infrastructure relies on optical fibers, rather than copper cables. The same principle applies in modern data centers. When massive volumes of data need to move quickly and with minimal energy loss, optical interconnects are the most effective approach.

Editor’s note: Interconnects are the links that move data between chips or systems—either electrically (using wires) or optically (using light). They include components such as wires, optical fibers, and waveguides that physically carry signals within and between computing devices

Photonic integrated chips (PICs) are used to convert data from the electronic domain, where CPUs and GPUs operate using electrons, to the optical domain, where data is transmitted using photons over optical fibers. This conversion needs to happen as quickly and efficiently as possible. Today’s PICs are often made out of silicon, the workhorse platform inherited from the electronics industry. But silicon photonics is approaching its performance limits. It can no longer perform this conversion fast enough or with sufficient energy efficiency. 

That’s where Lightium comes in. Our photonic chips form the core optical engine inside ultra-high-speed transceivers. We design and manufacture them at scale, with TFLN as the enabling material.

Editor’s note: Transceivers are the components that connect computing systems to interconnects. They sit at both ends of the link. In optical systems, transceivers convert electrical signals into optical signals and vice versa. In electrical systems, they facilitate the transmission and reception of electrical signals.

TFLN is a glass-like material with exceptional optical properties. It enables data transfer up to four times faster than silicon-based optical links, while consuming about four times less energy. 

Editor’s note: Thin-film Lithium Niobate (TFLN) is a transparent, crystalline material with excellent electro-optic properties. It’s increasingly used in photonic chips for modulating and routing light at high speeds, and is more efficient than traditional silicon in many optical applications.

What sets Lightium apart is our ability to scale this technology. We’ve solved one of the key barriers to commercializing TFLN technology: manufacturing high-performance chips at volume with the reliability and consistency needed for industrial deployment.

And it seems we reached product–market fit unusually early. We left our research institute in August 2023, incorporated Lightium in September, and by October, we were already receiving our first customer orders. From the very beginning, there was no question about demand. It was clear the market was ready.

What Is the Key Innovation Behind Your TFLN Manufacturing Process, and How Does It Set You Apart From Other Players in the Photonics Ecosystem?

The key differentiator isn’t the material itself, but the business model. Lightium operates as a foundry, more precisely, a virtual foundry. 

In the semiconductor world, companies like TSMC or GlobalFoundries are well-known examples of traditional silicon foundries with CMOS fabs. But building fabs from scratch is extremely capital-intensive. The equipment is costly, and getting everything up and running can take years. When we ran the numbers, we estimated it would take at least $200 million and three years just to get started. That simply wasn’t realistic for us at this stage.

Editor’s note: CMOS stands for Complementary Metal-Oxide-Semiconductor, a standard technology used to build most modern microchips, especially in logic and memory devices.

So we looked for another path. What we found is that across Europe and the U.S., there are legacy fabs that used to manufacture CMOS chips 20 years ago. They may not have cutting-edge resolution, but for our purposes in photonics, their capabilities are more than sufficient.

We decided to partner with a few of these fabs and use them as subcontracted manufacturing partners. They run our process using their existing infrastructure. This allowed us to bypass the massive CapEx and long lead times typically associated with building a fab from the ground up. We didn’t need to buy and install everything ourselves, but we can leverage what was already there.

That’s the real innovation. It’s not in the material, but in how we’ve structured the business to be capital-efficient, fast to scale, and ready for industrial deployment. It’s what allowed us to move from lab-scale prototypes to commercial production, without waiting years or raising hundreds of millions.

Can You Explain the Difference Between a Fab and a Foundry?

A fab is the physical facility where chips are manufactured, including the cleanroom, machines, technicians, and materials. It’s the factory floor. But just owning a fab doesn’t mean you can make useful chips. You also need design tools, building blocks, and validated processes. That’s where the foundry model comes in. 

Traditionally, a foundry both owns the fab and develops standardized process design kits (PDKs), libraries, and chip components. These building blocks are like LEGO pieces: standardized, modular, and designed to be easily combined into more complex systems. Chip designers can then use them to assemble a wide range of systems—from high-speed optical transceivers to switching fabrics, signal processors, or AI accelerators. This standardization is what makes large-scale chip design and manufacturing efficient and repeatable.

Lightium takes this one step further: We’re a foundry without a fab. We develop and qualify photonic building blocks (modular components, like LEGO pieces) and partner with existing fabs to manufacture them. This capital-efficient model enables us to focus on what we do best: facilitating scalable, high-performance photonic integration.

How Do Your Chips Outperform Established Alternatives in Data Centers? Is It “Just” Bandwidth and Power Consumption?

Yes, that’s really the essence of it. Apart from being faster and more efficient, everything else is the same.

That’s actually a big advantage. It keeps the barrier to adoption low. You often hear about emerging materials like graphene, but those are still unproven in real-world data center environments. They might work in the lab, but the risk of integrating them into production systems is high.

Editor’s note: Graphene is a single layer of carbon atoms arranged in a hexagonal lattice. It has exceptional electrical, thermal, and mechanical properties. However, its large-scale manufacturing and integration into existing systems remain major challenges.

Our material behaves like glass. It’s stable, well understood, and doesn’t introduce compatibility issues. Everything else (packaging, connectivity, integration) is the same. That makes it much easier for customers to adopt our technology without having to overhaul their existing infrastructure.

What Are Your Biggest Challenges in Terms of Production Capacity and Supply Chain Dependencies?

The most immediate risk is the lead time. Unlike fields such as quantum computing, which often have development timelines that extend over a decade, the industries we operate in run on much shorter horizons.

When speaking to hyperscalers, their expectations are clear: they need 10 million chips by 2029. This volume is significant, and the timeline is fixed. If we can meet it, that is great. If not, they will turn to other suppliers, even if those are less ideal, as they cannot afford delays.

For Lightium, this means we need to scale rapidly, operate efficiently, and remain focused. Internally, the team is fully aligned on that objective.

The second major challenge is the supply chain structure, which has two dimensions.

First, a challenge common to many deep tech startups: if only one supplier exists for a highly specialized component, large enterprises are unlikely to adopt the technology. Companies like Google or Amazon will not rely on a single-source supplier, especially a startup, for a mission-critical component.

That’s why, in our view, it’s positive that a few other companies are working on similar technology. You don’t want 30 players in the field, but having three or four creates the kind of redundancy and flexibility required for enterprise adoption.

The second dimension is upstream risk. However, we have built redundancy into our supply chain, especially in light of the current geopolitical tensions between the US and China.

Which Application Areas Do You See as the Highest Priority for Lightium’s Technology and Why?

Currently, the primary use case is telecom and datacom. These industries need higher bandwidth and lower power consumption, and that demand is only increasing. So for us, that’s the most immediate and commercially significant application.

That said, the material we’re working with (TFLN) is also interesting for a wide range of other fields, especially those related to optical communication.

In quantum computing, for example, some architectures rely heavily on photonics. Some companies are building quantum systems with photonic chips at their core. TFLN is particularly well-suited for those systems.

We’re also seeing growing interest from the aerospace sector, particularly around satellite communication. If you look at the design of modern telecom satellites, they’re massive, sometimes the size of a bus, with the antenna on one end and the processing unit on the other side. This illustrates just how large these satellites are. 

Right now, these systems are still connected using copper cables, with hundreds of them running through the satellite. That adds weight, creates signal loss, and limits performance. These companies want to replace all of that with optical fiber, and that’s where our chips come in. They’re compact, lightweight, and built for exactly this kind of high-performance optical link.

To summarize, while telecom and datacom are the immediate priorities, we see strong potential in quantum, aerospace, and any domain where fast and efficient optical communication is crucial.

What Advice Would You Give to Fellow Deep Tech Founders?

I’m not sure I’m in a position to give advice, but I can share something that’s been important for us.

In deep tech, hiring is one of the biggest challenges. You need people with specialized skills who also fit well with your company culture, and those profiles are hard to find. We realized early on that building the right team from the start is critical. It’s not just about finding technically strong candidates, but you also need to find people who fit the culture you’re trying to create.

On the technical side, we can usually tell within 10 minutes whether someone knows their stuff. However, determining whether they’re the right personal fit for the company requires significantly more time and effort.

We’ve spent a lot of time on hiring, and while it sometimes felt slow, it wasn’t wasted. In fact, I’d say it was time well invested. 

Because in a deep tech company, your first 20 or so people will define your culture. Finding those first 20 is much harder than in a pure software or sales-driven startup. But once they’re in place, they become the foundation for everything that follows.

So I strongly recommend investing real time in assessing cultural fit. It’s one of the most important things you can do.