Maxim Karpov: How Enlightra’s Multi-Color Lasers Enable High-Bandwidth Data Communication for AI Clusters
Photonics is a game changer for handling massive amounts of data quickly and efficiently.
One of its most powerful advantages comes from the fact that lasers with different wavelengths don’t interfere with each other. Thus, if you want to parallelize data transmission, you don’t have to scale through physical hardware but can use multiple lasers with different colors, i.e., different wavelengths, in the same optical waveguide.
Previously, building such a ‘multi-color’ laser required an entire optical table full of equipment. Since our last interview in the summer of 2022, Enlightra has made great progress building multi-color lasers on a silicon chip driven by a single continuous-wave laser.
We had the pleasure of speaking again with their co-founder and co-CEO, Maxim Karpov, about what to keep in mind when developing silicon photonics, how they shifted focus from telecommunications to short-range data communication for AI clusters, and how the Intel Ignite startup program accelerated their journey:
How Did You Get From a Prototype to Your First Product?
When we did our first interview in 2022, we had just moved to a new office, our lab was completely empty, we were in the midst of raising a funding round, and we were just about to finish the prototype testing of our first product.
Over the past two years, a lot has happened, including closing the round, building a strong team of more than 20 people, launching sales of our first product—a tabletop comb laser—and developing its highly miniaturized module version for high-volume communication applications. We are the first company to bring this unique comb laser technology, based on integrated photonics, to the market, and we have already built, certified, and launched sales of the first-ever product using this technology.
We have received highly positive feedback from our customers, with some even conducting long-term stability tests demonstrating over 2,000 hours (more than 2.5 months) of continuous operation without failure.
In addition to building our products, we have also developed a significant amount of ancillary technologies and know-how around them. We developed and accumulated a lot of knowledge about photonic chip design and packaging, electronics, and embedded software to control our photonic chips. It was surprising to see just how many new things we learned about seemingly simple integrated photonic circuits as we were working on the stability and reliability of our devices.
Our customers often push us to exceed current technological limits, and our ability to quickly prototype and build upon our early developments gives us a strong competitive advantage. This is especially important now as we develop module-sized comb laser engines, which must maintain their performance in a small form factor while being convenient and reliable for scalable manufacturing.
What Are the Challenges of Developing Silicon Photonics?
The main challenge for a fabless startup like Enlightra is the time it takes to go from an idea to a prototype and then iterate toward an MVP. Currently, the fabrication time for silicon photonics (SiPho) foundries takes several months. When you factor in the simulation and design time, fabrication time, chip testing, packaging, and potential delays, the total time for a single iteration can approach a year. This means the startup has virtually no room for error and is highly dependent on the stability of the fabrication process at the foundry.
Another challenge is that you literally need to develop and test almost all the photonic components you plan to put on your chip. It’s as if you had to design every resistor and capacitor for an electronic PCB by defining their sizes and materials.
Foundries recognized this problem early and are trying to alleviate it by providing some verified design blocks (known as process design kits, or PDKs). However, in practice, the PDKs are quite limited, covering only basic components and often being optimized for specific wavelengths. This need to develop your own basic components adds additional fabrication runs and further extends the prototyping timeline.
A third challenge is the cost of development. SiPho fabrication runs are expensive, and on top of that, you need to invest in software and hardware infrastructure for simulations, design, and chip testing.
What Applications Are You Targeting?
We target the area of optical communications because it’s one of the lowest-hanging fruits for our comb lasers and, at the same time, the biggest market opportunity. In the beginning, we focused a lot on connecting data centers, which involves rather long-distance communications.
However, we observed a tectonic shift in the past two years as the AI boom began. Driven by the training of large language models, the demand for short-range communication in data centers and AI clusters has skyrocketed to ensure fast training across thousands of GPUs. This is where we currently see the greatest opportunity: providing optical infrastructure for AI clusters.
Current GPU architectures are heavily limited by the speed it takes to move data and share it with other GPUs. Most interconnects between chips today are built with copper wires, which are reaching their limits with today’s training and inference workloads. Many companies have realized that, after one or two more hardware generations, we’ll need to switch to optical interconnects, which will require comb laser technologies to overcome the bandwidth bottleneck.
This represents a great opportunity for us, yet it’s not easy to address: those interconnects work with shorter wavelengths than telecommunications technology, which meant we had to redesign our chips geared toward these new requirements. Thanks to our prior developments, we managed to do it in no time and have already demonstrated a module-size comb laser engine that could work for interconnecting GPUs.
Why Did You Go Through Intel Ignite?
I should start with some pre-story. Every other week, we meet with fellow deep tech founders in Lausanne, and during one such gathering, I talked to Remco from Corintis, a chip cooling startup, who was going through the program at the time.
He was super excited about their time in Munich during the program, all the interactions with the Intel Ignite team, and mentors helping them to advance with their product and market understanding. That was the first time I heard about Intel Ignite. As we started to discover more about the program by talking to its alumni, we became sure that it would also be an excellent opportunity for Enlightra, which we shouldn’t miss.
First, because of the access to the Intel ecosystem and getting our ideas challenged by Intel experts. Second, because of the opportunity to meet and build relationships with other deep tech startups that were at a similar stage and shared similar challenges in building complex hardware or software products. Third, because of the deep tech focus of the program, which I want to explain in more detail.
Our first experience with startup accelerators started in 2022 when we had gone through the Y Combinator startup program earlier. This was an amazing and unique experience, which I’d recommend to anyone. Nevertheless, Y Combinator isn’t dedicated to deep and hard tech startups.
This doesn’t matter at a very early stage because startups face similar challenges on their way to building an MVP, attracting first customers, or fundraising, which doesn’t require deep knowledge of the technology and industry.
However, as startups mature and approach the market with the first product, they face very different problems, and a program like Intel Ignite is a great value-add since they understand our technology and market.
Finally, unlike other startup programs, Intel doesn’t take equity but provides a lot of exposure to Intel and other key industry players. Both are extremely important for a growing startup.
How Did Intel Ignite Accelerate Your Journey?
The 12-week Intel Ignite program includes three on-site weeks: one at the beginning, one in the middle, and one at the end, where startups from the current batch come together for workshops, networking, and other activities.
During the remaining weeks, we had weekly calls with mentors and check-ins with the Intel Ignite team, who held us accountable for our progress, tracked our KPIs, and helped us achieve the goals we had set. Since the goals were quite ambitious, the Intel Ignite period was very intense for us, and we worked hard to show progress on all fronts.
Another thing that was of great value to us was the engagement with our Intel mentors and other Intel experts. They were people with diverse backgrounds who had been working in the industry for many years and could ask us questions we hadn’t thought about, be it about manufacturing, reliability, or what makes a great product for our customers.
It also gave us a better understanding of how we should interact with the key market players and what it would take for them to adopt a new technology. Even though we finished the program nearly a year ago, we still maintain close relationships with our mentors.
Our Ignite batch brought together startups at a similar stage. There are a lot of challenges and uncertainty when you’re building a startup, and it’s great to have a community where you can openly discuss and learn how others are dealing with or have just dealt with similar situations. It is something that unites you and builds lasting relationships.
Another great outcome of the program is our joint project with one of our batchmates, Akhetonics.
All of the above would not be possible without the Intel Ignite team and the work they put into it, which no doubt has to be mentioned. They created a fantastic home-like and supportive atmosphere for everyone. We’ve never hesitated even a minute to reach out to them for any advice or support request. Fundraising, interaction with customers, hiring—you name it. We could just reach out and be 100% sure that the team will be very responsive and helpful.
What Was One of Your Key Learnings From Intel Ignite?
I learned how important it is to regularly zoom out from day-to-day execution and look at your startup from a much broader, or even completely different, perspective. It’s easy to get caught up in a blind execution mode—one more customer, one more email, one more meeting.
However, it’s crucial to regularly force yourself into a strategic mindset, allowing you to see what’s ahead and observe what’s happening around you, where the industry is headed, and where new opportunities may arise.