Matt Crowley is Chief Executive Officer of Scintil Photonics. A physicist by training, Matt built his career transitioning advanced semiconductor technologies from development to volume manufacturing. Before Scintil, he led MEMS technology company Vesper Technologies to high-volume production and through its acquisition by Qualcomm. He holds a degree from Princeton University.
Tell us about your company?
Scintil is a photonic integration company. We build dense wavelength division multiplexing (DWDM) light engines using a heterogeneous integration process we call SHIP™, for Scintil Heterogeneous Integrated Photonics. SHIP runs in production on Tower Semiconductor’s silicon photonics line. Sylvie Menezo, our founder and CTO, originated the technology out of CEA-Leti in Grenoble, where we are headquartered, with operations now expanding into the US.
The shorter version: photonics is going through the transition that semiconductors made 40 years ago, from discrete components on boards to integrated circuits on wafers. SHIP is how we participate in that transition. By bonding III-V gain material directly onto a silicon photonic wafer in a foundry process, we put lasers, modulators, detectors, and passive photonics on the same die, in the same flow, on the same lines that already produce tens of millions of optical transceivers a year.
LEAF Light™ is our first commercial product on SHIP, a single-chip DWDM laser source for AI scale-up networks. Our $58M Series B last year included NVIDIA as a participant.
What problems are you solving?
The hardest problem in AI infrastructure right now is not compute. It is the network between compute elements. As accelerator clusters scale into thousands of processors, the network determines how much of the compute is actually usable, and the metrics that decide it are energy per bit, latency, and bandwidth at the edge of the package.
Copper is at the wall on all three. Single-wavelength co-packaged optics (CPO) is in production now and works well for scale-out. The next step, where the system gains compound, is DWDM CPO for scale-up. That step needs a multi-wavelength light source that can be manufactured on existing silicon photonics flows, at hyperscale volumes, with the wavelength precision and reliability the architecture demands. That has been the missing piece. Building it is what we do.
What application areas are your strongest?
AI scale-up networks. The economics and the technical requirements both point to DWDM CPO as the destination architecture, and that is where our platform fits most directly. LEAF Light targets the DWDM laser source, which is the highest-value, hardest-to-manufacture element of that architecture.
The same process supports a broader set of integrated photonic devices, including transceivers with integrated lasers, optical circuit switches with semiconductor optical amplifiers, and high-speed modulator arrays. These are not separate process developments. They are products that share a single foundry-resident process flow. We are starting with the application that has the most immediate market pull.
What keeps your customers up at night?
Two things, in this order. First, manufacturing capacity at the volumes hyperscale build-outs already require. AI infrastructure committed capex at a scale that assumes the optical layer will be there, in the bandwidth, density, power, and reliability budgets the architecture needs, when the buildings switch on. The architectural debate is largely settled. The execution debate is wide open, and it lives at the foundry.
Second, headroom. A first-generation specification is a starting line, not a destination. Customers planning multi-rack scale-up clusters need to see a path where bandwidth per fiber scales without re-doing the fiber plant, the package, or the per-channel electronics every two years. They want to know that the supplier they choose for the first step has a process flow that supports the second and third steps on the same lines. That second question is the harder one for most of the industry to answer.
What does the competitive landscape look like and how do you differentiate?
Most of the energy in the sector right now is around three architectural choices for the DWDM light source. Discrete distributed feedback laser arrays externally combined into a fiber. Mode-locked lasers producing a wavelength comb in a single cavity. And heterogeneous integration of III-V gain material onto a silicon photonic wafer.
Each works in the lab. They separate on the manufacturing curve. With discrete arrays or external combiners, every additional wavelength is another assembly step, another alignment, another bill-of-materials line. With heterogeneous integration at the wafer level, the next wavelength is another circuit element. The cost curve follows a semiconductor learning curve, not a linear assembly curve.
SHIP is a backside-on-box flow built on Tower’s PH18M silicon photonics process. We start with standard PH18M, perform a handle exchange to expose the buried oxide as a bonding surface, bond III-V die, and process them into integrated photonic devices in alignment with the silicon waveguides beneath. The flow is foundry-resident, with a process design kit, design rules, and the manufacturability discipline a real product line requires. Our differentiation lives there.
What new features/technology are you working on?
Last month at OFC we launched the LEAF Light Evaluation Kit. It is the first DWDM laser source EVK to move from internal validation into a customer-facing program. Each unit hosts laser optical sub-assemblies in 8-wavelength or 16-wavelength configurations and provides a defined integration path into the ELSFP module form factor the industry is converging on.
A piece of the platform we are pushing hard on is what we call WaveGuard™, on-chip frequency monitoring and trimming that holds DWDM channel spacing within tight tolerances across temperature, ageing, and package stress. Wavelength precision is one of the things that has historically held DWDM back in production, and intelligent on-die control is how we solve it.
Beyond LEAF Light, the same SHIP flow supports integrated transceivers, optical circuit switches with semiconductor optical amplifiers, and other building blocks that hyperscalers will want next. We are walking that roadmap in step with customer evaluations rather than ahead of them.
How do customers normally engage with your company?
Through our website at www.scintil-photonics.com, the contact form, or directly at contact@scintil-photonics.com. Our LEAF Light EVK is available now to qualified customers through an early access program. Teams that want to bring real constraints to a technical discussion are exactly the conversations we are looking for.
Also Read:
CEO Interview with Dave Kelf, CEO of Breker Verification Systems
CEO Interview with Geoffrey Rodgers of Chameleon Semiconductor
CEO Interview with Xianxin Guo of Lumai
Share this post via:
Solving the EDA tool fragmentation crisis