By Scott Rauscher, Chief Innovation Officer, Rogue Valley Microdevices
Returning from the SPIE Defense + Security 2026 conference, I observed a few themes across the panels and the audience questions relevant to photonics applications in the defense industry and the semiconductor and MEMS manufacturing support that will drive this forward. The discussions uncovered a more nuanced picture than the usual “photonics is finally happening” headline. Here are my observations of where the ecosystem is headed:
Photonics will Stay Multi-Platform
Decades ago, digital electronics consolidated on silicon Complementary Metal-Oxide-Semiconductor (CMOS), and once that happened, the entire ecosystem (design tools, foundry process, packaging, test) standardized around it. Photonics looks unlikely to head toward an equivalent moment. Silicon photonics holds the largest share of short-reach hyperscale transceivers and continues to expand into co-packaged optics roles. Indium phosphide (InP) remains the dominant platform for long-haul coherent telecom, where native laser integration is a real advantage. Thin-film lithium niobate (TFLN) is gaining traction for high-bandwidth electro-optic modulation, though wafer supply, packaging, and yield remain early-volume considerations. Gallium arsenide (GaAs), indium gallium arsenide (InGaAs), polymer, and chalcogenide platforms each occupy specific application areas tied to their bandgaps, electro-optic coefficients, and spectrum windows. Radio Frequency (RF) electronics never converged on a single material either: silicon, GaAs, gallium nitride (GaN), and InP coexist because the physics drives different choices, and photonics is shaping up similarly.
That has consequences for the foundry ecosystem. The path forward calls for diverse foundry capability across multiple substrates and process flows. Interoperability and design portability across platforms matter more than chasing scale on any single material.
The Lab-to-Fab Gap is Mostly an Infrastructure Problem
Across the panels, one question kept coming up: Are we limited more by technology readiness or by ecosystem alignment?
The honest answer from the foundry side is mostly ecosystem alignment. The technology to ship a working photonic device has existed for years. What’s still missing is the supply-chain coordination (design help, packaging readiness, qualification pathways) that lets a designed device actually become a deployed product. These pieces are fragmented enough that startups often end up rebuilding them themselves. The bottleneck is coordination, with no single party owning the gaps that span the value chain.
Customer demand is no longer the binding constraint. This has shifted in the last 12 to 18 months. Customer pull is now leading ecosystem development, and design standardization (published design rules, characterized process tolerances, Process Design Kit (PDK) maturity) is where the foundry community can move fastest.
Photonics Is Already Deployed at Hyperscale, and MEMS is Part of That
Google has been running Microelectromechanical Systems (MEMS)-based optical circuit switches in their Jupiter network at massive scale for years, replacing traditional spine packet switching at the data center fabric level. It’s one of the largest MEMS deployments anywhere. Artificial Intelligence (AI) training workloads, with their predictable communication patterns, are an unusually good fit for circuit-switched optical fabrics, and as AI cluster sizes scale beyond a single rack, the question of whether NVIDIA-class architectures eventually follow that model is genuinely open.
Chip-scale photonics is already deployed at scale across multiple parts of the network. That includes wavelength-selective switching in long-haul transport, optical circuit switches in hyperscale data center fabrics, and Vertical-Cavity Surface-Emitting Laser (VCSEL) based optical interconnects across the data center floor. Co-Packaged Optics (CPO) sits at an earlier stage. Intel, IBM, TSMC, Marvell, Broadcom, and ASE are all actively working to bring CPO modules toward scale. Production volumes remain limited, and the timing of broad commercial availability is one of the open questions.
The U.S. has Real Capability Gaps in Specialty Photonics
For the defense audience at SPIE, this was the urgent topic. There are real gaps in U.S. sovereign capability for specialty photonic platforms today.
There isn’t a credible U.S. production foundry for TFLN at scale today. Trusted Foundry pathways for emerging photonic platforms are still being worked out, and several specialty substrates have thin domestic supply. The result is a real gap between defense application requirements and what U.S.-based fabs can produce inside International Traffic in Arms Regulations (ITAR) compatible flows.
Closing those gaps will be a long, capability-by-capability buildout, with National Institute of Standards and Technology (NIST) programs, CHIPS and Science Act-aligned funding, and individual foundry investments all playing impactful roles.
VLPI: Federal Origin and Community Momentum
Across both panels where I was a speaker, Very Large-scale Photonic Integration (VLPI) was the overarching theme. The term has two related but distinct meanings worth keeping straight.
The federal origin sits at the Defense Advanced Research Projects Agency (DARPA). DARPA’s Microsystems Technology Office (MTO) opened up VLPI as a technology category with its June 2025 Request for Information (DARPA-SN-25-88), and the follow-on program, Photonic Integrated Circuit Architectures for Scalable System Objectives (PICASSO), was announced in early 2026 with roughly $35M in anticipated funding and a July 2026 start.
In parallel, a community-driven VLPI initiative had its public launch at this year’s SPIE, with a Global VLPI Design Competition planned to begin at SPIE Defense + Commercial Sensing 2027. The community framing is consistent with the federal program. As AI systems scale, the binding engineering constraints shift toward data movement, with bandwidth, latency, and energy per bit emerging as the primary limits. VLPI groups co-packaged optics, photonic acceleration, reconfigurable architectures, and scalable system-level integration into a single architectural program targeting those constraints. VLPI is a reference to DARPA’s Very Large-Scale Integration (VLSI) program from the late 1970s, which helped catalyze the modern semiconductor industry.
DARPA was present at the panel discussions, which signals that VLPI and PICASSO are being actively connected to defense application requirements where bandwidth and power already drive architecture choices. The conversations at SPIE focused on translating into manufacturing pathways.
Where Rogue Valley Microdevices Fits
For us at Rogue Valley Microdevices, the SPIE conversations reinforced priorities we are already working on, including:
· Formalizing design rules across process modules towards PDK capability so customers can design with documented, measured tolerances in hand from day one · Offering foundry-flexible design services through RVM Design Services so startups and primes can get design help while keeping their IP and masks
· Continuing to develop our silicon nitride photonics platform, with pristine stoichiometric film deposition and etched waveguides backed by published optical loss data
· Bringing 300mm process control to MEMS and photonics-adjacent applications, where much of the industry still runs on 200mm tools
A real transition-and-deployment ecosystem requires the foundry layer to be transparent, accessible, and designed around the customer’s ability to keep their IP. That is the model we are building toward, and we left SPIE with even greater conviction in this model.
Finally, I want to thank Rouzbeh Borhani and the Plug and Play semiconductor team for organizing such a substantive set of panels, and all of my co-panelists across the VLPI: Engineering the Next Era of Photonic Integration and Plug and Play on Transition and Deployment session. Thanks to Hamed Dalir (University of Florida), Jorn Smeets (PhotonDelta), Shahab Ardalan (AMD), Antonio Teixeira (PICadvanced), Rayhane Ghane (ficonTEC), and Amirhossein Ghods (Mesa Quantum), for such collaborative conversations. The most useful panel discussions are the ones where everyone brings a different piece of the puzzle and engages with each other’s perspectives, and both of these panels achieved this goal.
The Real Challenge Ahead
The next phase of photonics growth will depend less on proving the technology and more on building the manufacturing ecosystems capable of deploying it at scale. The companies and foundries that can bridge design, process integration, packaging, and production readiness will help define how quickly the industry moves from promising demonstrations to widely deployed systems.
A version of this article was originally published in Semiconductor Digest’s June 2026 issue.
Rogue Valley Microdevices is a U.S.-based MEMS and semiconductor foundry offering design services and manufacturing from 50mm to 300mm. Learn more at roguevalleymicrodevices.com.