Lightwave Logic inks foundry PDK partnerships to integrate EO polymers into silicon photonics

Silicon photonics is the chip industry’s way of moving data with light instead of just pushing more electrons through hotter copper links. The bottleneck is no longer only compute — it’s getting data in and out of AI systems without blowing up power budgets. That is where Lightwave Logic has been trying to fit in with electro-optic polymers, which are materials meant to make optical modulators faster and more efficient. The news is that, in March 2026, those materials started showing up inside actual foundry design flows rather than staying at the demo stage. ### What actually got announced? Three separate March announcements matter here. On March 3, SilTerra said Lightwave Logic’s polymer modulator platform was available inside a Luceda Photonics PDK for SilTerra’s silicon photonics process. On March 11, Tower and Lightwave Logic signed a development agreement to integrate Lightwave’s modulator PDK for designs targeting GlobalFoundries’ silicon photonics platform. ### So was Marvell part of this? No — not in the announced partnerships. Marvell shows up in the broader market because it is a major optical-connectivity player for AI infrastructure, and it has been pushing 1.6T optics and silicon photonics products. But the specific Lightwave Logic foundry and PDK integrations disclosed in March were with Tower, GlobalFoundries’ design ecosystem through GDSFactory, and SilTerra through Luceda. ### Why does a PDK matter so much? A PDK is basically the kit designers use to build something a foundry can really manufacture. If a new device is not in the kit, it is still a science project for most customers. Once the modulator is packaged as parameterized cells, compact models, and design rules, engineers can simulate it, place it in a photonic circuit, and send it to tapeout without hand-building the whole flow. That is the real step change here. ### What are they targeting? The numbers are pretty specific. Tower’s deal targets bandwidths of 110GHz and beyond, aimed at 400G-per-lane applications. The Tower program also calls for multiple engineering tapeouts during 2026 to validate low-power 200G and 400G architectures. SilTerra’s flow targets 200 Gbit/s and 400 Gbit/s per lane and says an initial tapeout was already completed in early 2026, with characterization expected in mid-2026. ### What about GlobalFoundries? The GlobalFoundries path is a little different. Lightwave’s modulator blocks were integrated into GDSFactory’s PDK so customers can design for GF’s platform, and a new tapeout had already launched by March 16. More tapeouts were tentatively scheduled for later in 2026. GF was also described as optimizing its silicon photonics process for advanced slot waveguide manufacturing tied to Lightwave’s next-generation device architecture. ### Why is AI the hook here? Because AI clusters need absurd amounts of bandwidth, and every extra watt spent moving bits is a problem. The pitch for polymer modulators is that they can help deliver 200G and 400G per lane with lower power in a compact footprint. That is why every one of these announcements ties the technology to AI scale-up, scale-out, data centers, and high-performance computing. ### What is still missing? Commercial volume. PDK inclusion is not the same thing as customer production revenue. The work now is validation, tapeouts, characterization, packaging, and system integration — the messy middle between a promising device and a shipping optical engine. Basically, the story moved from “can this exist?” to “can customers design it into something manufacturable?” That is progress, but not the finish line. ### Bottom line The clean read is this: Lightwave Logic did not announce a Marvell partnership. It announced that its EO polymer modulators are entering real silicon photonics PDKs at Tower, GF’s ecosystem, and SilTerra. That makes the technology more usable, more testable, and a lot more relevant to the AI interconnect buildout — but 2026 still looks like a tapeout-and-validation year, not a full-scale deployment year.