LPKF‑Intel‑Ayar push chiplet optics
- Ayar Labs’ detachable-connector push has turned into a real ecosystem story, tying Intel Foundry’s glass photonic interconnects to deployable optical I/O for AI systems. - The concrete markers are 8 Tbps from Ayar’s UCIe TeraPHY chiplet, Intel’s 4 Tbps OCI demo, and Corning glass-waveguide links in FPGA proofs. - The bigger shift is packaging — optics are moving from lab demos toward serviceable, manufacturable chiplet interconnect stacks.
Optical chiplet I/O is the part of the AI hardware stack where a lot of wishful thinking used to live. Everyone agreed copper links were becoming the bottleneck. But getting light on and off a package at high density — without turning assembly into a science project — was the hard part. What changed over the last two years is that the pieces stopped showing up as isolated demos. Intel, Ayar Labs, Corning, Teramount, and glass-processing players like LPKF now look less like separate bets and more like an emerging packaging stack. (newsroom.intel.com) ### What is the actual object here? This is about chiplet-era optical I/O — moving data between processors, accelerators, switches, or memory using light instead of short-reach electrical SerDes over copper. Ayar’s pitch is an in-package optical chiplet, TeraPHY, paired with an external light source. Intel’s pitch is similar in destinati(newsroom.intel.com)more bandwidth, longer reach, and lower power than electrical I/O can comfortably deliver. (ayarlabs.com) ### Why does glass keep showing up? Glass is becoming the favored packaging material because it stays flat, routes dense signals cleanly, and plays well with panel-scale manufacturing. LPKF’s whole role is enabling that substrate transition with laser-based glass processing — blind vias, thin-glass work, and structures that hybrid electro-optical packages need. In plain English, glass is not the produ(ayarlabs.com)kes dense electrical-plus-optical packaging less painful. (lide.lpkf.com) ### What did Intel actually prove? Intel’s June 26, 2024 OCI demo mattered because it was not just a photonics die on a slide. Intel showed a fully integrated bidirectional optical I/O chiplet co-packaged with a prototype CPU and running live traffic. The published specs were 64 channels at 32 Gbps in each direction over up to 100 meters of fiber. That is basically a proof that co-packa(lide.lpkf.com)erconnect, not a lab curiosity. (newsroom.intel.com) ### Where does Ayar fit? Ayar is the startup pushing the commercial chiplet story hardest. Its TeraPHY optical I/O chiplet is positioned as a UCIe optical interconnect chiplet with 8 Tbps of bandwidth, and the company keeps showing it in increasingly system-like contexts — with Intel and Altera FPGAs, with Corning glass waveguides, and i(newsroom.intel.com)ard manufacturable systems. That is marketing, sure — but the ecosystem demos back up the direction of travel. (ayarlabs.com) ### Why are detachable connectors such a big deal? Because permanent fiber attach is miserable for manufacturing and service. If fibers are epoxied in place at the package edge, yield, rework, and field replacement all get ugly fast. Ayar has been highlighting a detachable optical connector ecosystem built around Intel Foundry’s detachable glass photonic interconnect appro(ayarlabs.com) — detachable, serviceable, passive-alignment fiber-to-chip connectivity for high-volume co-packaged optics. That sounds like plumbing, but it is the difference between a clever demo and something an OSAT can actually build and repair. (ayarlabs.com) ### So where does LPKF fit in this map? LPKF is not the optical-engine vendor. It is a picks-and-shovels supplier for the glass substrate layer underneath this whole trend. Its relevance is that advanced packaging is shifting toward glass interposers and panel processing, and LPKF has spent the last few years arguing that its LIDE process is ready for high-volume manufactu(ayarlabs.com)of chiplet packages, companies like LPKF become more strategically important because they help make the substrate manufacturable at scale. That last step is an inference — but it follows directly from where the ecosystem is heading. (lide.lpkf.com) ### What changed most recently? The newest signal is not a single breakthrough paper. It is consolidation and ecosystem thickening. Ayar spent 2024 and 2025 showcasing Intel Foundry, Corning, and Altera-linked optical I/O demos. Then in April 2026, Molex agreed to acquire Teramount, which suggests detachable fiber-to-chip connectivity is graduating from niche component tech into mainstr(lide.lpkf.com) means the packaging problem has become commercially real. (ayarlabs.com) ### Bottom line? The real story is not “optics are coming.” That part was obvious. The story is that chiplet optics now has a plausible manufacturing stack — optical engines, glass waveguides, detachable connectors, and glass-substrate processing — with Intel and Ayar proving the compute side and companies like LPKF h(ayarlabs.com)start treating it as a sourcing decision. (newsroom.intel.com)
