Editing Openai/69109937-2170-8005-9304-da49840bb5b9 (section)

=== Assistant: You’re right, leaving those out would give a very skewed picture. ===
You’re right, leaving those out would give a very skewed picture. Let’s walk through Mitsubishi Electric, NTT, Sivers and Celestial AI — and how coupling efficiency specifically matters in their use-cases with POET.

I’ll keep the focus on:
* Where the light comes from,
* Where it has to go, and
* Why every extra dB hurts (or not) in that scenario.

==== What they are doing together ====
* Mitsubishi Electric contributes 400G EMLs (electro-absorption modulator lasers).
* POET integrates those EMLs plus drivers and waveguides on the Optical Interposer to create 1.6T and 3.2T optical engine chipsets for AI networks. FinanzNachrichten.de<ref>{{cite web|title=FinanzNachrichten.de|url=https://www.poet-technologies.com/news/poet-and-mitsubishi-electric-collaborate-to-develop-3-2t-optical-engines-for-ai-networks|publisher=poet-technologies.com|access-date=2025-11-10}}</ref>

So the critical interface is:

: 

Why coupling efficiency is crucial here
* EMLs are “all-in-one” devices: they generate and modulate the light. You can’t just crank the driver indefinitely without blowing up your power/thermal budget.
* At 1.6T–3.2T, per-lane budgets are tight, lane counts are high, and cooling is already painful. If the EML-to-interposer coupling wastes, say, an extra 1 dB per channel, that’s a lot of extra laser power and heat across dozens of lanes.
* Mitsubishi is bringing premium EMLs to the party; if the interposer coupling were mediocre, Mitsubishi’s performance advantage would be partially squandered.

So in this project, coupling efficiency is:

: 

If they pull it off at volume, it’s a strong proof-point that passive alignment + Optical Interposer can play in the real top-end of the market, not just “cheap pluggables”.

==== What they are doing together ====
* POET and NTT Innovative Devices are co-developing a 100G bidirectional optical engine using the Optical Interposer and NTT’s optical components, targeting AI-enabled mobile networking (front-/backhaul, etc.).
* The design aims for 4× higher bandwidth efficiency than current solutions (squeezing more Gbit/s through the same fiber resources). Stock Titan<ref>{{cite web|title=Stock Titan|url=https://www.poet-technologies.com/news/poet-technologies-partners-with-ntt-innovative-devices-on-next-gen-connectivity-solutions-to-support-ai-mobile-networking|publisher=Stock Titan|access-date=2025-11-10}}</ref>

Key interfaces:

: 

Why coupling efficiency is ''even more'' sensitive here
* Bidirectional means TX and RX share one fiber. Any loss in combining/splitting optics (WDMs, filters, couplers) eats directly into both directions’ budgets.
* You have more optical elements in the path (combiner/splitter, filters), so the link budget per interface is smaller; each coupling needs to be low-loss and stable.
* Mobile gear is often in tighter power and temperature envelopes than cloud data-centers.

So for NTT:
* Good coupling efficiency at each interface = → feasible 100G bi-di on real network fiber, → with realistic power consumption in radio or transport equipment.
* If POET’s passive alignment required, say, 1–2 dB extra per interface vs a finely tuned active design, the entire 4× “bandwidth efficiency” story could become fragile.

In other words, here coupling efficiency is a core enabler of the “do more over the same fiber” pitch, not a nice-to-have optimization.

==== What they are doing together ====
* Sivers brings high-power DFB lasers, POET brings the Optical Interposer.
* Goal: scalable, energy-efficient light-engine solutions up to 6.4 Tb/s, including external light sources (ELS) for co-packaged optics (CPO) and future pluggables. cashumarkets.com<ref>{{cite web|title=cashumarkets.com|url=https://www.sivers-semiconductors.com/press/sivers-semiconductors-partners-with-poet-technologies-to-deliver-innovative-light-engines-and-strengthen-serviceable-market-offerings-in-next-gen-ai-infrastructure/|publisher=sivers-semiconductors.com|access-date=2025-11-10}}</ref>

Conceptually:

: 

Why coupling efficiency is almost ''the'' main story here

For ELS and CPO:
* You often have one light source feeding many channels (fan-out). Every dB lost near the source gets multiplied across all paths: - Lose 2 dB right after the lasers? That’s ~37 % of your total power gone before it even reaches the fabric.
* CPO targets insanely low energy/bit (sub-pJ) and tight thermal envelopes right next to GPUs/ASICs. There is almost no slack for “wasting” optical power in bad coupling.
* External light sources are being sold as a way to improve reliability and reduce thermal load inside packages. Too much coupling loss undermines both: you need more (or stronger) lasers and more cooling.

So for Sivers + POET:

: 

If POET’s passive alignment can hold low loss and tight variation across many channels and wafers, that’s practically a green light for ELS/CPO architectures. If not, ELS quickly looks like a science-fair project rather than a commercial winner.

==== What they are doing together ====
* Celestial AI co-developed a custom Optical Interposer platform with POET.
* POET supplies Starlight light engines, which Celestial uses as precision optical power sources for its Orion AI accelerator / Photonic Fabric. YouTube<ref>{{cite web|title=YouTube|url=https://www.poet-technologies.com/news/2022-feb-10|publisher=poet-technologies.com|access-date=2025-11-10}}</ref>

The architecture (simplified):

: 

Why coupling efficiency directly hits system-level AI metrics

Celestial’s value proposition isn’t “we sell transceivers”, it’s:
* Higher effective bandwidth between AI chips than electrical links can provide,
* At lower energy per bit.

If POET’s light engines have poor coupling:
* You need more lasers / more laser power to hit the same on-chip optical power → higher TDP for the AI box.
* Fan-out to multiple chips/tiles consumes more optical power on splitters and interfaces; any extra insertion loss at Starlight eats into what’s left for the actual compute fabric.
* System designers may be forced to shorten reach, reduce radix, or increase cooling — undermining the “better than copper” story.

That’s why Celestial’s CEO talks about “precision optical power sources in a highly integrated form factor”: the “precision” part is both power stability and predictable, low insertion loss from source into fabric. semiconductor-today.com<ref>{{cite web|title=semiconductor-today.com|url=https://www.semiconductor-today.com/news_items/2022/feb/poet-110222.shtml|publisher=semiconductor-today.com|access-date=2025-11-10}}</ref>

So in this relationship, coupling efficiency is a hidden but crucial factor in the AI system’s total flop/€ and flop/W. It’s not just a module spec; it’s part of whether optical fabrics actually beat Nvidia’s internal electrical fabrics on real workloads.

==== You didn’t mention them, but worth a quick nod: ====
* POET integrates Semtech FiberEdge® receive technology into 1.6T optical receivers at 200 G per lane, targeting high sensitivity and power efficiency. Finviz<ref>{{cite web|title=Finviz|url=https://finviz.com/news/179333/poet-technologies-and-semtech-launch-16t-optical-receivers-for-ai-networks|publisher=finviz.com|access-date=2025-11-10}}</ref>

Here coupling efficiency is about:

: 

If coupling is inefficient:
* You waste the very sensitivity advantage FiberEdge brings, and
* You may need higher launch power on the other end, increasing total system energy.

So again, coupling efficiency is a quiet but important piece of the promise “high sensitivity at low power”.

==== Across these “component-level” partners (Mitsubishi, NTT, Sivers, Celestial, Semtech) vs the “module-level” customers (Luxshare, Mentech, FIT, BFYY), coupling efficiency plays different roles: ====
* For Mitsubishi, NTT, Sivers, Semtech → It’s a joint technical validation: can POET’s passive, wafer-scale interposer preserve the performance their premium lasers/receivers are capable of?
* For Celestial AI → It’s a system-level economic factor: can Starlight deliver enough usable optical power, with low enough loss, that an optical fabric genuinely beats electrical fabrics on cost, bandwidth and watts per bit?
* For Luxshare, Mentech, FIT, BFYY → It’s a module-level cost/power knob: better coupling means cheaper modules, lower TDP, and more comfortable link margins.

The uncomfortable but honest point:
We still don’t have public, apples-to-apples dB numbers for these interfaces. What we do have is:
* A growing list of serious partners in very demanding applications, and
* No public sign yet that coupling efficiency is a blocking issue.

If you want, next step could be: we sketch typical link budgets for each use case (Celestial-type optical fabric vs NTT bi-di vs Sivers ELS/CPO) and mark where POET’s coupling sits in those budgets — that often makes it easier to see how much “slack” they probably have and where the real risk is.