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

=== Assistant: Short version: ===
Short version:
* Historically, active alignment has been the go-to when you want absolutely maximal coupling efficiency.
* With modern interposers (like POET’s), passive alignment has closed a lot of that gap.
* POET claims industry-leading low coupling loss, and their products clearly meet 400/800G specs, but there is no independent, apples-to-apples data set that says “POET is X dB better/worse than vendor Y”.

I’ll break it down.

==== Coupling efficiency η is simply: ====

: 

Usually quoted as loss in dB:

: 

Examples:
* 0.5 dB loss → η ≈ 89 %
* 1.0 dB loss → η ≈ 79 %
* 3.0 dB loss → η ≈ 50 %

In a real module you have several such interfaces:
# Laser → interposer waveguide
# Waveguide → on-interposer devices (modulators, PDs, AWGs, mirrors, etc.)
# Interposer → fiber (or to a silicon photonics chip, depending on architecture)

The system link budget cares about the sum of all these losses plus fiber, connectors, etc.

==== From neutral sources (e.g. Physik Instrumente, EPIC/AEMtec): ====
* Passive alignment uses highly precise mechanical features (etched pockets, fiducials, stops). It is: - Simple and cheap - Great when tolerances are “moderate” - But can struggle when you need sub-0.2 µm precision, especially in some cutting-edge silicon photonics couplers. pi-usa.us<ref>{{cite web|title=pi-usa.us|url=https://www.pi-usa.us/en/tech-blog/photonics-alignment-techniques-active-passive|publisher=pi-usa.us|access-date=2025-11-10}}</ref>
* Active alignment: - Moves parts in x/y/z/θ/φ while a detector measures optical power in real time - Can optimize whatever coupling interface you have and get very close to the theoretical maximum - Is slow and expensive in high-volume production

State-of-the-art fiber–chip couplers in research:
* Typical commercial silicon-photonics grating couplers: often ~1–3 dB loss (50–80 % efficiency). opg.optica.org<ref>{{cite web|title=opg.optica.org|url=https://opg.optica.org/prj/abstract.cfm?uri=prj-7-2-201|publisher=opg.optica.org|access-date=2025-11-10}}</ref>
* Advanced multi-layer grating designs in foundry processes: ~4.7 dB loss for very broadband/dual-pol devices (≈34 % efficiency) – optimized for bandwidth and manufacturability, not just minimal loss. opg.optica.org<ref>{{cite web|title=opg.optica.org|url=https://opg.optica.org/abstract.cfm?uri=oe-30-17-31058|publisher=opg.optica.org|access-date=2025-11-10}}</ref>

So my earlier statement “active alignment generally results in better coupling efficiency” is only strictly true if:
* You use the same optical structure and
* Compare “free” active alignment vs passive placement using no specially designed features.

As soon as you specifically engineer an interposer with precision pedestals, mode-matching structures, mirrors, etc., passive alignment can get very close to what active alignment can do — and sometimes within ~0.2–0.5 dB.

==== ### ====

From POET’s ECOC 2023 slide deck on the Silicon Optical Interposer: ECOC<ref>{{cite web|title=ECOC|url=https://www.ecocexhibition.com/wp-content/uploads/POET-presentation-ECOC2023-RK-sharing.pdf|publisher=ecocexhibition.com|access-date=2025-11-10}}</ref>
* “Wafer scale assembly. Passive alignments.”
* “Passive Laser placement with high coupling efficiency”
* “Mirror finish etched facet technology – Lowest coupling loss in industry”
* “Low loss micro mirrors for out of plane coupling”
* “CMOS compatible low loss waveguides”

A recent primer (investor-oriented) describes: shareholdersunite.substack.com<ref>{{cite web|title=shareholdersunite.substack.com|url=https://shareholdersunite.substack.com/p/a-primer-on-poet-technologies|publisher=shareholdersunite.substack.com|access-date=2025-11-10}}</ref>
* Optical transmission loss under 0.2 dB/cm in their waveguides (that’s propagation, not coupling, but it tells you the internal waveguide stack is very low-loss).

In investor communications that circulated on Reddit, POET has been quoted with a specific figure: reddit.com<ref>{{cite web|title=reddit.com|url=https://www.reddit.com/r/POETTechnologiesInc/comments/pq39df/our_waveguides_are_2x_as_efficient_as_industry/|publisher=reddit.com|access-date=2025-11-10}}</ref>
* Vertical “out-of-plane” coupling ≈ 0.5 dB (i.e. ~89 % efficiency), and
* Claim that their waveguides are “2× as efficient as industry standard”.

Important caveats:
* These are company / investor claims, not peer-reviewed metrology papers.
* They refer to specific interfaces (e.g. micro-mirror or MMF/PD coupling), not the entire module’s total loss.

However, they are at least consistent with the ECOC slides that emphasize “lowest coupling loss in industry” and “high coupling efficiency” with purely passive laser placement. ECOC<ref>{{cite web|title=ECOC|url=https://www.ecocexhibition.com/wp-content/uploads/POET-presentation-ECOC2023-RK-sharing.pdf|publisher=ecocexhibition.com|access-date=2025-11-10}}</ref>

===== A vertical coupling loss of ~0.5 dB is ambitious but not insane for: =====
* Butt-coupling or mirror-based coupling with very good mode matching, and
* Short distances / controlled geometry on an interposer.

If their propagation loss is <0.2 dB/cm, then internal routing contributes almost nothing for the typical few-mm path lengths inside an engine. shareholdersunite.substack.com<ref>{{cite web|title=shareholdersunite.substack.com|url=https://shareholdersunite.substack.com/p/a-primer-on-poet-technologies|publisher=shareholdersunite.substack.com|access-date=2025-11-10}}</ref>

So you can plausibly end up with:
* Laser → interposer: maybe ~0.5–1.0 dB total (depending on interface and lensing)
* Interposer internal → very low
* Interposer → fiber: depends heavily on the module design, but if it’s via a lens array or further optics, that may be the bigger contributor — and in many architectures, that part is actually handled by the module partner, not POET itself.

We also know that POET’s engines are in commercial 400G/800G designs (FR4, DR8, etc.), and are qualified by large customers. lightwaveonline.com<ref>{{cite web|title=lightwaveonline.com|url=https://www.lightwaveonline.com/home/article/55270138/poet-technologies-poet-optical-interposer|publisher=lightwaveonline.com|access-date=2025-11-10}}</ref>
If their coupling were significantly worse than incumbents, the power budgets on those standards would simply not close.

==== For “traditional” discrete TOSA/ROSA: ====
* Active alignment of laser to fiber / lens / SiPh chip can indeed squeeze out the last few tenths of a dB: - You might go from, say, 1.0 dB loss (good passive) to 0.5–0.7 dB with meticulous active alignment.
* But it costs: - High-precision stages - Long alignment cycles - Expensive benches and lower throughput

So the realistic picture for 100–800G pluggables is more like:
* Active alignment: - Maybe ~0.2–0.5 dB better in some interfaces if you really try. - But with significant per-unit assembly time and cost.
* POET-style passive alignment: - Claims coupling losses in the same ballpark, maybe 0.5–1 dB for key interfaces, achieved without active tuning. - You “spend” your engineering effort once in the interposer design and lithography, then reap gains at volume.

So I would now phrase it like this (correcting my previous oversimplification):

: 

==== What we do not have publicly: ====
* A peer-reviewed, independent paper that says: “POET’s interposer achieves X dB interface loss vs Y dB for an actively aligned reference, under identical conditions.”

We have:
* Company slide decks and marketing claims (low-loss waveguides, “lowest coupling loss in industry”). ECOC<ref>{{cite web|title=ECOC|url=https://www.ecocexhibition.com/wp-content/uploads/POET-presentation-ECOC2023-RK-sharing.pdf|publisher=ecocexhibition.com|access-date=2025-11-10}}</ref>
* A few data points quoted in investor channels (0.5 dB vertical coupling). reddit.com<ref>{{cite web|title=reddit.com|url=https://www.reddit.com/r/POETTechnologiesInc/comments/pq39df/our_waveguides_are_2x_as_efficient_as_industry/|publisher=reddit.com|access-date=2025-11-10}}</ref>
* Evidence that top-tier customers are sampling and designing in POET engines for 400G/800G and AI applications, which indirectly confirms that their effective coupling + power budget is competitive. semiconductor-today.com<ref>{{cite web|title=semiconductor-today.com|url=https://www.semiconductor-today.com/news_items/2025/mar/poet-240325.shtml|publisher=semiconductor-today.com|access-date=2025-11-10}}</ref>

If you want to be really strict as an investor or engineer, the key things to watch are:
# Hard numbers in future conference papers or app notes: - Measured coupling loss per interface (laser→interposer, interposer→SiPh, interposer→fiber) - Across temperature, aging, and volume.
# Comparisons to reference designs: - For example, “same laser, same fiber, active vs our passive approach”.
# Customer disclosures: - Sometimes an OEM will say “this engine achieves X dB link margin at Y reach with Z power”, which lets you back-calculate approximate coupling losses.

==== - You are right to zoom in on coupling efficiency — it’s one of the few technical parameters that can really kill the promised cost advantage if it is off by even 1–2 dB. ====
* Based on public info, POET’s claimed coupling performance is compatible with high-end 400G/800G specs, and their interposer architecture is explicitly optimized for low-loss passive alignment.
* The old rule of thumb “active = better efficiency, passive = cheaper” is too crude here: - For ad-hoc geometries, still true. - For purpose-designed interposer platforms, passive can be almost as good on efficiency and vastly better on cost/throughput.

If you like, next step could be to build a simple link-budget comparison:
* Plug in reasonable per-interface losses for an “old-school” actively aligned TOSA vs a POET-style interposer engine and see how much margin and power you gain/lose in each case. That often makes the trade-offs painfully clear.