My Authors
Read all threads
Optics teardown!

And this is probably my most exciting one so far! Meet the CFP optic! 100G Ethernet back when it was still hella hard.
And this thing is a big boi! Bigger than a deck of cards. 82mm wide x 144mm deep x 13mm thick.
Even though this is a 100G interface, it predates the 28G serdes used in the ultimately more popular QSFP28 100G cage, so this relied on 10x10.3Gbps lanes in both directions in the cage, which is... a lot of pins.
This being a 100G-LR10 optic means that all 10 lanes are handled separately as 10 different colors which get multiplexed onto a single Tx fiber.

Haven't been able to find much info about LR10 or what colors they used... Presumably you could also get PLR10 or SR10 with an MPO24
I've heard mention of there being 100G-LR4 optics available in the CFP form factor, which would need a gearbox ASIC on the pluggable to convert the 10x10 to 4x25 for that optical protocol.

🤷‍♂️ This predates me working in networking. QSFP28 had already taken over.
Awww hell yeah! 😍
Brains of the operation looks to be an STM32F103 with a 2Mb 25PE20 SPI flash chip next to it.
The Tx (bottom) and Rx (smaller, top) optical sub assemblies are the largest I've seen yet! Makes sense, since these need to somehow fit 10 channels and the mux in there! 🤯
Between the TOSA/ROSA and the CFP connector are some QFN chips that look to be Gennum GN2406 quad 10G Retimer/amp chips which condition 4x 10G for longer range so you can get through the connector. Presumably three more of these on the bottom of the PCBA for the other 12 lanes
I'm not sure since that discoloration might be from chemicals in the thermal pads, but based on that straw color, it seems like these amplifiers might get a little toasty.
You know, it would probably be a better guess that there's four more amps on the underside, based on those clusters of bypass caps...
You call those screw locks?

THESE are screw locks.
Finding tools SMALL enough to remove all these screws holding down the TOSA/ROSA was rather difficult.
Well... all the screw holes are empty... so why isn't anything coming loose?
Just needed to keep working it. Fuuuuuuck yesssssss 😍
Ohhhh man that's pretty.
Foil backed thermal paste for the TOSA/ROSA!
And those thermal pillars for the four more quad retimer chips.
So yeah, four more 4x10G amps on the bottom to compensate for the connector.
There is nothing about this optical packaging that I don't love.
When there's handwriting on an electrical component, you know it's expensive.
😬 now we're committed.
Very much don't know what I'm doing or how this originally went together, so we're improvising.
🤔 that lid seems to be much thicker than it seemed with the tap test
Thankfully there's more than one tap test...
Oh dang it! I was so close just with sanding!
WHAT THE HECK IS GOING ON INSIDE THIS ROSA‽‽‽
10 way optical demux to 10x10G receivers!!!
Same method for the TOSA. Kind of a tuna can vibe going on here opening it.
Time to pull out the big guns (smol guns?) and ENHANCE.
Ok, so we have microscope images of the receive and transmit optical assemblies, and while we're getting into things I'm not quite sure about... it's pretty dang good!
So first, let us look at the receive assembly!

Remember, this has coming into a single stand of optical fiber with 10 different colors around 1300nm of 10G Ethernet that need to be split apart to 10 different detectors.
The fiber comes in with a lens on it, and then another lens glued to a mounting block (the hazy white cube) glued to the main demultiplexer bed somehow lands the light into an optical channel!
What I'm not entirely sure of is what I'm actually looking at on this mux bed, from a materials perspective. Like, are these etched channels, doped material, etched and backfilled? Really not sure.
After the lens lands the light, it follows this channel, which seems to guide the light to the input of the actual demux
So... I thought optical muxes like these were based on diffraction gratings? But that thing on the left really looks like a concave prism... Doesn't that look like a prism?

Or there might be some kind of diffraction thing going on. I really don't know. Anyways, splitting happens
Once the light is split as a function of color, it looks like there's 13 more optical channels which I'm guessing just spread each wavelength far enough apart for the receive diodes.

Why there's 13 channels for a 10 channel mux is interesting...
I know that 12x10G was also popular since you could optionally split it out as 3x40G, so presumably this assembly is reusing a lot of the same components to save cost
There are a TON of wire bonds going on inside these things!

Here we're looking at what I presume is an amplifier stage since it's immediately after the rx diodes. Again, 12 channels and the outer-most channel on each side isn't used for 10x10G
The substrates look to all be ceramic. Here we see each differential pair get fanned out and then jumpers to the black ceramic of the main optical body, which then comes out of the package as pins.

Notice how there's a ground between each signal pair; important for impedance!
Lots of paralleling up on the power wire bonds on the edge of the rx amp IC. Also some digital logic, so this is probably where the RX DOM info is measured to report each channel's light level in dBm
Whacha doing there, wire bond buddy?

Just... kinda hanging out in the middle of a ground plane?
Here's the raw video of me panning around the ROSA under my scope, in case you wanted to get another look at anything.
The other optical assembly in this 100G optic is the Transmit OSA, which is just as good if not better than the ROSA!

10x10G transmitters at different colors that get combined into a single strand of single mode fiber.
This module takes 10x 10.3Gbps differential pairs from the switch on the edge, and use each of those to drive an LR laser with a laser driver

Which is pretty dang pretty.
Interestingly, each of these driver stages have a lot of notches cut in the substrate to open traces, so presumably those traces are used for manufacturing test/calibration, then cut.
Each of these 10 amplified single-ended signals then get routed to the laser and multiplex assembly and jumped onto the separate substrate as a bond wire guarded by two ground bond wires for each connection.
It looks like the return path for each of the ten lasers is by jumping up onto this little bridge spanning over the floating center assembly, and going through super cute little inductors as chokes probably?
Here's a full view of the jump onto the center assembly, the physical bridge with the inductors, and the row of ten lasers in the middle going into the optical mux on the left.
Finally the optical combiner that takes the ten different colors of 10G lanes (presumably centered around the standard 1310nm) and uses a diffraction grating on the left mirror to combine them all on the single exit route on the bottom.

The launching lens on the bottom left
Here's an angled view of the final lens on the left, which is anchored to the mux by the opaque block on the right, and the launch into the single mode fiber which leaves the assembly.
Here's the raw video of the Tx assembly, if you want a further look at anything.
Missing some Tweet in this thread? You can try to force a refresh.

Keep Current with Kenneth Finnegan

Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!