Looking for (Optically) isolated data transmission at 500+MBits over ~60cm

[Ysjoelfir]

Hi everyone,
a recent project I am challenged with is the transmission of data with (minimum) 500MBits over a high voltage isolation. Since this is going to be used in a commercial product for my employer I can't provide much data but I will try to describe the use case as good as possible while being as vague as needed

In the device we have a main processiong unit (basically an embedded system) which, in the recent generation, communicates with the interchangeable channels over a PCI-e backplane. The channels have a seperate FPGA on them, talking to the main processing unit.
Data coupling is direct right now, but we want to up the specification of the channels to be able to withstand over thousand volts, so isolation is becoming problematic. The new design uses a floating secondary supply for the FPGA on the channel, so data coupling has to be potential free.

It is possible to omit the backplane and couple the data directly to the processing unit, which would need around 60cm of distance.
Another possibility is to stay with the backplane, which reduces the isolation distance to around 8cm.
Also, it should be rather compact, using transceivers intended to be used for fiber channel networks isn't an option.

I have already looked at products from Broadcom/Avago, Foxcon and Finisair but didn't find many solution that meet all the specs, yet. We have some Fiber Optics Transceivers unter evaluation but since this process may take some time we would like to try several systems at once, so input is required here.

Cost isn't a direct limiting factor (sure, cheaper is always better), if it provides more long term stability, since the devices are build for quality, performance and reliability rather than price.

Other solutions as fiber optics are indeed possible, I have, for example, read about PCBs with integrated optical core, that would be a solution. Has anyone experiences with that, and knows about manufacturability, reliability, cost,..?

Data coupling using inductive elements was done in the past (using clip on ferrites and PCB inductors) but didn't meet the required specs for speed, also the failure rates while manufacturing were rather high and as the devices are sometimes used in environments that are subjected to vibrations and shock, I personally would like to use a more robust system.

Optocouplers... well, despite the needed data rates, the reliability wasn't to great in the past, so the team doesn't want them around any more

Thanks for all your ideas!

edit: corrected some phrases which sounded more restrictive than I meant.

[Marco]

Specifically, what's wrong with say AFBR-59F3Z? I know you looked, but it helps to give people a little more concrete reason why the products didn't meet your expectations.

[Rerouter]

I suppose the other part is, you say high voltage, but is this 1000V, 10000V or 50000V+ as at a point it becomes uncoupled laser diodes / recievers where its just placed on the boards with only air between them.

[Ysjoelfir]

Specifically, what's wrong with say AFBR-59F3Z? I know you looked, but it helps to give people a little more concrete reason why the products didn't meet your expectations.

Exactly those are right now under evaluation. I have edited my initial post since it sounded a bit to restrictive.

I suppose the other part is, you say high voltage, but is this 1000V, 10000V or 50000V+ as at a point it becomes uncoupled laser diodes / recievers where its just placed on the boards with only air between them.

Since this is one of the main selling points of the product I cant clarify this to much, but: over 1kV, quite a bit unter 10kV

[Whales]

You have PCI-E?  Anything running Linux on this embedded system or it is a more custom OS?

PCIE SFP card + 1Gbit SFP module + fibre cable.  If you're running linux then you can find something with existing driver support.  All off the shelf parts.

[Ysjoelfir]

That would be possible in theory, but the problems are:
- Size!
- We would need to re-engineer the whole communication system
- we would need several cards, as the device can have several channels, each needing its own (two) communication interfaces
- I guess that would be pretty expensive as well.

[capt bullshot]

Did you read this paper:
https://www.cleverscope.com/files/The%20Cleverscope%20CS448%20Development%20Journey.pdf
?

[Ysjoelfir]

I have not but I will read it now as it looks very promissing

[Marco]

Did you read this paper:
https://www.cleverscope.com/files/The%20Cleverscope%20CS448%20Development%20Journey.pdf
?

Those Murata FOT's look nice, but they seem to have been memory holed. A couple news releases is all that's on the web. Too good to be allowed to be sold for appropriate prices to plebians? When you look at normal 10+ gbps stuff they want to absolutely rape your wallet, I don't think cleverscope paid those kinds of prices.

[jhpadjustable]

If they don't like inductive isolation, how about capacitive? There are digital isolator ICs such as the Si86xx series, with up to 6 channels across up to 5kV in a wide SOIC-16, but they're too slow unless you have a bit of room on the board to (de)mux the data stream over multiple lanes.