1. Home
  2. Field-Programmable Gate Arrays
  3. Rocket I/O + Optical Fiber

Rocket I/O + Optical Fiber

Any good info locations online about simply tying Rocket I/O to an optical transmitter and receiver? I'm looking to do near 10 Gbps. We do a lot of stuff at my work that requires transferring lots of data from one box to the next. I'm thinking Aurora or maybe just the generic core. It's Virtex Pro X to Virtex Pro X in a closed system so I don't need any particular protocol.

The Xilinx site has lots of nice info, but they seem to be rather obsessed about backplane applications. My needs are more point to point between units mounted in a rack using optical fiber. There's a mention here and there on the site, but I can't find anything meaty, like an App Note.

I'm looking for info on recommended optical devices, PWB trace design between the FPGA and the optical device, and other point to point application info.

Reply to
Quiet Desperation
Loading thread data ...

Have you looked at XFP? The parts are readily available. The spec includes reference PCB designs, etc.

formatting link
formatting link

You can get expensive 1310nm plugin optics to go a few km, or even more expensive 1550nm plugin optics to go up to 80km. I'm not aware of any cheap 850nm short range plugins though. How far apart are the units in your rack?

Regards, Allan

Reply to
Allan Herriman

We might have up to 20 feet in some cases between different racks. Trivial for fiber, but very spendy when it's a bundle of expensive special order differential, delay matched cables for Gore or Meritec.

I recently did 18 differential signals in parallel at 1 Gbps succeessfully across 10 feet (LVDS signals from a Virtex). The timing window was 500 ps. It works great, thanks to the ability to set clock timing in the DCM, but the cable harness is heavy and big $$$.

I can't imagine four optical components and two fiber cables costing more. Not to mention all the mechanical design issues that go away on the units themselves- both the connector design and PWB routing from the I/O connector to the FPGAs. There's endless savings in parts and labor to be had here.

We're also forced right now to colocate units we might prefer to seperate by some distance.

Thanks for the info.

Reply to
Quiet Desperation

Where are you buying your cables? I find it hard to believe that the cost of the optics + fiber is cheaper than a cable. In my experience the cost is very much cheaper when you stay electrical, when your distances are not very far. Why not use standard cables like those made for Infiniband (meant for 3.125GHz)? There's also a standard for 10G Ethernet over copper cable. There's a reason it exists: cost!

-tom

Quiet Desperati> In article , Allan

Herriman

Reply to
tom

Compared to *a* cable, but I'm coming from a situation where we have custom built, electrically timed/trimmed bundles of differential cables. A harness like that can run well into five figures. And, like I said, given fiber we might want to no longer colocate units that previously were in the same rack simply to deal with the massive data transfer between them. It's a comm system, and there's powerful arguments for putting one of the units as close to the dish as possible, and having the other tucked away in the receive shack. That could be hundreds of feet.

If I can stay electrical, fine. That's why I'm looking for information. I can't get a feel on the Xilinx site as to how much cable a Rocket I/O port can drive. Cripes, it took me all day to discover it's CML. All the datashets and app notes are focused on backplane applications. I'll probably just buy a couple Pro X demo boards and connect them with various lengths of cable, and I guess I need to get a field engineer involved.

Reply to
Quiet Desperation

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.