MGT Power supply

Hi Heiner, I'd be interested in the response you get for this question. As linear regulators have a bandwidth of a 100kHz or so, I fail to see how they provide an advantage over a filtered switcher.

Is that in DS302 anywhere? Or the power estimate spreadsheet thingy?

UG076 table 6-1 gives some indication of the filtering required from which you might be able to make some assumptions.

You might want to get a development board and try removing some of the bypass stuff to see what happens. I doubt Xilinx will support what you suggest, but that doesn't mean it won't work. Good luck. Syms.

Hi,

Switching power supplies introduce ripple voltage on the voltage rails they are regulating, requiring multiple LC stages for filtering which further slows down the regulator's response time.

Linear regulators reject this ripple (act like an active filter) but they have finite tracking bandwidth which limits their usefulness beyond

100kHz. Linears are a good choice when the front-end switcher operates at lowish frequencies ( Hi Heiner,

--
Daniel Sauvageau
moc.xortam@egavuasd
Matrox Graphics Inc.
1155 St-Regis, Dorval, Qc, Canada
514-822-6000

voltages.

Hi Daniel, Thanks for your post, but I'm still somewhat confused. I agree that linear regulators only work well at removing noise at low frequencies, and that passive filtering is more effective at high frequencies. So, why not use a fast switcher, say switching at a few MHz, and passive filter its output. The passive filter wouldn't have a problem with "potentially nasty stuff across the whole spectrum". Are you saying that the MGTs are mostly adverse to low frequency noise? That seems somewhat strange for a multi-GHz device. BTW, I always make sure to defeat any burst modes in switcher circuits to reduce low frequency ripple. Thanks, Syms.

Using a high-speed switching regulator helps the power filtering requirements and the switcher's ~25kHz sense circuit bandwidth may be able to do a decent job for line-load regulation too but this still leaves a gap in the 25-100kHz area where passive filters are cumbersome.

BTW, when output drivers and internal logic switch, they too will generate noise on the power rails and this noise can span a very wide range depending on switching patterns. If I drive a video DAC with alternating black and white lines, I can generate some (barely measurable) noise down to half of what my horizontal sync frequency is. Every single-ended will generate similar noise on the power rails in addition to high-frequency transition spikes.

Since MGTs have dedicated power pins, it is reasonable to presume some special requirements have to be met. I looked at the XUP-V2P schematics, the MGTs on this board are powered using an LDO, bulk output capacitor and 1uH inductors between bulk and each MGT power pin... looks like the MGT circuits on these dedicated power pins are primarily constant-current.

Now, which frequency band are MGTs most sensitive to? I have no idea. But given that voltage variation will induce jitter and mess up thresholds in the clock recovery circuit, I would stick with LC to eliminate high-frequency noise down to ~75kHz followed by a linear to remove everything else. Spending $5 more to avoid junking a $100+ PCB and a $300+ FPGA sounds like a good deal to me - at least for smal-scale productions and prototypes.

--
Daniel Sauvageau
moc.xortam@egavuasd
Matrox Graphics Inc.
1155 St-Regis, Dorval, Qc, Canada
514-822-6000

from XILINX:

- Switching regulators can be used up to 3.12 with extensive passive filtering, for higher frequencies linears should be used.

- Unused MGT do need power supply but no filtering, actually they can be powered from the FPGA core supply.

- I still have no answer on how much current unused MGTs need

heiner