Xilinx VIIPro power supplies

you must keep them seperate. The UG states this.

Does it really? So I should have one 2.5V regulator for the RIO and another regulator for the 2.5V Vccaux etc.? I didn't realise.

Thanks,

Rog.

another

Extract from "The RocketIO=99 Transceiver User Guide UG024 (v2.5) December 9, 2004"

"PCB Design Requirements (Page 109)

To operate properly, the RocketIO transceiver requires a certain level of noise isolation from surrounding noise sources. For this reason, it is required that both dedicated voltage regulators and passive high-frequency filtering be used to power the RocketIO circuitry."

If you dont use the RIO's you still have to supply power, but you can use the VCCAUX supply in this case.

Hope this helps clarify the situation

Jason

Thanks for your replies. I was never considering 1 regulator per RIO but having never built anything that uses RIO before I'm keen to know what is considered best practice. I was just asking for opinions on the need (or otherwise) for a RIO regulator and a second "everything else on 2.5V" regulator.

Looking at the recent replies, there seems to be some confusion. However as the UG seems to imply separate 2.5V regulators, maybe that's the way I should play it.

Thanks,

Roger

"PCB Design Requirements (Page 109)

To operate properly, the RocketIO transceiver requires a certain level of noise isolation from surrounding noise sources. For this reason, it is required that both dedicated voltage regulators and passive high-frequency filtering be used to power the RocketIO circuitry."

If you dont use the RIO's you still have to supply power, but you can use the VCCAUX supply in this case.

Hope this helps clarify the situation

Jason

Roger,

The way I understood it, and therefore implemented it was to use a single linear regulator (LT1963) to power all of the RIO on the FPGA. If the LR is capable of supplying more than one FPGA's RIO circuitry, then that is acceptable. As long as all of the RIO supply pins are individually filtered with ferrite beads (and caps when they are not embedded), this should work. Under no circumstances should a switching regulator be used to power the RIO. Also, do not use the same LR that powers RIO to power the internal logic or IO of the FPGA.

As you said in your earlier post, a linear reg for RIO, and a seperate reg for everything else.

Regards

Jason

Jason,

Thanks. That's very clear.

Regards,

Roger.

Clear it may be, correct it's not. Switching regulators are just fine for RIO, as long as you provide for adequate filtering on the supplies. This may be easier than providing adequate heat sinking for linear regulators. As I said earlier. One big switching regulator for all your 2.5V, proper isolation and filtering to each RIO, Vccaux, each Vcco will work just fine. Sorry, I just get worked up when folks say 'under no circumstances'. It may well be that linear regulators are right for your design, but you should consider the options available to you. Cheers, Syms.

Syms,

Thanks again for your input.

I was actually intending to use linear LDO regulators in actual fact, as power and heat dissipation aren't an issue. However switching vs linear isn't really my query here. Basically, are you saying that I could supply my

4 RIOs, 4 x 3 Vcco pins and 4 Vccaux pins all from the same 2.5V LDO regulator (AVccaux RIO pins suitably filtered with Ferrite beads and capacitors of course)?

Thanks,

Roger.

Symon please don't take any offense at this, but this information is wrong and would like lead a designer into a serious issue with their high speed serial links if they were to take your advice.

For the record, Xilinx design requirements for the RocketIO supplies are that the supply is sourced from a linear regulator with capacitive filtering as noted in the RocketIO user guides. In addition each RocketIO voltage is to be isolated from the supply with a ferrite bead.

Depending upon the family and package type you may also need another cap on the FPGA side of the ferrite bead as noted in the RocketIO user guides.

Yes, you can have one supply to be the source for all RocketIO and you do not need to have individual regulators for each RocketIO or for each supply type.

We make these requirements for your benefit and link quality across all possible conditions.

In the real world, I have seen a few cases where switching power supplies have worked. I have also seen far more cases where they are the root of the problem in a totally dead link or have produced a marginal link.

Ed

Sym> Clear it may be, correct it's not. Switching regulators are just fine for

Hi Roger, Yes, that's what I'm saying. You will need to pay close attention to layout and filtering. I would not lay out the 2.5V as a plane. A plane couples everything together tightly, which isn't what you want in this case. Route it separately, i.e. star format, from your single LDO, through a ferrite and/or a low value resistor (10s of mOhm), then to any bulk capacitance you need, and then on to each of the sections of your circuit where you need

2.5V. When it gets to the destination, make a mini plane, add X5R ceramic decoupling capacitors to taste! The low value series resistor can save the day when you have any low frequency noise that the ferrite can't attenuate. Also, it's good for measuring current. If you find you don't need it, you can always short it out later! For bulk capacitance, I recommend Panasonic's specialty polymer parts. 5mOhm ESR! In my experience, the Xilinx app. notes are written to provide belt and braces solutions. If you do it their way, it'll work, but remember their app. notes are trying to cover a large number of design variations. They can't consider everyone's different requirements, so it's up to the designer to tailor their own solution. Good luck mate, Syms.

my

Ed, No offence taken Ed! Especially as you agree that switching regulators can work later on in your post. I did write "as long as you provide for adequate filtering on the supplies". I realise that you guys write your App. notes to protect your customers from when designs go bad, and you err on the side of safety, but all I'm trying to say is that it can work. I accept it may be a more risky route. Some ways to reduce the risk are:-

1) Defeat any burst mode the switcher may have. The bursts can give bad LF noise. 2) Provide for resistive low frequency filtering. 3) Switch as fast as possible, some modern switchers switch at several MHz, easier to filter. 4) Don't skimp on the layout effort and bypassing. 5) Don't use a board wide supply plane.

I wonder if you see mostly bad switcher designs because no-one rings Xilinx to complain when the switcher design works! Cheers, Syms.

[...]

If I used a linear regulator, it would be fed from a higher voltage from a switcher. I've observed that linear regulators provide almost no filtering of the switching noise, because their frequency response doesn't extend that high, so apparently the recommended capacitive filtering takes care of that? If so, why doesn't the capacitive filtering work adequately when using just a switcher?

Thanks, Eric