FPGA power supply design

Hi all,

I'm newbie in FPGA, and want to build my own FPGA board. Actually, I already have my own development board (which is Spartan-3E Starter Kit), but I would feel so guilty if I don't build my own board =P

Actually, which part of the FPGA datasheet that is everyone read to determine which power regulator IC that matches the FPGA ?

I've already read the Spartan-3E's power requirement. All that I can get is that I will need 1.2V, 2.5V, and 3.3V regulator (which can power up in any order), but I don't find current rating for it. The datasheet said that I should consult to XPower for the exact current requirement.

Does anyone do the same thing (measure the current and do simulation in XPower) before determining the power regulator IC ? Or there are some guidelines for typical applications ?

Thank you very much for your help

-daniel

Reply to
kunil
Loading thread data ...

Choosing power supplies for parts is as much art as science, especially in the world of programmable logic where the load can vary due to upgraded / 'fixed' functionality.

Unless you are certain of your power requirements, then provide a supply that can handle the worst case requirements. What *those* are depend on the core [which also determines the number of DCMs] and number of IO pins.

1.2V -> Core 2.5V -> DCMs, JTAG chain 3.3V -> IO

I usually use a linear regulator for the DCM power.

Cheers

PeteS

Reply to
PeteS

Hi Pete, Some comments in line:-

Only if you're a rubbish scientist! C'mon, it's not all that difficult! :-)

Aha, some science. Very good! ;-)

If you're feeding the regulator straight from a battery or a mains/transformer type circuit, the linear regulator works fine. Toasty too! However, if the linear regulator's fed from a switcher, then I suggest filtering out the switching noise with a passive network instead. The control bandwidth of linear regulators is maybe 100kHz tops. A nice LC filter gets rid of crud over a much wider bandwidth. That's science, not art! :-)

HTH, Syms.

Reply to
Symon

Snort - see my comment. The current requirements across the entire range of devices can vary by factors of 5 or more. Certainly one can simulate a board, but keep in mind that the FPGA on a real board is only a small part of the power system. I *do* add up the entire system, and then leave some overhead.

For a straight FPGA, one is rather in the dark because there's very little guidance in the data sheets as to current consumption (quite reasonably, in a way, as it depends on internal switching) and then one has to be careful about current across the entire temperature range.

So is it science? To the extent I can actually get solid numbers, yes. Beyond that, it's art ;)

The DCM supply is usually not a current hog, so for most situations a linear even from 5V is ok. I feed linear regulators all over the place from switchers, and only in the most sensitive caes (which *can* include the DCM if I have to ensure ultra low jitter) do I add extra filtering beyond normal bypassing and perhaps a series inductor.

Science is about figuring out why existing things work. Engineering is the art of figuring out new things ;)

Cheers

PeteS

Reply to
PeteS

Hi Pete, OK, maybe I got the wrong end of the stick, but from your post I thought you meant you used the linear regulator to generate a special quiet supply. I usually have a 2.5V rail in my designs for things other than Vccaux, which made me go off on one! Whatever, just to clarify my point, it's important to note that the only _filtering_ the linear regulator is providing above a few 10's of kHz is due to the resistance of its pass element together with the bypass caps after it forming an RC filter. If you _already_ have 2.5V on your board, e.g. for LVDS banks' I/O supply, I contend that it's better all round to generate Vccaux by filtering this 2.5V rail with passives than to include a separate linear device to regulate down a supply which comes from a switcher. Linear regulators make rotten filters as spending some time playing with Linear Technology's free simulator will quickly show! Cheers, Syms.

Reply to
Symon

Kunil,

formatting link

Use the "XPower Estimator" tool to find the intended power requirements.

Then scroll down to "Partners"

and pick your favorite vendor.

Follow the links.

Really quite simple. Not sure why others are making such a big deal out of this. It is really a trivial issue. The power requirements which require very low noise, are the MGT analog supplies, and there is a huge user's guide to help you there (as well as manufacturer part numbers, and sample schematics).

Every one of these vendors has met with Xilinx, and had their products reviewed. These vendors are used on our own boards.

The only really difficult part is estimating the power you will use, because we have no control over what you program into our parts. All we can do is provide the tools to make estimates (which are based off the quality of information you give them). So, if you put garbage in, you get garbage out.

Austin

Reply to
Austin Lesea

Now I'm home....

Hmmm - it's not a big deal if a design is strictly specified and simulated long before it's laid out. That may be possible (indeed, is, I am sure) within Xilinx (and all the other vendors), but out here in productland the PCB designer is left with, at best, a [hopefully educated] guess simply because the design is nowhere near complete before the board goes to layout. I prefer at least a SWAG to a mere WAG.

By that time, the power has to be specified of course. That's why I overspecify it (I have a specific bag of tricks I use for Xilinx FPGAs just as I have them for other vendors) based on core size, IOs, IO speeds I expect etc., but I don't normally get power estimates from the FPGA designer (when it isn't me) until long past the time it would be useful for the original design stage.

Then there's feature creep. What happens if I specify the core power minimally (from a design) to save money and then I suddenly find we're going to drop a high speed arbitrator into the device ?(your choice of a large piece of functionality here).

I'm sure I am not the only one that has to deal with that, but it is the reality of product design, whether we like it or not.

My answer is to have power designs that cover the worst case for all three of VIO, Vcore and Vaux.

Cheers

PeteS

Reply to
PeteS

I would hasten to add that the power estimator is a great tool, even after the fact, to help with overall board power calculations (where thermal issues might be handled after layout).

Cheers

PeteS

Reply to
PeteS

Dear all,

Thank you for all of your suggestions.

Regards,

-daniel

Reply to
kunil

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.