04-09-2019 03:29 AM
Hello dear engineers!
I have a complex digital design which is made of differnet modules and alot of FPGA resources are occupied. The design passes simulation and everything is ok. But when I programm FPGAs I see a weird issue! I have two boards with Kintex-7 FPGAs. I program an identical bitstream on two boards and in one of the boards I can't see the ILAs and a warning is shown that means there's a problem about the ILA clocks. I say again, that a completely identical bitstream has different behaviours on two different devices and I have faced with the sameproblem before, where different behaviours are seen on different FPGAs. I should inform that on the board which ILAs are not shown, other bitstreams can run correctly!
I know that Xilinx devices are reliable. (As we see here.) and I think there's something in my designs which I don't pay enough attention to!
I should add that I have checked almost everything out and there seems to be no problem, Issues like: Timing, possible defects on the boards, and ...
Any help is appreciated.
Regards, Herman Fisher
04-09-2019 03:59 AM
Are these custom boards, or off-the-shelf ones?
If you have a small timing failure in the bitstream (or undefined timing for I/O) then you're largely at the mercy of random chance. Sometimes it'll work, sometimes (even on the same device) it might not - just depending on temperature, voltage, etc. If you're operating the chip at a slightly incorrect voltage (or with more noise on the supply rails, or different decoupling capacitors, etc) then Xilinx makes no guarantees at all about how/if the chip will behave.
04-09-2019 04:43 AM
Thanks for your answer.
The board is a custom one.
There's no timing failure and the chip voltage is completely correct.
Do you have any idea of what other possible causes? Is there anything else, rather than timing and board voltage, to check out?
04-09-2019 05:41 PM
Check the clock source for your board/FPGA with a spectrum analyzer. -or try another clock source. -or swap clock sources for your two boards.
We recently started buying clock sources from another vendor. The new clock sources have the same jitter specification as the old clock sources. However, FPGA boards using the new clock source had unusual problems - similar to what you are seeing.
We eventually found the problem to be a high 2nd harmonic coming from the new clock sources. Filtering off the 2nd harmonic with an inexpensive low-pass filter solved the problem.
-and, as I'm sure you're doing, tell Vivado how much jitter is on your clock source.
04-09-2019 09:21 PM
About your last sentence: Yes, I'm doing this. And I will try checking the clocks more precisely to find out is anything to do or not.
04-09-2019 11:48 PM
04-13-2019 08:52 PM
Thank you for your answer Yash!
Maybe that's correct, but what can we say when another bitstream with some different ILAs runs correctly on the board which seems to be damaged?
This test says that the problem can't be related to hardware!
04-14-2019 01:02 AM - edited 04-14-2019 01:03 AM
If all else fails, it's possible that one of the FPGAs is damaged - and it may just be a single path which affects one bitstream but not the other. Do you have the ability to swap the FPGA and re-test on the same board?
Edit: actually ... grab a good microscope and check the decoupling capacitors. We've seen a few boards where the capacitor dielectric was cracked (which is invisible to the naked eye for most SMT components) and causing all sorts of fun issues.