X4000 bad configuration

It's been a long time since we've used 4000 series parts here, but we have had a similar issue with Virtex (original) parts. It turns out the bad lots were counterfeit. If you got these parts from a franchised Xilinx distributor, your parts are most likely not counterfeit. In that case I would suggest attempting to return them and to open a web case with Xilinx as well.

However if you got them through a part broker, which is common for older parts like these, it is possible that they are counterfeit. In the case of the counterfeit Virtex parts, the actual parts were marked with the Xilinx part number and logo, but the chips inside had no resemblance at all to the part marked, and in fact caused a direct short from +3.3V to ground on our boards.

Good Luck, Gabor

Jacques GEN> HI,

Gabor a écrit :

Effectively, I got those parts from a broker. I do not think they are counterfeit because they have a part of the behaviour I expect from "good" components. They do not seem to be dumb stones in packages. Anyway, I'm OK : when buying obsolete parts, one must be more and more carefull !

Thanks, anyhow Jacques

Woww...I 've seen virtex parts doing the configuration properly in master mode (CCLK, PGM, INIT...), but DONE never goes high, it took a while to figure out the part was a counterfeit, cost for corrections is far beyond one could expect :(

"DONE did not go high"

Could also describe a piece of toast at the end of a JTAG cable.

The I package in question could be completely empty, and, yes,

"DONE did not go high"

There is a sequence of events (CCLK, INIT, etc.) that should be present if the part is really a Xilinx part inside that package.

Please provide more information, as "DONE did not go high" doesn't provide enough information to even tell if there is a die in the package.

I am sure there are those here on the newsgroup who would like to help, but with just the one symptom, everyone is placing bets the part is counterfeit.

Sorry.

Austin

Agree, those pieces could be anthing, but working

I wouldn't buy that. However, one might be thinking why did they bother to have an "intelligent statemachine" in a fake?

Sorry, no more information

You bet.

In my case (old XC4010E) the first symptom was that M0, M1, M2, left floating, did not polarize high as expected. So the parts entered master parallel mode with address pins toggling. When I put external pull-ups, the parts entered slave serial as expected and DONE went high - I did check that DONE came high on due time and not before the configuration file had been completely sent. After that configuration, my board did not operate well, but I did not investigate further. Using good parts, I checked that those external pull_ups did not prevent my board from correct operation.

JAG

Comments below,

That could be a completely empty case, too. No pullups. No nothing.

Only what you expect a real part to do. An empty package would not do this.

This is a good sign. It must be a Xilinx FPGA if DONE goes high, and does so exactly after the serial bitstream finishes entering the part!

OK. "Did not operate" in what way? Everything remains tristate? Did INIT go high indicating a bad bistream (CRC error)?

Not sure what this tells us. On a good part, it will configure serially, and work? That seems to be what you are saying.

Are you sure the parts are identical? There have been various "flavors" of 4K parts (4, 4E, 4X, 4XL, 4XLA, 4XLV...). They are not the same from a bit stream point of view.

Austin Lesea a écrit :

My board-level tests failed. I did not investigate further at chip-level. I'll do.

All my board-level tests were OK.

Sure (XC4010E-3PC84).

All my bad components had been delivered in tape. I had noproblem with parts delivered in sticks. Is there any reason for tape-delivered parts to worse ? My bad components seemed to be older. Is there a way to derive fabrication date from markings, for example XC4010E PC84CKM0141 A2083146A 3C ?

Thanks for any help

JAG

The date code is after the package type on line 2, in the above case 0141 or 41st week of 2001.

See answer record #1067

Gabor a écrit :

Thanks for that; Oups ! My bad parts are the most recent I ever got...

Peter Alfke a écrit :

I designed a PCI board with 3 XC4010E on it (PCI with AMCC part, not in FPGA). The first FPGA is configured asynchronous peripheral. The other FPGA are daisy chained and configured slave peripheral (cf. Data book p. 4-74). During ten years, I had had no problem with that design. Since the beginning of this year, as XC4010 is a discontinued part, I buy it more and more often from brokers. I got three deliveries of "bad" parts. Using those "bad" parts, configuration pins M0, M1, M2, if left floating, do not all polarize high as they should, because of internal pull-ups. Then, the parts do not enter the configuration mode implied by my design. If I set external pull-ups, configuration pins polarize correctly and configuration process seems to be successful, but the behaviour of the part is wrong, ie my board does not work. I did not investigate much at this time, but the pins seem to remain floating and the parts do not start working. Though, these "bad" parts are not counterfeits, as would be empty cases, because configuration process progresses correctly : DONE raises on time, with no external pull-ups, parts may enter master serial with address pins toggling.

The only difference is that "good" parts had mainly been delivered in sticks and "bad" parts had always been delivered in tapes.

Further, "bad" parts were recent, but seemed older (oxyded pins).

I'd like to know whether someone else experienced same problem.

Jacques GENIN

I still remember what we used to suggest 10 years ago:

Look at the Dout pin. There, the serial bitstream is passed on to the next device, but only after the device otself has received its configuration. So you should see the preamble' length count etc, and then a continuous High level, until the end of the internal configuration. And then the remainder of the concatenated bitstream is passed on to the next device.

You have > Peter Alfke a =E9crit :

Peter Alfke a écrit :

Too late, now. I'll do that tomorrow...

That's the most likely... Do you know the source? The brokers won't tell... I'd better buy elsewhere...

Thanks Jacques GENIN

Are grey market parts a serious problem these days? What does "grey" actually mean?

I remember in the old TTL days when it was easy to get parts that were out of spec.

I'd assume Xilinx would have pretty good control over the factories where their chips are packaged/tested. Too much trouble with leaks and the factory looses the contract. But maybe I don't know how things actually work in places where labor is cheap.

I'd expect to find real chips getting recycled when startups go out of business. Some of them may not have been handled correctly so I wouldn't be surprised by problems if I got chips from a non-official source. But they have to be somewhat careful or their reputation for junk will get out.

Is there a mechanism to make a whole batch of chips go bad without visible damage? I'm assuming idiot or short cuts rather than malicious. The sort of thing that would happen in a startup about to go under.

I thought I read Peter mention something about humidity -- if the humidity isn't kept right during storage, something bad can happen to the pads, so when the part is soldered down to a PCB something won't be right...or something.

-Dave