Screening bad FPGA parts

[Gribo]

Hi all,
 A customer of mine is using Spartan 2 parts in BGA package. Recent batches have >20% failure rate for the FPGA - IO failures, no JTAG etc.
I want to screen them before assembly and plan on using an Ironwood BGA socket (6K USD) with the target board. There is no high speed requirement, the design is mainly SPI and IO buffers, so the socket will not affect it much.
My assumption is that if a part is passing this functional test, it will be good after assembly. Am I missing something? I would have liked to do a full boundary scan of all the balls, but I don't have this capability - only a Xilinx USB programming adapter.

[mon2]

Be aware of the possibility of old stock having soldering issues / moisture contamination. Unless the parts are in the factory original packaging, you may / should bake the parts before use in the soldering process to dry out the moisture.

Also, you may face soldering issues with older stock - like we have recently with SEALED TQFP devices in our warehouse for ~10 years of this EOL component. The CM claims many of the parts cannot be soldered. We will inspect each and every PCBA upon arrival but this is a risk that must be considered.

[Gribo]

Thanks, The components are sourced from the customer. They are sealed, with the humidity tags and silica gel. We had the same issue recently with NXP68332 parts.. Baking is now a standard process for them.

[AndyC_772]

There is no high speed requirement, the design is mainly SPI and IO buffers, so the socket will not affect it much.

Do bear in mind that whether or not a signal "is high speed" for signal integrity purposes depends entirely on its edge rate in V/sec, which is a function of the driver, and not on the number of edges/sec (ie. the clock speed in MHz). The fact that a particular interface "is slow" from a functional point of view (ie. low useful data bandwidth) makes no difference at all.

If you have a high failure rate from one particular batch, it sounds as though it would be worthwhile doing a full root cause analysis on one or two boards. Send them away for X-ray, check for popcorning or soldering issues. Find exactly which balls aren't working, and (crucially) why those particular balls don't work. Chances are even if different balls are affected on different PCBs, the root cause is still the same (moisture, ESD damage, oxidation etc).

[Gribo]

The parts are baked prior to assembly, X-ray is done on 50% of the units (that's a very high rate), Some lots have higher failure rates than others, so we don't use them. Sadly, we still have ~4000 of these parts. After all these measures, the part still has ~10% failure rate. I will probably have to create an image that exercises all the IOs.

[wraper]

A customer of mine is using Spartan 2 parts in BGA package. Recent batches have >20% failure rate for the FPGA - IO failures, no JTAG etc.

I bet your customer brings recycled garbage. I would look for signs of relabeling. As these are old parts, it's possible that original marking was made by paint which can be easily removed without any need of sanding. I would look if PCB substrate when looking on the edges is darker than usual or if there is color mismatch between the ICs.

[SiliconWizard]

I'm not even sure how you get ahold of Spartan 2 FPGAs these days unless through dodgy resellers selling old stock and probably scavenged, untested chips. I'm curious.

[asmi]

I'm not even sure how you get ahold of Spartan 2 FPGAs these days unless through dodgy resellers selling old stock and probably scavenged, untested chips. I'm curious.

You'd be surprised to find out that they are still sold on Mouser, and are even in stock right now.

[SiliconWizard]

I'm not even sure how you get ahold of Spartan 2 FPGAs these days unless through dodgy resellers selling old stock and probably scavenged, untested chips. I'm curious.

You'd be surprised to find out that they are still sold on Mouser, and are even in stock right now.

I am indeed, although the available stock is tiny and the price is insane.

[asmi]

I am indeed, although the available stock is tiny and the price is insane.

That is still eons better than the situation with more modern FPGA, which are impossible to buy for any amount of money via official channels. The only place where I was able to find them is Aliexpress

[NorthGuy]

That is still eons better than the situation with more modern FPGA, which are impossible to buy for any amount of money via official channels. The only place where I was able to find them is Aliexpress

Prices are unreasonable anyway (roughly 15x of pre-crisis). Moreover, if you try to buy, there's no guarantee that they have it.

[Gribo]

It's the customer's own stock, they bought it years ago, and kept it sealed in storage and unlike other customers, this one doesn't mess with 2nd tier distributors (medical devices). My question was whether it is a sensible approach to implement a design that exercises the part before assembly - Build a breakout board with a BGA sockets, lots of IO buffers, and hook it to a PXI DIO board.

[ebastler]

My question was whether it is a sensible approach to implement a design that exercises the part before assembly

It's pretty clear that this was your question, but it is apparently not what this group has chosen to discuss... 

More seriously -- the caveat that the reliability problem might be due to bad solder joints, caused by humidity in the parts, was a valid one. If that were the underlying problem, your test fixture would obviously not help in identifying problematic parts. But if you are sure that the parts have been sourced, stored, and baked properly, and can hence rule out soldering problems, I think the test adapter you propose is a plausible approach.