Intel Xeon E5-2699V4; apparent short on power

[tmbinc]

I bought a - correctly described as - damaged Intel Xeon E5-2699V4 (for way too much money...). It was described as "fell down, capacitor ripped out, doesn't boot". (In retrospect I don't doubt that the CPU doesn't work, but I doubt that it worked before it "fell down" - I think it was probably dead before and someone attempted a repair. Anyway - it was sold as broken and I respect it as that.)

It arrived in good shape except for the ripped out capacitors. I It looked like someone attempted a botched repair (see attached picture), so I had good hopes initially in just fixing up the ripped ground plane. There are no other visible signs of either damage or repair attempts, except for one corner that has a very little dent (which I checked to not affect the isolation between the power planes).

I measured the impedance across a remaining capacitor, and indeed, ~40mOhm (measured with 34401A in 4W mode, but including the impedance of the measurement tips). A reference measurement on a working CPU showed ~4 Ohm (unstable due to capacitor charging and strongly depending on test voltage of course), distinctively different.

I carefully cleaned up the area with a sharp knife, and ensured that there are at least no optically visible connections. But still - low impedance. I looked up the polarity from the socket pinout, and applied current, in the hope of being able to see the fault with a thermal camera. No such luck - the voltage drop across the chip is tiny (for example ~4mV at 1A), and nothing heats up significantly. I tried higher currents - after all, what's to lose - 2A, 10A, even 30A, but the voltage drop only linearly increases, indicating a resistive load somewhere.

Now I'm really confused. At that point I have pretty much given up on repair - but I'm still curious on what defect mechanism this could be. I'm not aware of a failure mode of silicon that results in 40mOhm impedance between power and ground.

I've heard of capacitors that went conductive before, but with such a low impedance? Unfortunately the huge ground plane makes it almost impossible see the effect of a tiny heatup, and the low voltage drop means that even at 30A(!!), I would only see only 3.6W of dissipated power. I removed a few capacitors (around the "impact point") that had visible damage - no change.

There is no visible damage to the ground plane, at least on the bottom.

I mapped out the voltage drop across the bank of capacitors (by measuring VCC and GND on each of the 8x13 center capacitor array), in the hope to see a local minimum pointing me to point of the shortcut, but it doesn't appear to be within the capacitors but rather outside - the voltage drops the further we go, but only a tiny bit (~4.75mV to 4.71mV). The fault must be elsewhere, and it must be freaking low-impedance.

De-lidding on this chip is difficult because the heatsink is soldered (with Indium - melting point of 156°C) to the silicon. I have no experience with de-lidding so it's a last-resort as it will likely destroy - if it isn't already - the CPU, so maybe someone has an idea?

[stojke]

It is normal to have low impedance on microchip power supply lines even in the values of 2 to 8 ohms between ground.

[tmbinc]

stojke - 2 to 8 Ohms (at low voltages) seems okay, and a good CPU is within that range - but I'm measuring ~4..40 milli-Ohms, so it's a real shortcut.

[stojke]

Ive had experience with intel chips shorting to ground on their comm lines but never on their power supply. Id guess its long gone. Some one probably tried to overclock it on a crap board or way too much.

[vaualbus]

why the hell have you spend money on this?? I wish you not spend to much. CPU are not fixable!
Also you could even demage the cpu ourself when measuring the resistence with the multimeter. This CPU work at very low voltage sometime voltages below 1.0V.
The fact that the CPU getting warmer is normal, this kind of cpu can apsorbe when workin something like 80A on the VCCIO voltage line so I guess that if you wanna make the short dissapear you would need a lot of current anyway.
Still buy a working CPU would had been better.

[wraper]

Still buy a working CPU would had been better.

I don't think many would buy working E5-2699V4 with $4k MSRP, even engineering samples cost over $1k.

CPU are not fixable!

They may be fixable if there was mechanical damage. And seller probably sold it by falsely presenting it as such to mislead the buyer. All in all, ebay is pretty crap place to buy broken equipment. Most of the stuff is already attempted to fix and made even worse or knowingly misrepresented. Like you have a dead motherboard, barely worth anything. Bend a few pins on CPU socket and sell it as such, many suckers will buy it hoping for easy fix.

[TiN]

MLCCs often can crack and have a short. Since at this point and that level of substrate butchery which indeed looks like multiple layers deep already, i think you aren't risking much to remove all MLCCs (sorting by location in case you need put them back) and check if the short is there. Also this is Broadwell-EP, so it have FIVR. Unlike usual older CPUs (pre-Haswell), this means that VCC input to CPU package does not actually go into CPU (logic/cores etc) but goes to FIVR mosfets and power stages. You can google up some info about Intel FIVR online to learn more. So if any of those internal (ondie) mosfets fail as short circuit, then you get yourseft a pretty keychain souvenir. 

[stj]

i had a shorted E8400, same familly as xeon afaik.
i was going to remove the related caps but couldnt - god knows what they "solder" them on with.

anyway, now we know they are all compromised, it doesnt matter.
i'm looking for used AMD stuff on ebay/ local recycling. 

[wraper]

i had a shorted E8400, same familly as xeon afaik.
i was going to remove the related caps but couldnt - god knows what they "solder" them on with.]

I suspect you were using crappy soldering iron(s) with tiny tip or weak hot air and no preheating. CPU is a piece of heatsink, and multilayer PCB substrate transfers heat quiet well.

[tmbinc]

Yes, I was using my Ersa i-Tool (which I like a lot), and had very little success - the thermal mass is simply not sufficient to heat up the solder quick enough, and the thermal resistance to the ground and VIN plane is too small to heat it locally. I then switched to the Ersa Power Tool that I had lying around, and it was much better. Also, I tried - per eevblog recommendation - Rose's metal (as a cheap alternative to ChipQuik, which seems really hard to get in Europe) and that worked great as well.

Anyway - I think I can conclude that the MLCCs on the bottom side are not at fault - it they would, I would see the minimum voltage drop across one of them. Instead, the minimum voltage drop I can see are between a pair of VIN/GND pins at the outer package edge (with no visible damage around), so I conclude that it must be something on the upper side, under the IHS. It sure could be a blown FIVR but I would not expect silicon to have such low impedance (~5mV drop @1A -> 5mOhm). I understand there is littlezero chance of fixing this, so I'll put on my "root cause hat" instead.

So seems I don't have a choice than to de-lid the part. My plan was to cut the glue carefully with a razor blade, and then heat up the IHS to ~170°C to melt the indium. Any suggestions? (Other than to not do it, of course...)

[TiN]

And what removing IHS will help you how, exactly?
Blown mosfets can have microohm resistances, no worries .

Cracked MLCCs often presented as dead short, so you would not see any voltage drop over them. If you pump few hundred amps into them, you can see them shine nicely though 

[andy2000]

I've had some experience salvaging CPUs like this.  One trick I use to locate a shorted cap is to monitor the resistance of the short with a micro ohm meter while I touch my soldering iron to each cap.  When you heat the shorted cap, the resistance will increase slightly.

I have seen a number of CPUs where the short seems to be in the silicon itself.  There should be several isolated power rails that show different resistances.  If you're measuring a short across all the caps, then it's probably in the silicon, or you have several shorted caps.  You could simply remove all the caps before you scrap it.

As a last resort, you can try powering the chip in a motherboard to see if the short will burn open.  There's some risk of damaging the motherboard VRM, or socket, but it usually goes into over current shutdown.

[Bicurico]

In order to get a better idea of this processor and, especially, the price tag and so to understand why tmbinc is trying to fix a broken one, I googled for it and found a seller on Amazon. The Q&A are pricesless!!!

Regards,
Vitor

PS: A used CPU like this costs around US$3.000...
 

[Bicurico]

The reviews are a pearl, too.

[Bicurico]

On a serious note, here is an interesting site regarding on how to remove the CPU cover: https://www.ekwb.com/blog/what-is-delidding/

Good luck tmbinc!

[macboy]

I recently delidded an old Core 2 Quad too get a look at the two dies it was made from. I stuck it in a bench vise and literally sheared the lid off. It came off with no drama at all; the indium solder is extremely soft. The CPU had been in working condition but as it was >10 years old I didn't care if I damaged it. I don't recommend this technique on a $4000 CPU.