So, it looks, so far, like we don't have any ohmic connections between the VIOs and GNDs on any of the first-silicon chips.
This suggests that something was just screwy in this last silicon run. The only change to the pad ring was the PLL filter settings, which wouldn't have had such effects, and only connects to V2431, anyway.
I've not seen any lateral current measurements yet ? (Shows in the waveforms above, and I measure ~10mA here )
Done with VDD powered and measure the current from V2431 to GND, all other VIOs isolated
Did you simulate to find where that flows thru ?
Terrel at Parallax did a bunch of measurements on the seven remaining glob tops we have.
He found one VIO pin on one device which had a 190-ohm path to GND. See page two of attachment.
He read ~8M-ohms on every pin, except the bad one. These pins were single and not grouped in sets of two, like on the P2 Eval.
We ought to be able to initially screen the bad dies on the wafer prober by measuring resistance between VIO (all grouped) and GND (all grouped) and looking for >100k ohms to pass the die. That would keep the ball rolling, while ON Semi figures out what happened.
P2 Eval Board Owners, I need your help!
**snip**
I would think the negative meter probe would go to GND (any of those four GND posts) and the positive meter probe would go to the Vxxxx pin header pin being measured. You'll need to measure with the VIO pin being subjected to the positive voltage. Otherwise, you may wind up measuring the clamp diode to GND, which we don't care about.
If you could report your measurements as such, it would be most helpful:
We ought to be able to initially screen the bad dies on the wafer prober by measuring resistance between VIO (all grouped) and GND (all grouped) and looking for >100k ohms to pass the die. That would keep the ball rolling, while ON Semi figures out what happened.
Are they confident the (possibly damaged) probes are not damaging die ?
I'm surprised they do not have a 'soft start' approach, that injects some low mA, and checks the pins are above some test level. Effectively a resistance to gnd test.
Curious what yields do they get ? (ignoring the peak drain issue)
Any chance this could be ESD related ? What ESD tests did OnSemi do on ES1 and ES2 ?
It just seems unlikely that a PAD metal part of the die could be erratically variable. Generally the clearances there, are much less tight than the die core, and photo imaging is usually very stable.
Did they have any issues on other companies dies, in the same shuttle run ? (tho they might not admit to that...)
Is anyone else getting "open diode" when measuring those pins with diode multimeter setting? I found that on 2 of the VIO-to-GND pairs. Rest were close to either 1V or 2V.
Edit: Scratch that. Might be the pin fet reacting to measurement voltages.
Yes, first and last.
EDIT: Ha! You're right, testing those all again today and the results are mostly 4+ volts all around.
Those seem all very low ? Is there a missing 'M' ? Does the P2 work ?
There is no missing M
I am using a EEV blog 121 GW multimeter
Does the P2 work ?
I can't say because I haven't been using it yet. I've been too busy lately but I just dusted it off last weekend and had a look at the accessory boards. I am hoping to free up some time this weekend or next to run a few programs.
You are the first one reporting such low resistance values.
Since you've never powered-up your Eval board, perhaps it's advisable not doing so for a while, before giving Chip a chance to express any ideas he can have about it.
Does the P2 work ?
I can't say because I haven't been using it yet. I've been too busy lately but I just dusted it off last weekend and had a look at the accessory boards. I am hoping to free up some time this weekend or next to run a few programs.
Is it easy to plug it in, to quickly check if you get 3.3V and 1.8V and the chip is not hot ?
@jmg
re:Is it easy to plug it in, to quickly check if you get 3.3V and 1.8V and the chip is not hot ?
I plugged it in and the 1.9 VDC and the 3.3VDC voltages are fine. The chip feels cool after about 10 minutes. There is no apparent short.
Good, seems those ohms readings might be bogus then... an average of say 15 ohms, would be 0.726W, which would be warm...
Thats interesting. Are you able to measure the open circuit voltage of that eevblog 121 GW, Bob?
Just put the 121 GW into resistance measuring mode, and measure its open circuit DC voltage using a second voltmeter. I think values that small would indicate its a relatively high voltage that its measuring with
Those seem all very low ? Is there a missing 'M' ? Does the P2 work ?
There is no missing M
I am using a EEV blog 121 GW multimeter
Does the P2 work ?
I can't say because I haven't been using it yet. I've been too busy lately but I just dusted it off last weekend and had a look at the accessory boards. I am hoping to free up some time this weekend or next to run a few programs.
Hi Bob,
I would be cautious about that EEV Blog multimeter. Try to measure with another multimeter. Also, make sure that the 121GW multimeter does not output more than 3.3V when measuring resistance (you can measure the voltage the 121GW outputs with another multimeter, with the probes in the air, and then with the probes measuring on the circuit). I suspect that it is doing so, because that multimeter is powered by four AA cells in series, totaling 6V!
I would be cautious about that EEV Blog multimeter. Try to measure with another multimeter. Also, make sure that the 121GW multimeter does not output more than 3.3V when measuring resistance (you can measure the voltage the 121GW outputs with another multimeter, with the probes in the air, and then with the probes measuring on the circuit). I suspect that it is doing so, because that multimeter is powered by four AA cells in series, totaling 6V!
If you are unsure about the ohms readings, (CAPS can confuse some), you can try a simple 100R~1K series resistor power, and measure the pin voltage in voltage mode. (or drop over resistor)
It looks pretty certain that this VIO short to GND is only in the new silicon.
It's not on every die, though. So, we should be able to screen away the bad ones with a resistance test on the wafer prober. Then, we can still get some initial chips packaged.
.. So, we should be able to screen away the bad ones with a resistance test on the wafer prober. Then, we can still get some initial chips packaged.
Can that probe test report the actual pin(s), and the current or resistance seen, when it tests ?
That would help isolate the offending pins, if some never fail, vs some often fail, you can compare the clearances / metal etc
Were those with the VIO jumpers removed, to isolate the VIO pins ?
No the VIO jumpers were on for the measurements
...
You are measuring the "output resistance" of either the LDOs, or the DC-DC converter, depending on your jumpers configuration. You should remove them, definitely.
It looks pretty certain that this VIO short to GND is only in the new silicon.
It's not on every die, though. So, we should be able to screen away the bad ones with a resistance test on the wafer prober. Then, we can still get some initial chips packaged.
Hi Chip,
It is good to know that not all dies are defective. What would be the yield, by the way?
If it is not on every die does that basically suggest this is only a manufacturing/yield issue as opposed to a design change issue? Or could there still be some possible offending design change that is somehow marginal affecting some chips and not others? Seems very weird that a design change would do that, one might imagine it would more likely to be a more consistent issue over all chips and less random if that were the case. Also is the probe tip burnout and this issue directly related? Did these high currents on VIO do the initial damage to those probe tips? Or, could a partially damaged probe tip have caused these low resistance paths to start to appear on subsequently tested chips, by somehow frying something internally due to arcing etc? Pure speculation and I don't know their test methodologies.
To add what rogloh said - When I was in big corporate defense industry as a design engineer - We always had GS - Golden Samples. These were usually sub-assemblies or final products that were more perfect than perfect: dimensionally, electrically and functionally. Problem in the field or on the production floor, we tested the offending product / component against the Golden Sample to quickly isolate the problem. Was it a manufacturing defect, contamination, assembly, electrical - whatever. We assumed nothing, but many times it was human error, or test fixtures themselves that caused most of the problems.
To add what rogloh said - When I was in big corporate defense industry as a design engineer - We always had GS - Golden Samples. These were usually sub-assemblies or final products that were more perfect than perfect: dimensionally, electrically and functionally. Problem in the field or on the production floor, we tested the offending product / component against the Golden Sample to quickly isolate the problem. Was it a manufacturing defect, contamination, assembly, electrical - whatever. We assumed nothing, but many times it was human error, or test fixtures themselves that caused most of the problems.
It is common practice in the industry to have golden samples. But that is justifiable in bleeding edge processes where the lithography can fail, or samples can be slower than normal. Actually, this often happens in the same waffer, where you can have SS graded samples, and FF samples, and the in-between.
However, I think this is a completely different case, where process just got botched. Mind that this is a 180nm process, and this should never happen. Anyway, correct me if I'm wrong.
Comments
Done with VDD powered and measure the current from V2431 to GND, all other VIOs isolated
Did you simulate to find where that flows thru ?
He found one VIO pin on one device which had a 190-ohm path to GND. See page two of attachment.
He read ~8M-ohms on every pin, except the bad one. These pins were single and not grouped in sets of two, like on the P2 Eval.
We ought to be able to initially screen the bad dies on the wafer prober by measuring resistance between VIO (all grouped) and GND (all grouped) and looking for >100k ohms to pass the die. That would keep the ball rolling, while ON Semi figures out what happened.
V0007 = 4.7M ohms
V0815 = 4.7M ohms
V1623 = 4.7M
V2431 = OPEN ohms
V3239 = 4.6M ohms
V4047 = 4.6M ohms
V4855 = 4.6M ohms
V5663 = OPEN ohms
Can you mount those and check they work ok.
Even the 190 ohms indicates ~ 17mA if that is purely resistive and does not degrade further.
Are they confident the (possibly damaged) probes are not damaging die ?
I'm surprised they do not have a 'soft start' approach, that injects some low mA, and checks the pins are above some test level. Effectively a resistance to gnd test.
Curious what yields do they get ? (ignoring the peak drain issue)
Any chance this could be ESD related ? What ESD tests did OnSemi do on ES1 and ES2 ?
It just seems unlikely that a PAD metal part of the die could be erratically variable. Generally the clearances there, are much less tight than the die core, and photo imaging is usually very stable.
Did they have any issues on other companies dies, in the same shuttle run ? (tho they might not admit to that...)
V0007 = 6 M ohms
V0815 = 9 M ohms
V1623 = 8 M ohms
V2431 = 8 M ohms
V3239 = 14 M ohms
V4047 = 9 M ohms
V4855 = 8 M ohms
V5663 = 8 M ohms
V0815 = 20.66 ohms
V1623 = 7.54ohms
V2431 = 2.37ohms
V3239 = 9.47 ohms
V4047 = 12.21 ohms
V4855 = 11.51 ohms
V5663 = 10.47 ohms
With my meter leads shorted together I get .047 ohms
Yes, first and last.
EDIT: Ha! You're right, testing those all again today and the results are mostly 4+ volts all around.
Those seem all very low ? Is there a missing 'M' ? Does the P2 work ?
There is no missing M
I am using a EEV blog 121 GW multimeter
Does the P2 work ?
I can't say because I haven't been using it yet. I've been too busy lately but I just dusted it off last weekend and had a look at the accessory boards. I am hoping to free up some time this weekend or next to run a few programs.
Looks like you did found something new ....
You are the first one reporting such low resistance values.
Since you've never powered-up your Eval board, perhaps it's advisable not doing so for a while, before giving Chip a chance to express any ideas he can have about it.
Only a thought
Henrique
Is it easy to plug it in, to quickly check if you get 3.3V and 1.8V and the chip is not hot ?
re:Is it easy to plug it in, to quickly check if you get 3.3V and 1.8V and the chip is not hot ?
I plugged it in and the 1.9 VDC and the 3.3VDC voltages are fine. The chip feels cool after about 10 minutes. There is no apparent short.
Good, seems those ohms readings might be bogus then... an average of say 15 ohms, would be 0.726W, which would be warm...
V0815 = 14.0 Mohms
V1623 = 14.0 Mohms
V2431 = 14.0 Mohms
V3239 = 14.0 Mohms
V4047 = 14.0 Mohms
V4855 = 14.1 Mohms
V5663 = 13.3 Mohms
did you try reversing the measurement leads to reverse polarity?
Just put the 121 GW into resistance measuring mode, and measure its open circuit DC voltage using a second voltmeter. I think values that small would indicate its a relatively high voltage that its measuring with
I would be cautious about that EEV Blog multimeter. Try to measure with another multimeter. Also, make sure that the 121GW multimeter does not output more than 3.3V when measuring resistance (you can measure the voltage the 121GW outputs with another multimeter, with the probes in the air, and then with the probes measuring on the circuit). I suspect that it is doing so, because that multimeter is powered by four AA cells in series, totaling 6V!
Kind regards, Samuel Lourenço
Were those with the VIO jumpers removed, to isolate the VIO pins ?
No the VIO jumpers were on for the measurements
@Tubular
re:Are you able to measure the open circuit voltage of that eevblog 121 GW, Bob?
Yes,
.222 volts DC
0007 = 2.624 M
0815 = 2.971 M
1623 = 3.449 M
2431 = OL (Meter shows OL at > 3.999 M)
3239 = 3.331 M
4047 = 2.769 M
4855 = 2.233 M
5663 = 2.593 M
C.W.
V0007 = 6.72 Mohms
V0815 = 6.72 Mohms
V1623 = 6.75 Mohms
V2431 = 6.73 Mohms
V3239 = 6.76 Mohms
V4047 = 6.76 Mohms
V4855 = 6.75 Mohms
V5663 = 5.25 Mohms
Amprobe 37XR-A
It looks pretty certain that this VIO short to GND is only in the new silicon.
It's not on every die, though. So, we should be able to screen away the bad ones with a resistance test on the wafer prober. Then, we can still get some initial chips packaged.
Can that probe test report the actual pin(s), and the current or resistance seen, when it tests ?
That would help isolate the offending pins, if some never fail, vs some often fail, you can compare the clearances / metal etc
Hi Chip,
It is good to know that not all dies are defective. What would be the yield, by the way?
Kind regards, Samuel Lourenço
However, I think this is a completely different case, where process just got botched. Mind that this is a 180nm process, and this should never happen. Anyway, correct me if I'm wrong.
Kind regards, Samuel Lourenço