Keithley 196 losing calibration values

[Krampmeier]

Hi all,

I'd like to share last weekend's story...
I have a Keithley 196 6.5 digit Multimeter which suddenly stopped working in the 300 mVDC range. It showed 96 mV when the input terminals were shorted and overflow when any significant voltage was applied to the inputs. The other DC ranges and the Ohms ranges seemed to be fine.

I suspected a fault in the range switching or input amplifier, so I opened the thing, removed the shielding plate, and powered it up again. That was a mistake, because the PCBA had moved a bit on the metal poles in the middle, which then had shorted something on the board out. Learned the hard way when and where to use a strip of Kapton tape...

The supply voltages were now all gone, +15 rail down to 2 V and alike. Transformer buzzing. Not easy to debug from there, especially as the mains fuse blew after a short while, too.

Long story short, the LM399 reference heater was fried, and that's all what was broken. I fortunately still had another scavenged LM399 reference in the empire of dirt, and the meter was back to the initial state, with the 300 mV range not working (and the others now way out of cal).

Luckily TiN has all the required documents on his site, including hi-res schematic scans, and the manual includes a description of the low-level test modes the Keithley 196 has built in. Thank you so much for that, TiN!

I spent many hours going through the schematics, understanding only little, and probing the analog board. Everything seemed good and working until I got to the input of the ADC section. The input voltage drops to 0 there, but I suppose that's how it is supposed to work, as it is the case for all ranges.

Stopped probing, connected my (way out of cal) Adret 103 A and a Keithley 2010, and did the front panel calibration for the 300 mV range. Works like a charm again now 
Same for the AC ranges which were completely gone as well. Adjusted the lower ones with a frequency generator (and K2010), and they were bang on again.

It seems the X2816CP EEPROM does not have great data retention time, so the DMM just lost its calibration values. Has anyone here ever seen that happen?

That particular EEPROM is only available from eBay, and those chips are probably no younger than the one I have. Can I still trust any of these when I re-calibrate the unit? Is there any suitable drop-in replacement?

I'll re-cap the meter soon, then take it to my employer's cal lab, and hope for the best...

[alm]

Based on your story, I am not exactly sure when the EEPROM lost its calibration values. Did it give a checksum failure at some point? Did it happen when you (presumably) shorted something killing the LM399? Any what killed the '399? Apart from the mains input or an external voltage (both likely to kill a lot more than just the '399), there should be no voltage inside that unit that would blow the heater. Either the +/- 15 V rails would have to get reversed, or one of them got shorted to mains. Depending what was happening to the +5 V digital rail, I could imagine the EEPROM getting unhappy.

Erasing the entire EEPROM and recalibrating it should recharge the cells, so that might help it retain information longer. Assuming the EEPROM is not defective. I would probably do a basic calibration and then leave it for a while to see if it retains information, otherwise that calibration is a waste.

[saturnin]

It seems the X2816CP EEPROM does not have great data retention time, so the DMM just lost its calibration values. Has anyone here ever seen that happen?

I think it is rather unlikely EEPROM lost its cal data (some X2816CP datasheets claim data retention is greater than 100 years). Moreover, K196 tests ROM, RAM and EEPROM upon power up. If EEPROM test fails, "UNCAL" message is displayed...

Still, there can be something wrong with the input section (particularly, you reported more ranges were affected). I would check readings on 300mV DC range are stable in time (i.e. they don't drift)...

[The Soulman]

Mind that lm399 don't have good initial accuracy and can differ up to 2% on 6,95V
So replacing one with another could potentially create a 4% error on your dmm.

[Krampmeier]

The first thing that happened was the 300 mV range failing, without me opening the housing. The unit had just been powered off for a couple of weeks. I also did not apply any excessive voltage.

I did not investigate further on the short circuit which killed the reference.  I am sure mains voltage was not involved. It could be that a pin of the the heater supply (+/- 15 V) and the zener were connected for a moment. I suppose that this could have killed the circuit, but I didn't try with another one to make sure. Only one left in the "special parts" bin...

I did not get the "uncal" message, or any other error code. We don't know if there is a checksum test in this old meter though - it could as well be that it just checks for some magic number or prints that message if all the cells read FF...

I traced the signal through the input section, and it looked OK as far as I can tell. I connected 100 mV and 200 mV as test voltages, and the corresponding voltage was present at the resistor which feeds it right into the ADC. It was only possible to do that in test mode though, as the meter cycles through several signals (0V, reference, input) all the time during normal operation.
I did the same test in both the 300 mV and 3 V ranges, and got consistent results. The input amp toggled between x1 and x10 just fine, and that is the only difference between these ranges. The readings on the display were fine in the 3 V range, but total garbage in the 300 mV range. In my understanding, this indicates that the problem seems to be not on the signal path side.

So far, the 300 mV range also does not seem to have drifted a lot yet (within 3 days). It is still pretty consistent with the K2010 which I used as a reference during alignment. It was totally off before.

I agree the Xicor data sheet claims 100 years data retention. I don't know if maybe ageing electrolytic caps could have compromised that somehow, but I can't think of anything else...

@The Soulman
Thanks for the reminder! I know that the new reference will make it necessary to re-align all ranges, and it will also have increased drift for a couple of years because of the soldering process. That is OK with me, as long as the meter works again.

[The Soulman]

@The Soulman
Thanks for the reminder! I know that the new reference will make it necessary to re-align all ranges, and it will also have increased drift for a couple of years because of the soldering process. That is OK with me, as long as the meter works again.

Ok, I misunderstood and was under the impression you already replaced the LM399 and after that all other ranges
were out of spec.

Long story short, the LM399 reference heater was fried, and that's all what was broken. I fortunately still had another scavenged LM399 reference in the empire of dirt, and the meter was back to the initial state, with the 300 mV range not working (and the others now way out of cal).

[Bushougoma]

I have seen two (one in my presence) Xicor X2444 16 Byte no that's not a misprint (EEPROM backed SRAMs) lose the calibration constants.

The first meter a Keithley 175 I purchased listed as not working it was actually working because it was displaying CeRR. Which usually means the calibration constants are corrupted. I would say it was bit rot because when I put the X2444 into my TL866 and programmed it with random data it was retained and verified properly. Not wanting to take any chances I purchased some NOS replacements installed a socket, replaced it, and sent it to the cal lab. I made sure to backup the constants after it came back.

The second Keithley meter I purchased a 197 didn't have a calibration issue but I purchased it with the intention to have it calibrated. Since I was going to back up the constants after it came back from the cal lab (due to my previous experience with the 175 which used the same X2444 chip). I desoldered it with my Hakko FR300 (so the absolute minimum of heat was applied) to install a socket.

For no reason other than to compare the old constants to the new ones when I received it back from the cal lab I put it in my TL866 to read it and the calibration constants were gone. Not corrupted just gone I believe the bytes were CFFF over and over. To be sure it wasn't just the programmer not reading it properly I reinstalled it into the meter and sure enough CeRR was displayed. For the heck of it I was also able to program it with random data and it verified properly.

I again replaced the component as a matter of course luckily I bought a few extra when I was working on the 175.

All of the work was done on an ESD mat so I doubt that was the cuprit. The first Xicor in the Keithley 175 I can understand the hex "looked right" and it was likely simple bit rot a single bit flipped and the constants were useless. But the second one on the 197 was just too uniform CFFF over and over that one has me dumbfounded . If heat damaged the chip which was somehow causing that repetitive data to appear I wouldn't have been able to write to it and verify the data.

In any case the meters have been in service for over two years and no issues yet but even if there are at least this time I have the calibration constants to restore them.

So the short answer is yes it does happen .