E-bay AD588BQ references

[BradC]

I've just had 4 E-bay special AD588BQ references arrive : https://www.ebay.com.au/itm/AD588-High-Precision-Voltage-Reference-Module-AD588BQ-without-battery/222382448519

If I read the data sheet correctly, the BQ should give 10V +/- 1mV. So maximum of 10.001V
The measurement sheet that comes with them says :
1 - 10.00282
2 - 10.00127
3 - 10.00156
4 - 10.00169

Measured with an unknown 34401A. So out of the box all 4 refs are out of spec. My rough measurements with an in-cal 3457a are within a couple of hundred uV of these measurements. I need to get them run up for a couple of days to measure them again. At the moment I have them powered with +/- 18VDC from a Tek PS503A, but my eventual aim is to use 3 15-0-15 transformers to float the references, then put them in series for a more stable 30V source, and then divide that down with a caddock divider to get 3V, 300mV & 30mV.

These could be anything, even factory rejects. But out of the bag they exceed the maximum tolerance specified in the data sheet.

Is my initial observation correct or have I missed a trap for young players ?

On the topic of the caddock divider : https://www.digikey.com.au/product-detail/en/1776-C6815/1776-C6815-ND, what would be a reasonable voltage follower to use to ensure the divider remains unloaded? Chopper stabilised op-amp?

[Kleinstein]

Those cheap parts from Chinese or similar sources could be recycled parts. So they could have drifted, just from age or unsoldering. Pre-aged parts could be a good thing too. Its possible to have relabeled relabeled parts too (at best a different initial tolerance class).

For the buffer amplifier an AZ OP may be an option. Keep in mind the bias current and maybe current noise - so the super low noise types like LTC2057, MCP6V91 may not be the best.  If the circuit can adjust for zero offset at reasonable time frame an precision JFET OP like OPA140 would be OK too.

It also depends which part of the divider is used - using the full 10 M for 30 V is quite a lot. Ignoring the 9 M part could be an option to get lower impedance.

[RandallMcRee]

Use an ADA4522 for the buffer. They just work. They are low noise (lower noise than your vref).

It seems to be in spec for an "AQ". What are the markings on the chip?

[Andreas]

Is my initial observation correct or have I missed a trap for young players ?

Besides that it might be a remarked JQ device,
I would also check if the voltage after the voltage regulator is correct.

Are the outputs on the PCB correctly kelvin sensed or is there a large voltage difference betwen output GND and Gnd Sense In+ (pin 9) of the device?

with best regards

Andreas

[Kleinstein]

Use an ADA4522 for the buffer. They just work. They are low noise (lower noise than your vref).

It seems to be in spec for an "AQ". What are the markings on the chip?

The reference noise is somewhere in the maybe 500 nV/sqrt(Hz) at 1 Hz.  With a divider of 1:1000 with would be less than the about 6 nV from the OP.  So a low noise OP could really help. 

As there is quite some current noise the ADA4522 may not be the best choice for the 1 M source impedance. Some extra cap to ground could help, but the current noise could also include an 1/f part. For the 3 V output, a higher impedance OP like the AD8551 could work better.

[RandallMcRee]

It seems to me that the largest error source in this plan is the Caddock voltage divider, +- 30ppm/C.

So I would worry about that more than current noise.

OP, do you have a schematic? What are your requirements?

[BradC]

It seems to be in spec for an "AQ". What are the markings on the chip?

They have 3 line markings. The 1st two are lined up with the AD logo. The 3rd is under the logo. I don't have hardware that allows me to get a usable photo.
1 - AD588BQ / 0828 / E73245
2 - AD588BQ / 0828 / E73245
3 - AD588BQ / 1028 / E73245
4 - AD588BQ / 0828 / E73245

It has been a while since I looked at markings on cerdips. They could well be re-marked.

Are the outputs on the PCB correctly kelvin sensed or is there a large voltage difference betwen output GND and Gnd Sense In+ (pin 9) of the device?

The outputs are not correctly kelvin sensed. They are joined under the chip. There is no discernible difference between output GND and pin 9. They are almost co-located on the board and are both joined to the ground plane. I do get 150uV(ish) difference between pin 6 & pin 1 (unbuffered output vs buffered output - measured both in reference to gnd and between the pins)

I thought that may have been the additional 10M load applied by the meter, but I chucked a 1M load on there and got no change, so the output buffer appears to have a 150uV(ish) offset.

I'm also seeing 110uV between GND and pins 8/10 which if I read the schematic right would imply the internal ground is not at system ground, but measuring between pin 1 and pin 8/10 gives me the same reading as pin1/gnd, so maybe I'm reading the chip schematic incorrectly.

It seems to me that the largest error source in this plan is the Caddock voltage divider, +- 30ppm/C.

So I would worry about that more than current noise.

OP, do you have a schematic? What are your requirements?

Good question. I have no schematic as I'm still spitballing. Getting a stable 30V is my first step. I figured I'd buy 4 references and pick the best 3.

I already have the divider, so the 3, .3 & .03 outputs are next. I'm also considering tapping the references to allow me to switch using 10, 20 & 30V as a base for a bit more flexibility. I've been using a Tek PG506 but while it is flexible enough, it's also a bit drifty.

My requirements are to provide a short to medium term stable set of voltages 30mV to 300V (I'll deal with the 300V later, but at the moment I'm looking towards 30 REF5010 in series as a shunt regulator) as a transfer standard to periodically calibrate (not adjust) a pair of HP3478a against a 3457a. I'm chasing a short term (stable temperature ~10 to 30min) of a couple of PPM (<10 if I can manage it). If I can get <100PPM over 24 hours with a couple of degrees C ambient change then I'm doing better than I need.

Stability rather than accuracy is the aim here.

[TiN]

set of voltages 30mV to 300V (I'll deal with the 300V later, but at the moment I'm looking towards 30 REF5010 in series as a shunt regulator) as a transfer standard to periodically calibrate (not adjust) a pair of HP3458a against a 3457a.

I'm sure you messed something up in this sentence, as 3458A would not like *3 ratio voltages anyway. And 3458A being capable of ppm transfers, testing it against 3457, noise of which is much larger than 3458A 24-hour spec. Even if you just borrow DIY LTZ-reference from some EEVBlog member and calibrate 3458A to that 7.xx volt DC, even with shipping and conditions effect, you'd end up in way better uncertainty than anything 3457A. 

[BradC]

set of voltages 30mV to 300V (I'll deal with the 300V later, but at the moment I'm looking towards 30 REF5010 in series as a shunt regulator) as a transfer standard to periodically calibrate (not adjust) a pair of HP3458a against a 3457a.

I'm sure you messed something up in this sentence, as 3458A would not like *3 ratio voltages anyway. And 3458A being capable of ppm transfers, testing it against 3457, noise of which is much larger than 3458A 24-hour spec. Even if you just borrow DIY LTZ-reference from some EEVBlog member and calibrate 3458A to that 7.xx volt DC, even with shipping and conditions effect, you'd end up in way better uncertainty than anything 3457A. 

I absolutely got that wrong. HP3478a. If I had a 3458a I wouldn't be farting around with 10's of PPM.

[BradC]

I sent an E-bay contact note to the seller asking if they were flogging re-marked or counterfeit chips as even by their measurement all 4 references they sold were out of tolerance (their add parrots the datasheet about +/- 1mV initial accuracy).

The came back and told me I hadn't let them warm up enough, must have been using the wrong input voltage and +/- 1mV is 2mV in total and therefore since the references were lower than 10.002V they were all in spec. I've countered pointing out their own measurements showed them all high, and +/-1mV means between 9.990 & 10.001. Let's see what they come back with.

I've decided I'm not using these refs for the 30V source anyway (I now have 3 REF5010 to stack), but since they were advertising the specs I want to see what they intend to do about 4 references that don't meet the advertised spec. I'm not going to play hardball, and I can find a use for these as-is, but if you don't ask you never get.

[Kleinstein]

I would not worry so much about the absolute value being off a little more than specs. With the absolute value measured reasonable well, there is no real need to have the value spot on at 10 V.

I like the argument that +-1 V means a 2 mV range and thus 2 mV off is OK  .  Repeat that a few times and everything is OK. It is odd that they still specify +- 1 V when they have measured the actual voltage that is more like 2 mV away from the nominal values.

Have you checked the voltage directly at the chip (to avoid possible errors from the layout). So it could be +-1 mV from the chip and +-1 mV from a poor board.

[BradC]

I would not worry so much about the absolute value being off a little more than specs. With the absolute value measured reasonable well, there is no real need to have the value spot on at 10 V.

I'm not really fussed or worried, but if you sell a device with a specification it should really meet it as a minimum. Am I being unreasonable?

I like the argument that +-1 V means a 2 mV range and thus 2 mV off is OK  .  Repeat that a few times and everything is OK. It is odd that they still specify +- 1 V when they have measured the actual voltage that is more like 2 mV away from the nominal values.

And this is my point. Add says (A), and they deliver (B).

Have you checked the voltage directly at the chip (to avoid possible errors from the layout). So it could be +-1 mV from the chip and +-1 mV from a poor board.

Yes I have. There is a combination of what appears to be an offset in the chips buffers, but even without that offset they are > 1mV.

I just thought I'd ask the question and see what sort of answer I got. I absolutely expected an accusation of me "doing it wrong", but then I also might have been pleasantly surprised.  (I wasn't).

[Muxr]

I have one of those I got last year. Mine seems to be within the spec. Although different markings.

[Andreas]

Although different markings.

Hmm looks like black topping on a ceramic case.
What do you see if you scratch the black topping off?

AD usually lasers the ceramics directly.

with best regards

Andreas

[Muxr]

Don't think it can be scratched off without grinding it. The top part is pretty thick:



That's the official image from the product listing on DigiKey. https://www.digikey.com/product-detail/en/analog-devices-inc/AD588BQ/AD588BQ-ND/617694

[BradC]

Hmm looks like black topping on a ceramic case.
What do you see if you scratch the black topping off?

I attacked one of mine with a razor blade and the black topping came straight off. The engraving went through the black topping into the ceramic though and I was unable to scrape that off.

What if they were ground/polished to remove the old markings and leave a smooth ceramic, black-topped and then engraved through to the ceramic? The engraving in the bare ceramic isn't particularly deep.

edit : Most certainly black-topped.
I gave one of the others a quick scrub with a cotton bud and some isopropyl alcohol and the black-top wiped clean off. On this one you can clearly see the radial grinding marks in the ceramic and the engraving is a lot shallower. I'll see if I can get a usable image.

edit2: I hit all 4 chips. 2 wiped clean off, 2 didn't, so I don't really know now. Image of chip with grinding marks attached.

[Andreas]

edit2: I hit all 4 chips. 2 wiped clean off, 2 didn't, so I don't really know now. Image of chip with grinding marks attached.

So obviously re-marked parts: in best case they made a lower grade part to the BQ grade part.

Don't think it can be scratched off without grinding it. The top part is pretty thick:

I asked to remove the paint only (black topping) not the ceramics.
Black topping is never used by semiconductor manufacturers.
If its easyly done with acetone (or even isopropyl alcohol) it is in every case a re-marking.

with best regards

Andreas

[Muxr]

I asked to remove the paint only (black topping) not the ceramics.
Black topping is never used by semiconductor manufacturers.
If its easyly done with acetone (or even isopropyl alcohol) it is in every case a re-marking.

Ahh I see. Sorry I misunderstood.

[BradC]

I sent an E-bay contact note to the seller asking if they were flogging re-marked or counterfeit chips as even by their measurement all 4 references they sold were out of tolerance (their add parrots the datasheet about +/- 1mV initial accuracy).

Got a response. "No they are most certainly genuine, my supplier who I trust with my life tells me they are genuine and nobody has ever said anything before."
Sent polite response with description and image of chip with grinding marks and got "oh dear, I must question whether I continue to work with that supplier, you can return them or I'll give you a $10 refund". I just politely let it go. Of course they know they are selling dodgy stuff. Still, had to ask the question.