More voltage references - die pictures

[Noopy]

Hi all!


I don´t want to blow the metrology section apart, so I will post less important references in this topic.


Today I have a LT1009 for you:












I have also done some reverse engineering.

It seems that they can change the reference voltage by changing the metal layer. Most parts of the 6k6-resistors are shorted here.

In the upper right corner there are some parts that are not connected. Perhaps a spare function? A start-up-circuit? It seems that this circuit has also a fusible link.

And I found a small hidden capacitor looking like an F. At least I think it´s a capacitor.

See more at my website:

https://www.richis-lab.de/REF10.htm

 

[Noopy]

Today I have a LT1029 for you:

https://richis-lab.de/REF11.htm




The specifications are quite simliar to the LT1009 but the die is much more complex. It seems that it´s based on the LT1019 design because you can read 1019B on the die.




The die could be kelvin connected. The two bondpads placed at the ground potential and the two placed at the v+ potential are isolated from eacht other so you can compensate the bondwire resistance even without kelvin connection outside the package. That seems more intelligent than in the LT1236 where the kelvin connection pads are connected on the die.




A nice bandgap configuration!


You can compensate the temperature drift with the fusible resistors in the bottom left corner that belong to the bandgap circuit. Fusing the resistors in the upper right courner you can adjust the output voltage because they are placed in the feedback path.


 

[Andreas]

Hmm,

the LT1019 has a heater pin and a heater resistor (hidden in the newer datasheets).

https://www.datasheetarchive.com/pdf/download.php?id=4628e6e0c37324eca3aefd62a37ec0b715ab84&type=P&term=LT1019-5

the problem is that there is no extra ground pin and the heater current shifts the output voltage due to common impedance effects.

But if its the same die the heater structure should be somewhere visible

With best regards

Andreas

[Noopy]

Where did you find a heater in the datasheet? 

Well I think I will have to decap a LT1019... 

Best Regards,

Richard

[Andreas]

In the above linked datasheet (old version Rev B) page 2 + page 3 (see PIN 7)

see also AN42 page 15
https://www.analog.com/media/en/technical-documentation/application-notes/an42.pdf

with best regards

Andreas

[Noopy]

Sorry, I was somehow blind. I need more coffee... 
You are absolutely right.
Perhaps the heater is not on the same die? Although that would not be very reasonable...

I have to decap a LT1019! 

[Noopy]

I have some new die pictures for you.
I have taken a closer look at a AD588 (revision A).
That´s a quite interesting reference. Two outputs, bipolar, bipolar floating, Kelvin Connection... 






The AD588 uses two metal layers as the AD587 does (https://www.richis-lab.de/REF06.htm).




It´s quite easy to identify the big blocks (reference, opamps A1, A2, A3, A4).




Analog used an interesting resistor to tune the reference (I assume the temperature coefficient). They used a "normal" laser trimmed resistor and connected it in series with a "big, splitted resistor" in which they cutted whole resistors.




The AD588 uses the "Analog-buried-Zener" you can also find in the AD587 (https://www.richis-lab.de/REF06.htm) and in in the AD565 (https://www.richis-lab.de/DAC06.htm).




You can find a buried zener in every opamp! 


More pictures here:

https://www.richis-lab.de/REF12.htm

 

[magic]

That looks similar to the 587. The opamps use µA741 input stages again. The zeners are probably for local bias generation in each opamp - they don't seem to have any connection to the primary reference cell to derive their bias from it.

[Noopy]

That looks similar to the 587. The opamps use µA741 input stages again. The zeners are probably for local bias generation in each opamp - they don't seem to have any connection to the primary reference cell to derive their bias from it.

 

Nevertheless it´s interesting they used buried zeners for the bias generation. 

[magic]

Analog used an interesting resistor to tune the reference (I assume the temperature coefficient). They used a "normal" laser trimmed resistor and connected it in series with a "big, splitted resistor" in which they cutted whole resistors.

Yes, that's tempco trim. The reference circuit is about the same as AD587 and this is the R1/R2 divider. Q1-Q3 are the big NPNs near the zener and they used a quad PNP Wilson mirror this time for Q1 base current cancellation (visible above). Ground connections from the reference cell seem to use Kelvin routing to the output stage of A2. I wonder how thermal gradients from A2 affect performance, given that those three NPNs are supposed to compensate each other in various ways and that the input stage of A1 also is laid out along the gradient. Possibly not a good idea to load Vlow.

[Noopy]

...

It´s always a pleassure to read your explanations. 

[Noopy]

I decapped some more AD588 (three).
Someone bought a batch AD588 over eBay. He said some didn´t work as they should. The first AD588 was a good one, the following AD588 is one of the buggy parts.
They all (the good and the bad) had the same package marking engraved in the metal lid.




What´s that? A newer revision. It is called C588. The C pretty sure stands vor revision C. It was designed 1987 (revision A was designed 1985).
They changed a lot on the die it´s not just a die shrink or some minor modifications. Nevertheless they didn´t change the lid marking.
Furthermore it seems that in 2005 they still used an old revision (A, 1985)! Perhaps they had a big stock pile of the old revision. Perhaps they had problems with the newer revision and had to go back to the old revision but still had a big stock pile of the new revision. 
Strange things happen...




- They moved the zener and the transistors away from the output stage of A2 and arranged the transistors around the zener.

[...]
I wonder how thermal gradients from A2 affect performance, given that those three NPNs are supposed to compensate each other in various ways and that the input stage of A1 also is laid out along the gradient. Possibly not a good idea to load Vlow.
[...]

 

- They integrated an additional buried zener. It seems that the zener is used to supply some bias generator.
- They moved the capacitor in the bottom left corner of revision A upwards and integrated one more capacitor.
- There are unused transistors in the upper left corner. These transistors didn´t exist in revision A. I assume there was a revision B in which these transistors had been integrated in the reference block.




The opamp A2 is quite similar to the revision A but there is some more circuit at GND-SENSE+. It seems that this circuit adjusts the bias in the differential input stage in relation to the common mode voltage at the input.




The opamps A3 and A4 have some additional parts at the output stage. It seems like they integrated some protection circuits here.


More pictures here:
https://www.richis-lab.de/REF13.htm
Overview over all decapped references:
https://www.richis-lab.de/REF00.htm

 

[Andreas]

What´s that? A newer revision. It is called C588. The C pretty sure stands vor revision C. It was designed 1987 (revision A was designed 1985).
They changed a lot on the die it´s not just a die shrink or some minor modifications. Nevertheless they didn´t change the lid marking.
Furthermore it seems that in 2005 they still used an old revision (A, 1985)!

Hello,

from the picture above the lid looks sanded. (no original manufacturer would do this).
So I guess that those EBAY parts are pulled and remarked parts.
I would not make much guesses about the manufacturing in this case.

In 1992 Databook the side brazed DIP (D) package is still listed.
https://archive.org/details/bitsavers_analogDevilogDevicesDataConverterReferenceManualVo_129751587/page/n1251/mode/2up

In 1996 Databook only the CERDIP (Q) package is still available.
https://archive.org/details/bitsavers_analogDevilogDevicesDesignersReferenceManual_112939486/page/n687/mode/2up

So for me a side brazed dip in 2005 smells like counterfeit.

With best regards

Andreas

[Noopy]

Hello Andreas,

probably you are right. That sounds more plausible too.
I could not believe that this metal lid got a new marking. The writing looked so nice...

Best regards,

Richard

[magic]

Just wanted to say the same

And I think you swapped GND SENSE + and -. Your order doesn't match the pinout.

The new circuit in A2 is input bias cancellation, I think. It makes sense, because the inverting input of this amplifier can see high impedance in the ±5V configuration, which is not matched by an equal impedance on the noninverting input. Its bias current could potentially disturb the 50:50 divider too, maybe.

It seems that the whole current of the input stage and the current sink which biases the input stage (4x the current of one input transistor) goes through an NPN identical to the input ones and the base current of that NPN is mirrored at 25% ratio into a common base PNP (not sure what that one is doing) and from there goes to the input pin.

This offers simple means of determining the die revision nondestructively.

You also missed another tweak, which is found in the 1987 AD587 too: the connections between the input stage NPNs and PNPs are crossed over. This presumably offers some thermal gradient cancellation - if there is a vertical gradient, the NPN of one branch and the PNP of the other branch get hotter, and offset voltage of both branches shifts about equally.

[Andreas]

The writing looked so nice...

Hello Richard,

from the first look: yes
So if there were no different chip revisions in the same date code I would not have been sceptical.

In worst case you can remove the markings with acetone.

https://www.eevblog.com/forum/metrology/ad587lq-from-aliexpress/msg2023066/#msg2023066

with best regards

Andreas

[mawyatt]

You also missed another tweak, which is found in the 1987 AD587 too: the connections between the input stage NPNs and PNPs are crossed over. This presumably offers some thermal gradient cancellation - if there is a vertical gradient, the NPN of one branch and the PNP of the other branch get hotter, and offset voltage of both branches shifts about equally.

This is a cross-coupled quad type arrangement, common with precision differential stages. Originally developed by George Erdi long ago at Fairchild and utilized in the 1st precision op-amp the ua725 I believe.

It not only helps with thermal gradients, but also process gradients.

Best,

[Noopy]

Thanks for all your input! I will add that to my website soon. 

The writing looked so nice...

Hello Richard,

from the first look: yes
So if there were no different chip revisions in the same date code I would not have been sceptical.

In worst case you can remove the markings with acetone.

https://www.eevblog.com/forum/metrology/ad587lq-from-aliexpress/msg2023066/#msg2023066

with best regards

Andreas

That was the point that puzzled me: That´s no cheap print on the AD588 lid there are nice deep laser drillings.
If the original marking was made with the same technique you have to sand a lot to get rid of it. 

[magic]

This is a cross-coupled quad type arrangement, common with precision differential stages. Originally developed by George Erdi long ago at Fairchild and utilized in the 1st precision op-amp the ua725 I believe.

It not only helps with thermal gradients, but also process gradients.

Yes indeed. Kinda, sorta.
The precision opamp input stages use four identical transistors. This one is a mix of NPNs and PNPs.
I'm not sure if it ends up cancelling much process gradients, given that they likely affect the two types differently.
The primary goal here is stability, not ultimate precision, or they wouldn't be using the 741 topology.

[Noopy]

Today I have a MAX6325 for you:







The die is HUGE: 3,74mm x 2,19mm in a SO8-package! 

Unfortunatelly the picture quality is not as good as I would have wished it to be. (Is that really english or more german?  )




The design was built by Maxim in 1997.




I assume that´s the +/-15mA-Push-Pull-output-stage.
It´s quite big for +/-15mA but it has to be good enough for a voltage reference.
It seems that the lowsider has it´s one ground pad. Would be reasonable.




Maxim left "a lot" of room between the components.
Unfortunatelly I didn´t find the buried zener. The structures are to small... 




Some testpads...
Not very important but interesting: The connections to the capacitors at the bottom are quite different. Hey that are capacitors the connections are quite irrelevant. 


https://www.richis-lab.de/REF14.htm


 

[razvan784]

W...T...F... that looks more like a standard cell ASIC than a precision voltage reference...
Digital trim / temperature compensation? Many features look analog though...
One can wonder how all these complications effect long-term drift...

[Noopy]

Quite strange...

But I don´t see any digital circuit. I assume Maxim integrated a lot of analog circuits to compensate whatever...

If you want to check my pictures in detail:

https://www.richis-lab.de/temp/01.jpg
https://www.richis-lab.de/temp/02.jpg

[T3sl4co1l]

Y'don't suppose G-T, E-E, O-G are the designers..?

Sure seems like there's enough circuitry in there to measure temperature and compensate for it, perhaps digitally.  But even then, did they just... make 10 of everything and wire it in parallel, just because..?!

Tim

[Noopy]

Y'don't suppose G-T, E-E, O-G are the designers..?

Aha! Sounds very reasonable! 

[magic]

A vast majority of this die looks like low density analog to me
I think they may have a complementary process here given the additional diffusions surrounding some transistors, presumably PNPs. A wild guess as to what is what:


As can be deduced, I'm not exactly sure what WAT and WTF might be. Perhaps larger BJTs, or maybe MOS? Who knows, MAX6250 datasheet says it's fabbed on a CMOS process, maybe this one is BiCMOS. But MAX6350 datasheet also says 435 transistors and I have no idea where they are.

I agree with Noopy that the left side is probably the output stage. This means the bondig pads are, top-bottom and left-right:
output stage ground, precision ground, noise reduce, input voltage, do not connect
trim, output voltage, probably output sense, dnc, dnc

It seems to make sense. Ground, input and output are quite fat. Noise reduce appears to cross under a Vin trace and go to some circuitry in the top right corner. Trim is another thin trace which goes somewhere to the right.

edit
For instance, the pair of structures below the P in NPN appears to be a simple current mirror, with the top tranny being the input (B-C shorted). The one I labeled PNP also seems B-C shorted, but an identical transistor on the right is not. I'm pretty sure that the "zeners" are indeed zeners because they look similar to other zeners we have seen and what else could those things be with apparently only two terminals?