Decapping the LT1236LS8

[iMo]

Any idea what function could those pins A,B,C represent when I blow up the 3 fuses (in yellow circles)??

perhaps different output voltages?
The LT1236 is available in 5V and 10V.
The LT1021 is available in 5V, 7V and 10V

It could be. The 1021-7 has none resistor divider there, so I guess they flash a fuse with the resistor's string off (and therefore it has got the best long term stability). I have here a bag of 1021s, knowing more on the A,B,C related fuses I may try to "reconfigure" it..

[magic]

There are two parallel resistors above the Vin pad and below the triple lateral PNP which appears to be the zener current source.

Those fuses short their bottom ends to ground. If you blow the fuses, the resistors will be connected to ground through additional resistance, increasing their effective values up to perhaps a little under 2x of what they are now.

The other ends go somewhere into the amplifier, as Noopy says. I'm not yet sure what they do.

Somewhere there has to be a gain option to implement the 10V variant, unless it uses a different die. It doesn't appear to be any of the fuses connected to the unbonded pads in the center.

[notfaded1]

This isn't at home is it?  If it is wow... I can imagine taking the pictures but you took the lid off with some big gear there.

[Noopy]

perhaps different output voltages?
The LT1236 is available in 5V and 10V.
The LT1021 is available in 5V, 7V and 10V

It could be. The 1021-7 has none resistor divider there, so I guess they flash a fuse with the resistor's string off (and therefore it has got the best long term stability). I have here a bag of 1021s, knowing more on the A,B,C related fuses I may try to "reconfigure" it..

There are two parallel resistors above the Vin pad and below the triple lateral PNP which appears to be the zener current source.

Those fuses short their bottom ends to ground. If you blow the fuses, the resistors will be connected to ground through additional resistance, increasing their effective values up to perhaps a little under 2x of what they are now.

The other ends go somewhere into the amplifier, as Noopy says. I'm not yet sure what they do.

Somewhere there has to be a gain option to implement the 10V variant, unless it uses a different die. It doesn't appear to be any of the fuses connected to the unbonded pads in the center.

Sounds very reasonable. 
…I need more parts… 

...You guys are probably right, but I must warn not to trust Zeptobars blindly unless you know where he bought the chips from or who sent them in. Now I also recall that more recently he was "surprised" to find NE5532 inside OP275. I suppose if they used some less familiar die instead, he would have fallen for it.

I´m also very cautious with logical consequences when I look at dies.
Another example (and off-topic  ): Everybody calls the smaller TI-NE555-die a fake
https://www.richis-lab.de/555_10.htm
Zeptobars has found a new TI-design with big nice structures and callls that one the original design of today. To the contrary the small cheaper design without TI-symbol must be a fake.
I´m not sure about that. You find the small "fake-design" only in TI NE555 marked packages and they did some updates on the design. Why did they put it only in TI NE555 marked packages?
Who knows? Perhaps TI has still an own NE555-design for military or whatever and buys additional very cheap NE555-dies from china. Sound more convincing to me...

[magic]

It could be. The 1021-7 has none resistor divider there, so I guess they flash a fuse with the resistor's string off (and therefore it has got the best long term stability). I have here a bag of 1021s, knowing more on the A,B,C related fuses I may try to "reconfigure" it..

There are some problems with this theory.

Part of R1 located near the LT 1205 logo can't be bypassed by fuses.

The datasheet shows the zener as 6.3V rather than 7V. That's different than the LT1021-7 spec. Also, either 6.3V or 7V zener ought to have significant positive tempco. LT1021-7 might be using a different metalization mask which hooks up that dummy NPN in series with the zener for compensation.

The datasheet also shows the reference output being a weighted average of the zener and two series diodes, presumably cancelling out opposite tempcos. Now, where the diodes should be we see a PNP emitter instead and some pretty complex circuit driving this transistor, but chances are that this thing still has negative tempco and you can't easily change the division ratio without affecting thermal drift of the overall reference. Sure enough, the datasheet confirms that LT1021-5 has drifty TRIM pin voltage. Interestingly, the 10V version doesn't. Another different metal?

The two resistors trimmable by externally accessible fuses degenerate some NPN current mirror. I still don't know how it works. There is like a dozen transistors in this central part but everything seems entangled with everything

[iMo]

The LT1021/LT1236 and LT1236LS8 have got different schematics, btw.
From datasheets:

[notfaded1]

Similar... but NOT the same.

[iMo]

The LT1021/31 and LT1236 are the same schematics as they may support 10V/7V/5V.
The LT1236LS8 is 5V only, therefore the divider at the zener side.
FYI - below the LT1027LS8 (5V) schematics.

[Noopy]

It could be. The 1021-7 has none resistor divider there, so I guess they flash a fuse with the resistor's string off (and therefore it has got the best long term stability). I have here a bag of 1021s, knowing more on the A,B,C related fuses I may try to "reconfigure" it..

There are some problems with this theory.

Part of R1 located near the LT 1205 logo can't be bypassed by fuses.

The datasheet shows the zener as 6.3V rather than 7V. That's different than the LT1021-7 spec. Also, either 6.3V or 7V zener ought to have significant positive tempco. LT1021-7 might be using a different metalization mask which hooks up that dummy NPN in series with the zener for compensation.

The datasheet also shows the reference output being a weighted average of the zener and two series diodes, presumably cancelling out opposite tempcos. Now, where the diodes should be we see a PNP emitter instead and some pretty complex circuit driving this transistor, but chances are that this thing still has negative tempco and you can't easily change the division ratio without affecting thermal drift of the overall reference. Sure enough, the datasheet confirms that LT1021-5 has drifty TRIM pin voltage. Interestingly, the 10V version doesn't. Another different metal?

The two resistors trimmable by externally accessible fuses degenerate some NPN current mirror. I still don't know how it works. There is like a dozen transistors in this central part but everything seems entangled with everything

In my opinion it´s possible that the change in the amplifier leads to an offset somewhere in the amplifier stage which then gives a different output voltage.
That would lead to a constant reference voltage and only modify the amplifier behavior.
Of course you will have to change the amplifier before you tune the reference voltage but that´s no Problem.

The LT1021/LT1236 and LT1236LS8 have got different schematics, btw.
From datasheets:

Perhaps one was an old design and eventually they decided to put the new design in both packages, leaving the datasheet as it was.
In "new" PMI REF01 you don´t get a REF01-design but a ADR01-design because it´s the new reference from analog:
https://www.richis-lab.de/REF01.htm

[magic]

I overdramatized a little, the input stage of the buffer amplifier isn't actually that hard to isolate. Here's the animal with all its inputs/outputs annotated for convenience. I used Zeptobars LT1021-5 image because of higher resolution; looks similar enough.

REF - input from the reference divider. I agree with Noopy that R1~R4 are wired exactly as per the datasheet and that the trim pads can only adjust R1.
FB - feedback from the output pin.
OUT - this drives the output stage.
BIAS - goes to some NPN collector, presumably a constant sink.
RE1/RE2 - mirror emitter degeneration, goes to a pair of resistors with external trim.


The topology is similar to LM101. What's on the right must be NPNs because it wouldn't make sense to connect the inputs to PNP collectors. They work as emitter followers, collectors powered from FB at the very bottom. The lower transistor has some tiny spot connected to REF on its base, probably a clamping diode. Notably, their emitter areas aren't equal.

Next up, split collector PNPs. One collector of each is pulled down by BIAS, another goes to an NPN current mirror. The bases are also connected to BIAS, but one is level-shifted down by a series resistor on the BIAS connection.

I don't understand the purpose of those asymmetries, but I can imagine that they introduce some minor offset voltage and/or a controlled amount of thermal drift.

Trimming the externally accessible resistors RE1/RE2 will imbalance the mirror, forcing a few mV of input offset voltage to develop and change the proportion of currents going through each PNP to make the mirror happy again.

This isn't going to turn a 5V reference into 7V or 10V. These versions must use different metal layers, or more.

[Noopy]

...
I don't understand the purpose of those asymmetries, but I can imagine that they introduce some minor offset voltage and/or a controlled amount of thermal drift.

Trimming the externally accessible resistors RE1/RE2 will imbalance the mirror, forcing a few mV of input offset voltage to develop and change the proportion of currents going through each PNP to make the mirror happy again.

This isn't going to turn a 5V reference into 7V or 10V. These versions must use different metal layers, or more.

Very interesting! Thanks a lot! 

Sure you can´t get 7V or 10V?
A quick guess is that if you trim the mirror to a 2:1 mirror or something like that you will get a quite different amplification factor at the output?

[magic]

Sure you can´t get 7V or 10V?

Yes, I think so. The voltage between PNP bases is determined by voltage drop across these two paralleled resistors on the left side of the lower PNP. This drop depends only on the externally generated bias current (and base currents of the PNPs, but come on).

Then the voltage at each PNP emitter is one diode drop higher and the voltage at the corresponding NPN base is another diode drop higher.

You can change the exact value of "one diode drop" by increasing or decreasing the bias current of the transistor, but that's only some millivolts of difference.

Due to global feedback action, the amount of current which flows from the mirror to OUT has to be constant. It is exactly the amount of current which causes the output to end up at roughly 5V such that the input pair is roughly balanced and the mirror receives the proportion of currents that it expects.

edit
I also see no means of dividing down the feedback voltage fed into the buffer amplifier. It is hardwired for unity gain.
I see no means of adjusting the reference divider, other than the R1 fuses you have already identified.

[Noopy]

Sure you can´t get 7V or 10V?

Yes, I think so. The voltage between PNP bases is determined by voltage drop across those two paralleled resistors on the left side of the lower PNP. This drop depends only on the externally generated bias current.

Then the voltage at each PNP emitter is one diode drop higher and the voltage at the corresponding NPN base is another diode drop higher.

You can change the exact value of "one diode drop" by increasing or decreasing the bias current of the transistor, but that's only some millivolts of difference.

Due to global feedback action, the amount of current which flows from the mirror to OUT has to be constant. It is exactly the amount of current which causes the output to end up at roughly 5V such that the input pair is roughly balanced and the mirror receives the proportion of currents that it expects.

Hm...
Perhaps I misunderstood your interpretation. Let the circuit here aside.

Let´s assume you have a "normal" opamp with a current mirror above the collectors of the input Transistors and a not differential output. That configuration gives an output of X.
Now I put different resistors in the emitter of the mirror so that the mirror gets a factor 1:5.
With this setup the voltages at the inputs and the currents flowing into the collectors are the same. But the mirror will now multiply the current of the one side and source it on the other side with a factor of 5. As the current through the collectors won´t change a five times bigger current will leave the output.
And so I can give the output a factor of 5 by switching the emitter resistors.

Were did I miss the track? Or is your understanding of the circuit a different?

Hope my english isn´t too bad… 

[magic]

That's correct, but you forgot about feedback.

The 5 times larger current leaving the input stage is multiplied by the second stage to a 500 or 50000 times larger current. This goes into the common emitter "current to voltage" transistor at the end of the second stage, which decides the output voltage. The output immediately starts moving towards ground pretty fast and causes an offset voltage to appear between the inverting and noninverting input. A few millivolts of offset is enough to make the input stage currents become 5:1. The mirror is happy, its output current drops down to the original value and the output of the opamp stops moving. In the end, you have moved the output voltage by a few millivolts.

edit
Okay, for 5:1 maybe it would be closer to a few tens of millivolts. But you are still just trimming the offset voltage of the opamp. It can't go too far.

Check out the TL071 datasheet. It uses exactly this method to implement external offset nulling. The exact offset is equal to the difference between input JFET currents divided by input JFET transconductance. Or maybe half or twice that value, whatever, do the math

This is standard opamp theory, nothing specific to TL071 or LT1236.

[Noopy]

 

You are right. 

My working day was too long... 

[Andreas]

The LT1021/LT1236 and LT1236LS8 have got different schematics, btw.
From datasheets:

I think not different schematics.
Only different simplifications of the whole schematic.
With a 6.2 V Zener you need a output voltage divider to get the 10V of the LT1021-10 and LT1236-10 devices.
This is what is shown in the LT1021 / LT1236 datasheets. (the 10V version).

the 5V devices need a voltage divider to divide down the 6.2V for the 5V output.
Since the LS8-devices are only 5V devices they show also only the voltage divider on the zener side.

And this schematic also explains why the trim cirquit for the 5V devices needs a additional diode.

with best regards

Andreas

[magic]

I think I need to seek therapy from my reverse engineering addiction. I just wanted to have a look at that current source for the zener and before I realized here's what I ended up with



The current source is indeed interesting. Q1 represents the small collectors in the corners of Q2~Q5 which are shorted to the base by a strip of metal. There is no other connection to the base so what the heck? Well, the base is the same chunk of silicon as the collector of Q6, that tiny little wart on the big Q3~Q5 collector. It draws current from the base on behalf of the bias generator below, shielding the generator from variations in power supply voltage. Nice.

Q2 is a feedback output, its current being 1/3 of the zener current. It really serves no other purpose. BE junctions of Q6~Q9 ensure that its collector voltage is about equal to Q3~Q5 collector voltage, so it tracks the zener current very well indeed. Clever.

Q7 drives the mirror so as to ensure that Q2 current is equal to the sum of Q11,Q14 currents. The latter are driven by a somewhat complex system of current mirrors, whose input is R6 - the reference divider chain (R1~R3 on LT's schematic) tied to the zener voltage on the other end.

Q12,Q20 are bias sinks for the buffer amplifier. Q21 sinks some small current from the PNP mirror and provides means of startup for the whole circuit.

Speaking of which, there is absolutely nothing wrong with the PNP mirror being cut off and all the other nodes in the bias generator being at ground potential. So what makes the circuit start? Well, without input stage bias, the buffer amplifier can't drive the output stage sink transistor. The output starts to rise and the clamping circuit that I found in the input stage pulls the reference node up, sending current through Q10,R8,Q21 and powering up the reference. Hopefully all of that happens before VCC reaches 5V so the user never sees an output overshoot. That's a hell of a hack

[Noopy]

I think I need to seek therapy from my reverse engineering addiction. I just wanted to have a look at that current source for the zener and before I realized here's what I ended up with
…

 

Thanks for the analysis! 
But we still don´t know what the "ground-fuses" are for, do we?

[magic]

They are offset trim of the buffer amplifier. Perhaps to salvage units which drifted out of spec during the package molding process.

Alternatively, AFAIK, changing current density of a transistor (amps per die area) changes its Vbe tempco. I think
So unbalancing the mirror could be a way to adjust tempco of the buffer amplifier. It also introduces offset voltage, which would need to be compensated by changing the reference divider. This could perhaps save a badly drifting die during factory calibration. But there is no point connecting those fuses to external pins in such case.

The input stage of the buffer amp is certainly designed in a weird way. The upper PNP is maintained at slightly lower base voltage with respect to ground than the lower PNP. The upper NPN has smaller emitter area / higher current density. It's possible that these effects cancel out and base voltages of the NPNs are almost the same. OTOH, one may expect that thermal drifts of the two NPNs operating at different current densities will be different, like in bandgap references.

That's all I know. Maybe somebody can say more
 

edit
If you own such chip you could blow the fuses and see what happens
I think it would be safe to apply 3.3V to those pins (after the grounding fuse has blown) so a resistor-limited low voltage PSU might suffice...

[Noopy]

They are offset trim of the buffer amplifier. Perhaps to salvage units which drifted out of spec during the package molding process.

Alternatively, AFAIK, changing current density of a transistor (amps per die area) changes its Vbe tempco. I think
So unbalancing the mirror could be a way to adjust tempco of the buffer amplifier. It also introduces offset voltage, which would need to be compensated by changing the reference divider. This could perhaps save a badly drifting die during factory calibration. But there is no point connecting those fuses to external pins in such case.

The input stage of the buffer amp is certainly designed in a weird way. The upper PNP is maintained at slightly lower base voltage with respect to ground than the lower PNP. The upper NPN has smaller emitter area / higher current density. It's possible that these effects cancel out and base voltages of the NPNs are almost the same. OTOH, one may expect that thermal drifts of the two NPNs operating at different current densities will be different, like in bandgap references.

That's all I know. Maybe somebody can say more
 

edit
If you own such chip you could blow the fuses and see what happens
I think it would be safe to apply 3.3V to those pins (after the grounding fuse has blown) so a resistor-limited low voltage PSU might suffice...

Very interesting… 

I will have to ask branadic whether his LT1236 was working in the beginning and whether I´m allowed to torture it. 

[magic]

Somebody more familiar with thermal behavior of BJTs could perhaps do the math on that and simply tell you whether tweaking this mirror will make more difference for offset or for thermal drift.

Truth be told, playing with those things is probably pointless if you don't have a good DMM and a thermal chamber. Maybe you will see some difference, maybe you won't even notice, maybe the difference will be due to change in ambient temperature

The chip is probably calibrated optimally from the factory. But hey, maybe it would be useful if you really must have that 5,001V reference without external trimpots.

edit
Two rules of thumb I once calculated from Ebers-Moll for 70°C junction temperature:

+2mV Vbe gives +6% Ic
+20mV Vbe doubles Ic

Look at the ratios of those resistors on the die and you will have some clue by how much the offset can be adjusted and how large the steps are. Note that there are two transistors at each input (NPN and PNP) so the offset doubles. My bet: more than 5mV step with the shortest available segment after one long segment has already been inserted on each side.

It will be slightly different at 25°C, which proves that monkeying with offset affects thermal stability. How much? See Ebers-Moll and calculate. I don't know if that's the most accurate model out there or if it's good enough, but at least you will have some idea of the order of magnitude. Or see what SPICE thinks about it.

[iMo]

There are actually 4 fuses you may break via pins:

pin8 - fuse - pin7 - fuse - pin4_GND
pin1 - fuse - pin3 - fuse - pin4_GND

[Noopy]

Truth be told, playing with those things is probably pointless if you don't have a good DMM and a thermal chamber. Maybe you will see some difference, maybe you won't even notice, maybe the difference will be due to change in ambient temperature

With my old Fluke 45 (and no thermal chamber) it would only be possible to see if the behavior is completely different.
I still haven´t set up GPIB-control... 

There are actually 4 fuses you may break via pins:

pin8 - fuse - pin7 - fuse - pin4_GND
pin1 - fuse - pin3 - fuse - pin4_GND

 

[notfaded1]

With my old Fluke 45 (and no thermal chamber) it would only be possible to see if the behavior is completely different.
I still haven´t set up GPIB-control... 

Get with the program Noopy!  I just did it with a raspberry pi3 with some help from TiN's guide pointing me in the right direction of where to look for stuff at least... I had to improvise some but not too hard to figure out.  I'm kiddin' I put it off for quite a while too.  Funny thing... I ended up testing visa was working with the same device TiN used in his guide.  It just happened to be the device on top near me on the bench...


root@raspberrypi:/home/linux-gpib/linux-gpib-code/linux-gpib-user/language/python# python ./testvisa.py
HEWLETT-PACKARD,33120A,0,8.0-5.0-1.0

root@raspberrypi:/home/linux-gpib/linux-gpib-code/linux-gpib-user/language/python#

Bill

[Noopy]

On my To-Do-list!