LM723 die pictures

[Noopy]

No, it looks the same. No way in hell that this matches the schematic from current National/TI datasheet.

Well perhaps "very different" was a little too hard. The basic schematic could be the same and the blocks are similar arranged but it´s clearly a different ... well ... my english isn´t the best  ... it´s a little bit different.

I´ll have to check the whole circuit as soon as I have more time...

edit
Not sure if it means that there are two versions of National LM723 or maybe you just bought this part from AliExpress

This one came from ebay... 

[magic]

Here is a 723 schematics from electronicprojectsforfun page, with the current sense base resistor (and even more mess):

That's the modern National schematic. And yes, there is a base resistor on the limiter. But it also shows VREF output being connected to the collector of the noninverrting input of the regulation amplifier and that's simply not the case on this die. On the die, regulator amplifier is wired exactly as per the Motorola/TI schematic.

On the other hand, the biasing circuit looks interesting. There are two NPNs connected in a weird way near the JFET, and indeed, it looks like something similar might exist on the die. So perhaps it is some older version of genuine National LM723.

edit
Indeed, the transistors near the JFET look the same, assuming that this wide, snaking thingy is the JFET
I can't see anything resembling Q25, though. And the whole reference section is definitely different. For starters, observe that the National schematic shows several PNPs in the reference circuitry, and here there are no PNPs except for the current sources at the top.

[chris_leyson]

PS: I like the 723 (because I built my first PSU around it). On the other hand the stories about its "Vref stability and low noise" are just urban myths, imho.

Ulrich Rohde published a very low noise regulator circuit using an LM723 and a PNP pass transistor in "Mircrowave and Wireless Synthesizers: Theory and Design". Noise performance turned out to be much better than any of the ultra low noise regulators from National and TI and the 723 ended up as part of a low phase noise crystal oscillator. Circuit below.

[iMo]

PS: I like the 723 (because I built my first PSU around it). On the other hand the stories about its "Vref stability and low noise" are just urban myths, imho.

Ulrich Rohde published a very low noise regulator circuit using an LM723 and a PNP pass transistor in "Mircrowave and Wireless Synthesizers: Theory and Design".

You quoted me: Well, experienced voltnuts here (equipped with suitable TE) may build such a source with a "new" 723 off Digikey or Mouser and publish some noise/stability data.
PS: to demonstrate my positive attitude to the 723 - here are 4x "low noise" floating sources (2x15V, 2x5V) I made last fall - with 2x723(15V/60mA/20mA_foldback)+PNP, 723's Vrefs filtered 4k7/220u.

[chris_leyson]

I used a pair of AC coupled amplifiers either AD797 or LT1028 and I think the overall gain was 40dB to get the noise up to a level I could measure with an HP3585. Used a TI uA723C without an external pass transistor. It would interesting to test a few "new" 723s given the design variations.

[David Hess]

Nice hand drawn PCB artwork...

For the record, my CAD drawn printed circuit boards look like that also.  Circuit board layouts are about the only place where I made good use of geometry class.

No buried zener in 723. And frankly, I doubt there is even a zener as the Vref.
The Vref is quite noisy, I saw 2mV p-p noise there, moreover, Wolfgang confirmed that in his measurements.
PS: I like the 723 (because I built my first PSU around it). On the other hand the stories about its "Vref stability and low noise" are just urban myths, imho.

The ones I have tested compared very favorably with bandgap references when the 723 was operated as a reference instead of a regulator.  Bandgaps have to be multiplied up or the 723 divided for comparison purposes.

The real advantage should be using the output transistor to operate the 723 at a constant temperature.

[magic]

These days LM723H seems to cost about as much as LM399 so this oven trick is getting pointless, unless there really is a serious advantage in noise even before filtering.

BTW, I'm not sure why people bother biasing the noninverting input with resistors in those designs. I would leave it open circuit so that the inverting input is pulled firmly to ground by the tail current source and can be used for temperature sensing. It could even be combined into a darlington with the current sense transistor for more gain.

In fact, this technique seems applicable to just about any opamp, so a much more interesting question is if it would work on AD588. The latter is not cheap either, but at least has a fair bit of added functionality which may justify the effort.

[iMo]

FYI - self heating 723 Vref
https://www.eevblog.com/forum/projects/playing-with-723-voltage-regulator-oldtimer/msg1779983/#msg1779983

[David Hess]

These days LM723H seems to cost about as much as LM399 so this oven trick is getting pointless, unless there really is a serious advantage in noise even before filtering.

Why would I need the expensive metal can package?  I am referring to using the 723 output transistor as the heater and the current limit transistor as the temperature sensor.  One example is shown below.

BTW, I'm not sure why people bother biasing the noninverting input with resistors in those designs. I would leave it open circuit so that the inverting input is pulled firmly to ground by the tail current source and can be used for temperature sensing. It could even be combined into a darlington with the current sense transistor for more gain.

In fact, this technique seems applicable to just about any opamp, so a much more interesting question is if it would work on AD588. The latter is not cheap either, but at least has a fair bit of added functionality which may justify the effort.

If the current mirror which controls the tail current saturates, then all of the connected mirror outputs shut down.  The positive and negative current mirrors used for biasing in the 723 are in a positive feedback loop so both may be affected.

Doing this also removes the Vbe compensation from the error amplifier.

[magic]

If the current mirror which controls the tail current saturates, then all of the connected mirror outputs shut down.  The positive and negative current mirrors used for biasing in the 723 are in a positive feedback loop so both may be affected.

Haven't thought of it

That being said, the lower current mirror of µA723 has no other outputs so no problem here. And IMO the fault will not propagate to the upper mirror, because the saturated mirror's input will not go high impedance.

In this particular mirror, I think the "driver" transistor Q9 will take on generation of the missing half of Q13 emitter current and feed it there directly through Q13 base, restoring normal operating point of Q10 and permitting sinking of all Q7 output.

I concede that this doesn't generalize to arbitrary opamps as easily as I hoped.

Doing this also removes the Vbe compensation from the error amplifier.

Sure, the idea is to use the input stage of an opamp as a temperature sensor / regulator

[David Hess]

If the current mirror which controls the tail current saturates, then all of the connected mirror outputs shut down.  The positive and negative current mirrors used for biasing in the 723 are in a positive feedback loop so both may be affected.

Haven't thought of it

That being said, the lower current mirror of µA723 has no other outputs so no problem here. And IMO the fault will not propagate to the upper mirror, because the saturated mirror's input will not go high impedance.

The schematic I checked showed one other output ... to drive the positive current mirrors in parallel with the startup circuit through Q25 thereby controlling the available current to the input of the bottom current mirror, hence the positive feedback.

In this particular mirror, I think the "driver" transistor Q9 will take on generation of the missing half of Q13 emitter current and feed it there directly through Q13 base, restoring normal operating point of Q10 and permitting sinking of all Q7 output.

As mentioned previously, that schematic is wrong in some details so I am not sure if what you would suggest would work or not.  Unlike the example I gave, I would probably not be as clever and instead use an external error amplifier for the thermal control for more gain.

I concede that this doesn't generalize to arbitrary opamps as easily as I hoped.

Some operational amplifiers do weird things if the input common mode range is exceeded or the output is allowed to saturate.  Duals and quads where the bias circuitry is shared can be especially problematic with the condition of one amplifier affecting others.  More modern parts tend to be designed to at least behave benignly.

Doing this also removes the Vbe compensation from the error amplifier.

Sure, the idea is to use the input stage of an opamp as a temperature sensor / regulator [/quote]

It will also pick up the saturated temperature coefficient of the current mirror output.  Offhand I do not remember if that would add to or subtract from the Vbe temperature coefficient.

[magic]

Oh, Q25, forgot about that one. It's unique to National and not even present in the one National chip we have seen so far. But I think it's not a problem, I think the mirror will work all right thanks to Q9. A subtle issue may occur in chips following the original design, because diverting current from Q11 to Q9 will reduce standing current in one of the VREF output transistors, likely affecting VREF load regulation.

I assume that saturation tempco is low enough to be ignored. For what it's worth, ZTX689B datasheet shows about +25mV going from 25°C to 100°C at 1A. That's 15% loss of gain. Almost no difference at low currents and almost no difference at -55°C - not sure what to think of that, but who cares.

You are perhaps right that I'm trying to be too clever, but I like the minimalism in elimination of those external resistors on IN+

[Noopy]

I got a LM723J:

https://richis-lab.de/LM723_02.htm

Wolfgang said the LM723CJ is the consumer version and the LM723J shows less drift.

Actually the die looks very different:



It even has testpads!

Perhaps it´s only one of the newer (complexer) designs or the LM723J is really different to the LM723CJ...
I think I need one or two more of this LM723... 

[magic]

This might be the new National circuit. And look at that green snake meandering over isolation diffusions; I was wondering how they are going connect Q10 with Q25 on the opposite side of the die and there you go

Now, where's the reference diode and what are those combs on the left side? Schematic suggests capacitors?

Those test pads seem to be Zener zaps. They are connected to reverse biased BE junctions and resistors in parallel with them. Is this thing really calibrated in production?

[Noopy]

This might be the new National circuit. And look at that green snake meandering over isolation diffusions; I was wondering how they are going connect Q10 with Q25 on the opposite side of the die and there you go

That resistor is really freaky! 

Now, where's the reference diode and what are those combs on the left side? Schematic suggests capacitors?

Those test pads seem to be Zener zaps. They are connected to reverse biased BE junctions and resistors in parallel with them. Is this thing really calibrated in production?

Have to do some more research. Right now it´s a bit to late...

[magic]

I'm starting to wonder if the whole reference section could actually be a bandgap cell

For starters, there is no DC connection between D2 and anything else on the schematic, only some AC coupling. This wouldn't be the first time a datasheet schematic is wrong, but I'm not sure what the missing connection could be and where to find it on the die.

On the other hand, Q21, Q22, R14, R15 could be a bandgap reference. Note that Q22 consists of 10 identical copies of Q21.
Q19, Q20 current mirror attempts to establish equal current through them, Q19 collector drives the emitter followers Q6, D5, Q4, Q5 which regulate VREF and indirectly control Q21, Q22 base voltage. This is how bandgaps work AFAIK.

Note that the VREF node is driven by Q5 emitter and I don't really see other feedback path to Q5 base besides the above.

It remains a mystery what Q23, Q24 are doing. Some thermal compensation?
And I glossed over R8.

edit
It has to be a bloody bandgap. Zener fanboys are running for cover

[Noopy]

Interesting!

I have to do some more research regarding the LM712. But at the moment you know much more about the LM723 than me...  :popcorn:

[iMo]

No buried zener in 723. And frankly, I doubt there is even a zener as the Vref.

[David Hess]

I'm starting to wonder if the whole reference section could actually be a bandgap cell

I have been told that some versions used a bandgap instead of a zener reference.

For starters, there is no DC connection between D2 and anything else on the schematic, only some AC coupling. This wouldn't be the first time a datasheet schematic is wrong, but I'm not sure what the missing connection could be and where to find it on the die.

I have noticed that before many times now.  I think the schematic is wrong.

[magic]

The schematic looks mostly right so far. VREF is driven by Q5, the NPN right next to the bias PNPs. Q5 base is driven by Q4 emitter, Q4 has D5 embedded on its collector (there seems to be a tiny fractional collector shorted to the base by a patch of metal) and that is driven by Q6, whose base is a high impedance node between Q21 and Q19 collectors. Long story short, VREF is two diode drops above the voltage between Q19 and Q21. And this voltage is determined solely by the ratio of Q21 and Q22 currents due to the PNP mirror above them. That's a bandgap cell, and it seems to control the output exclusively.

I assume that Q4 and Q6 are substrate PNPs and the yellow-brown stuff around their emitters is just bulk epitaxial layer silicon with no buried layer underneath (buried layer seems to show up as red-brown). That's consistent with grounded-collector PNPs I have seen elsewhere. But surface P diffusions (NPN bases) are visually almost identical to bulk silicon here, so maybe I'm wrong and it is actually a P diffusion which acts as the GND collector and it covers almost the whole island and makes contact with P isolation diffusions surrounding it to rid itself of the incoming current

I don't quite understand the nature of the three objects hanging off the left collector of the split PNP at the top. One of them ought to be D2. Perhaps D2 is a patch of N+ over the large grounded P collector of Q4, assuming that Q4 has a surface collector. Then there is that green strip (green is N+ elsewhere) which leaves the isolation island and apparently connects to GND metal. It could be that it activates when Q4 collector / D2 anode rises 0.7V above ground due to P diffusion's resistance and shunts the current to ground. No idea

By they way, if you wonder what's the point of D2 if the chip uses a bandgap reference, it might serve to reduce voltage across the compensation capacitors, which are presumably of "reverse-biased BE junction" type, and prevent breakdown. But I'm not sure, those combs look totally weird.

[BravoV]

I'm starting to wonder if the whole reference section could actually be a bandgap cell

I have been told that some versions used a bandgap instead of a zener reference.

Which version exactly ?

[David Hess]

I'm starting to wonder if the whole reference section could actually be a bandgap cell

I have been told that some versions used a bandgap instead of a zener reference.

Which version exactly ?

Foreign knock offs maybe?  The comment was made in connection with why some 723s are much noisier than others.  I never found a schematic for one with a bandgap reference or at least one that I recognized as such.

[Wolfgang]

So a few remarks for Wolfgang, if you are reading this

It wasn't exactly the legendary Widlar and, pending further research, the National schematic may not even be true to the original Fairchild design.

And I also doubt it's a buried zener. Looks like any other NPN BE junction visually and the Fairchild appnote (page 1.12) brags about a state of the art process capable of integrating even N-JFETs and capacitors in addition to NPNs and PNPs

Hi,

have you seen Noopy's LM723J die photos ? They look substantially different from the consumer stuff.
https://www.richis-lab.de/LM723.htm
The J/833 parts are a lot more stable than the CN versions, I still have no info how they have been made.

[magic]

Hi, we discussed this new die a few days ago. It appears to match the schematic from National datasheet and I think it's a bandgap, crazy as it sounds.

Look at transistors Q21 and Q22, the latter made of ten paralleled Q21 emitters, and the resistors between them. If I understand correctly, that's exactly what a bandgap reference is supposed to look like. Read my post from February 14th, where I explained why I think that VREF output is determined solely by a need to keep Q21 and Q22 currents equal (maybe up to a constant, not sure what the influence of R8 is). This is achieved by driving their bases to some appropriate level and no higher.

And I just calculated from the values of the resistive divider that with VREF=7.15V their base voltage is about 1.23V, go figure.*

We have also seen a different National part (or so Noopy's eBay vendor said ), which used a similar bias generator to this one but otherwise was identical with all the mainstream xx723s. Perhaps an older National version, maybe a fake, or maybe they really put different dice in the J and CJ versions

You can verify which version you have by testing resistance from VREF to GND. It should show open circuit on the classic design chips and about 15kΩ on this National weirdo.


*There is a bit of influence by Q23 and Q24 collector current. This part of the circuit is completely above my head, but I think it makes little difference because emitter resistors are relatively high and base voltages are only 600~650mV above ground. Maybe I should do the math instead of guessing but I'm lazy

[Wolfgang]

There is one pragmatic point that speaks against a bandgap version. The 7.something volts is exactly what you would choose for a Zener with minimum tempco. For a bandgap, you can make any voltage, and with a lower voltage you would be able to run the chip at lower supply voltages too, widening its market (723 minimum 10V is not ideal for many cases). So I still doubt the did a bandgap.

723 clones with improved data (SG3532,...) did use bandgaps, but they also offered a much lower minimum supply voltage.

The remedy would be to consult the inventors, bit I'm afraid all them have already moved over to the eternal soldering grounds.