Author Topic: LM723 die pictures  (Read 8673 times)

0 Members and 1 Guest are viewing this topic.

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #50 on: February 17, 2020, 08:47:36 am »
But they weren't drop-in replacements. National LM723 is a drop-in replacement for µA723 and could readily be sold to any µA723 user desiring freedom from zener walkout or whatever ;)

And actually, there are people who complained about just that on TI support forum a few years ago.


Perhaps I'm wrong, perhaps that bandgap circuit is somehow sneakily used to compensate a zener or maybe it isn't even a bandgap at all. But we have the datasheet and we have the die image and yet no concrete alternative theory of operation has been proposed so far, despite a few people stating they don't believe that this chip uses a bandgap reference ;)

Noopy, do you have a full resolution pic of the area between the compensation pad and those two lateral PNPs next to it? Something like the closeup of the test pads that's on your website.
« Last Edit: February 17, 2020, 09:13:55 am by magic »
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 2774
  • Country: li
Re: LM723 die pictures
« Reply #51 on: February 17, 2020, 09:08:48 am »
I think a zener is more process steps in manufacturing and less yield. The bandgap is more compatible with the processes they had. When you look at the dies of the Vrefs with zeners you may see the zeners are always pretty visible structures there.
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #52 on: February 17, 2020, 09:24:10 am »
Two of Noopy's dice appear to use an NPN base emitter junction as a zener. That's no rocket science.

You are talking about high-spec chips using subsurface zeners. I linked a source which claims that this technology was introduced for the first time in LM199 in mid-1970s, while µA723 was introduced in late 60s.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 2774
  • Country: li
Re: LM723 die pictures
« Reply #53 on: February 17, 2020, 09:41:00 am »
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.

15xMAA723 DIL14 resistance between pin 6 (+Vref) and 7 (Gnd)

15x 110-115kOhm (34401A kOhm)
14x 1.85-1.89MOhm 1x0.85MOhm (34401A Mohm).
« Last Edit: February 17, 2020, 10:45:19 am by imo »
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #54 on: February 17, 2020, 10:19:47 am »
 :palm: :palm: :palm:

Right, there is a parasitic PN junction from the 100Ω resistor on VREF to VCC. But that's still considerably more than 15kΩ, good enough :)

The 190kΩ outlier may be an autoranging artifact. You will likely get a different reading if you force the megaohm range.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 2774
  • Country: li
Re: LM723 die pictures
« Reply #55 on: February 17, 2020, 10:37:04 am »
Fixed above.
 

Offline Wolfgang

  • Super Contributor
  • ***
  • Posts: 1591
  • Country: de
  • Its great if it finally works !
    • Electronic Projects for Fun
Re: LM723 die pictures
« Reply #56 on: February 17, 2020, 11:18:57 am »
But they weren't drop-in replacements. National LM723 is a drop-in replacement for µA723 and could readily be sold to any µA723 user desiring freedom from zener walkout or whatever ;)

And actually, there are people who complained about just that on TI support forum a few years ago.


Perhaps I'm wrong, perhaps that bandgap circuit is somehow sneakily used to compensate a zener or maybe it isn't even a bandgap at all. But we have the datasheet and we have the die image and yet no concrete alternative theory of operation has been proposed so far, despite a few people stating they don't believe that this chip uses a bandgap reference ;)

Noopy, do you have a full resolution pic of the area between the compensation pad and those two lateral PNPs next to it? Something like the closeup of the test pads that's on your website.

Complaints about drift were common a few years ago, thats true. BUT, it only applied to the consumer version, and it was attributed to some surface charge topic, because you could "heat them out" using elevated temperatures. My guess: New, fast process, cheaper to make, but not as good as the old one (or only after some months of service). Even the higher drift is within specs. TI recommended the MIL version as an alternative.
 

Online Noopy

  • Frequent Contributor
  • **
  • Posts: 858
  • Country: de
    • Richis-Lab
Re: LM723 die pictures
« Reply #57 on: February 17, 2020, 04:40:15 pm »
Coincidentally (and with thanks to floobydust) I have a MIL723 for you:

https://richis-lab.de/LM723_03.htm

 8) :popcorn:







But I don´t know who built this one...

 
The following users thanked this post: floobydust

Offline Wolfgang

  • Super Contributor
  • ***
  • Posts: 1591
  • Country: de
  • Its great if it finally works !
    • Electronic Projects for Fun
Re: LM723 die pictures
« Reply #58 on: February 17, 2020, 04:55:01 pm »
... this one looks like a fake part to me. The military never made parts with a C temperature class.
In TO-99 the MIL parts were suffixed -HM or HJ, IIRC.
« Last Edit: February 17, 2020, 07:58:10 pm by Wolfgang »
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #59 on: February 17, 2020, 07:20:34 pm »
No parasitic diode from VREF to VCC :-+

Somebody saved himself the headache of routing a few ground traces and used the substrate :)

H suffix indicates the package, C the temperature class. There are LM723CH still available at Mouser. But MIL-C, yeah, what's that :-//
« Last Edit: February 17, 2020, 07:26:45 pm by magic »
 

Offline floobydust

  • Super Contributor
  • ***
  • Posts: 4508
  • Country: ca
Re: LM723 die pictures
« Reply #60 on: February 17, 2020, 08:55:20 pm »
Thanks to Noopy for popping the hood on that MIL723 which has been in my parts bin for decades.
I don't remember where I got it from. Gold leads and bonding wires, it's not a fake by modern standards. But the marking is plain and no logo. If I did buy it, must have been from Poly-Paks around 1977 and they were notorious for selling blems and rejects. Have to compare die fonts to see who made it.
I checked ads and could not find it. Did see ads for the LM300 sold as a "super 723".

Signetics NE550 almost identical to LM723 but a lower reference voltage 1.6V despite a zener being there?
I wonder how IC designs were licensed back there. The shenanigans with the reference must be over patent issues, or to get a little better numbers off the datasheet for sales.
« Last Edit: February 17, 2020, 09:04:24 pm by floobydust »
 

Offline Wolfgang

  • Super Contributor
  • ***
  • Posts: 1591
  • Country: de
  • Its great if it finally works !
    • Electronic Projects for Fun
Re: LM723 die pictures
« Reply #61 on: February 17, 2020, 09:19:48 pm »
What speaks for a fake is the unlabeled chip and the MIL/consumer temp range combo.
Maybe east block ? Or some spec-missing surplus ? But then the chips would at least have a label.
 

Online Noopy

  • Frequent Contributor
  • **
  • Posts: 858
  • Country: de
    • Richis-Lab
Re: LM723 die pictures
« Reply #62 on: February 17, 2020, 09:23:09 pm »
What speaks for a fake is the unlabeled chip and the MIL/consumer temp range combo.
Maybe east block ? Or some spec-missing surplus ? But then the chips would at least have a label.

I have already opened a MAA723 (Tesla). That one looks different.
Coming soon...
 
The following users thanked this post: Wolfgang

Offline Wolfgang

  • Super Contributor
  • ***
  • Posts: 1591
  • Country: de
  • Its great if it finally works !
    • Electronic Projects for Fun
Re: LM723 die pictures
« Reply #63 on: February 17, 2020, 09:35:10 pm »
Thanks to Noopy for popping the hood on that MIL723 which has been in my parts bin for decades.
I don't remember where I got it from. Gold leads and bonding wires, it's not a fake by modern standards. But the marking is plain and no logo. If I did buy it, must have been from Poly-Paks around 1977 and they were notorious for selling blems and rejects. Have to compare die fonts to see who made it.
I checked ads and could not find it. Did see ads for the LM300 sold as a "super 723".

Signetics NE550 almost identical to LM723 but a lower reference voltage 1.6V despite a zener being there?
I wonder how IC designs were licensed back there. The shenanigans with the reference must be over patent issues, or to get a little better numbers off the datasheet for sales.

The large headroom requirement is certainly one of the downsides of the original 723. All clones used a lower reference voltage (down to 1.2V), so you could make, e.g., a 2.5V regulator fed from 5V supplies, impossible with an original 723 with a 7.2V Zener in it.
 

Offline floobydust

  • Super Contributor
  • ***
  • Posts: 4508
  • Country: ca
Re: LM723 die pictures
« Reply #64 on: February 18, 2020, 04:02:39 am »
I notice the extra pad, perhaps for burn-in. I'm not sure what transistor it connects to and why you'd bother there.
Looking at µA723 aerospace vendors, the usual Fairchild, Philips, TI were the big manufacturers. Raytheon Systems LTD Weapons, Intersil, AMD, Bunker Ramo, Adelco Elektronik, Lockheed Martin were lesser known sources for the 723.
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #65 on: February 18, 2020, 08:09:06 am »
The extra pad is Vz which is missing on the metal can package. It's not a transistor but a zener.

Green is P, grey is N, you can imagine how that works. The N also makes emitter balancing resistors for the power tranny.
« Last Edit: February 18, 2020, 08:14:07 am by magic »
 

Offline floobydust

  • Super Contributor
  • ***
  • Posts: 4508
  • Country: ca
Re: LM723 die pictures
« Reply #66 on: February 18, 2020, 08:21:43 pm »
I'm not fluent in die-> transistor interpretation, especially seeing 8 mask layers.
Added "History of Semiconductor Engineering" by Dr. Bo Lojek to my reading list. Quote the hate:
"A Fairchild researcher trained a frog to jump at the sound of a bell. The researcher measured the distance the frog would jump, then removed the frog’s legs and rang the bell again. The frog did not move, thus proving the Fairchild R&D group hypothesis that removing a frog’s legs deafens the animal.”  Robert J. Widlar, describing Fairchild’s R&D group in 1967.
 

Offline Wolfgang

  • Super Contributor
  • ***
  • Posts: 1591
  • Country: de
  • Its great if it finally works !
    • Electronic Projects for Fun
Re: LM723 die pictures
« Reply #67 on: February 18, 2020, 08:32:06 pm »
I'm not fluent in die-> transistor interpretation, especially seeing 8 mask layers.
Added "History of Semiconductor Engineering" by Dr. Bo Lojek to my reading list. Quote the hate:
"A Fairchild researcher trained a frog to jump at the sound of a bell. The researcher measured the distance the frog would jump, then removed the frog’s legs and rang the bell again. The frog did not move, thus proving the Fairchild R&D group hypothesis that removing a frog’s legs deafens the animal.”  Robert J. Widlar, describing Fairchild’s R&D group in 1967.
A classic. Another one:
A patient consults a doctor and tries to persuade the doctor that he is alread dead. The doctor uses all kinds of scientific and pragmatic arguments that this could not be the case, like the patient breathes, has an audible heratbeat, can roll his eyes and move, talk and what not - to no avail, the patient insists that he is dead. Then, as a last result, he asked the patient: do dead people bleed ? The patient thinks a while, and then replies: no, dead people do not bleed. In a swift movement, the doctor takes a small  knife and makes a small cut in one of the patients fingers - and a stream of blood comes out. The doctor says: See - you cannot be dead. The patient says: No - I just was wrong saying that corpses dont bleed.
« Last Edit: February 18, 2020, 09:25:23 pm by Wolfgang »
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #68 on: February 18, 2020, 09:16:16 pm »
I'm not fluent in die-> transistor interpretation, especially seeing 8 mask layers.
I don't need to understand anything, I know it has to be a zener from the pinout :-DD

A quick glance at obviously-NPN-transistors reveals that grey on green is emitter on base. This is a BE junction intended for reverse biasing, though actually no one can stop you from forward biasing it externally if that's your thing ;)
 

Online Noopy

  • Frequent Contributor
  • **
  • Posts: 858
  • Country: de
    • Richis-Lab
Re: LM723 die pictures
« Reply #69 on: February 19, 2020, 07:25:48 pm »
Here we have a (three) Tesla MAA723:

https://www.richis-lab.de/LM723_04.htm




Three different packages and date codes but always the same die:





Hey there are cracks all over the surface!  ???
 
The following users thanked this post: Wolfgang, imo

Online Noopy

  • Frequent Contributor
  • **
  • Posts: 858
  • Country: de
    • Richis-Lab
Re: LM723 die pictures
« Reply #70 on: February 20, 2020, 10:37:59 pm »
And here a very old LM723CN:

https://www.richis-lab.de/LM723_05.htm








It´s the same die as in the LM723CJ:

https://www.richis-lab.de/LM723_01.htm


Some corrosion on some bondpads:

 
The following users thanked this post: Wolfgang

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #71 on: February 21, 2020, 10:34:58 am »
A small analysis of the National bandgap circuit.

The divider driving Q21,Q22 has total resistance of 14.879kΩ including 2.58kΩ on the lower side. Q21,Q22 base voltage is therefore roughly 1.240V. Not sure why I got 1.23V before.

Q24 base is then 710mV, yielding Q24 current of give or take 100µA for an additional voltage drop of 40mV on R16 and actual Q21 voltage closer to 1.23V. Ha, I was right even if I got my math wrong :) Q23 influence is even less and both are insignificant in the grand scheme of things.

R15 including the trimmable part is 12kΩ and voltage across that is circa 600mV. We estimate 50µA combined Q21,Q22 current.

Assuming 1µA base current in Q20, we get 2mV drop across R8 and therefore mirror gain of +6%. That's insignificant, probably on the order of base current errors. The mirror is unity gain. (The PNPs themselves are identical).

Each of Q21,Q22 has 25µA current and therefore about 1mS transconductance / 1kΩ intrinsic emitter resistance. Q22 has an additional 2kΩ of physical resistance in R14.

Assuming 50V Early voltage for Q21 and Q19 (pulled outta my ass for easy calculations) we get 1MΩ combined collector impedance on their junction node / Q6 base.

Let's do the loop gain. Rising Q21 Vbe by 3mV we get 3µA extra current in Q21 and 1µA in Q22. That's 4µA total and 48mV extra across R15. So for a 51mV rise of Q21,Q22 base wrt ground we get 3µA rise in Q21 and 1µA in Q22 and therefore -2µA at the Q21,Q19 junction and a -2V shift of Q6 base voltage and the VREF node.

That's not a lot :-//

Could it be that this circuit just compensates a zener? But how would that work :wtf:
 
The following users thanked this post: Noopy

Offline magic

  • Super Contributor
  • ***
  • Posts: 3345
  • Country: pl
Re: LM723 die pictures
« Reply #72 on: March 01, 2020, 08:03:38 pm »
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).
I learned that this is wrong and that buried layer is not visible externally. The only sign of its presence is a slight vertical depression on the surface of the epitaxial layer and even that is shifted vertically with respect to the real location of the buried structures. This phenomenon is known as "buried layer pattern shift" and a lot has been written about it because it's a nuisance for IC fabs - it makes it difficult to align surface features to the buried ones.

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 :-//
This appears to be the case. The mustard colored areas at the perimeter of isolation islands have to be P diffusions produced in the same step as NPN bases. I really tried to find some confirmation in literature available online and failed, except for this somewhat ambiguous suggestion in "Analysis and Design of Analog Integrated Circuits":

Quote
The terraced effect on the surface of the device results from the fact that additional oxide is grown during each diffusion cycle, so that the oxide is thickest over the epitaxial region, where no oxide has been removed, is less thick over the base and isolation regions, which are both opened at the base mask step, and is thinnest over the emitter diffusion.

Additional evidence is the fact that the long green N+ silicon resistor on the LM723J die appears to overlap slightly some isolation islands and yet makes no electrical contact with them, so it needs to have some grounded P silicon below it for isolation. Also, the MIL723 die has places where a P resistor connects an NPN base with the perimeter of the isolation island and it all looks perfectly smooth.

Armed with this knowledge we can have another look at the National LM723J and its D2 diode.

[attachimg=1]

On the right, we see the emitter and base of Q6, covered by metal and surrounded by "mustard", which has to be a surface diffused collector extending all the way to the isolation diffusion around the island. Above Q6, an N+ diffusion is made into the collector and part of it branches off to the right and crosses two vertical metal traces. I thought that it connects with the metal on the far right, which is GND, but with some magnification of the image it becomes apparent that it actually does not.

Since the N+ diffusion is fed constant current from Q3, it forms a zener diode with the grounded P silicon below (D2 on the National schematic) and assumes a fixed potential of a few volts above ground. So what's the point of the stub extending to the right? The stub is just long enough to leave the isolation island of Q6 and reach an isolation diffusion, which has higher concentration of P dopant than Q6 collector. Higher doping is known to reduce breakdown voltage (see Linear AN-82, buried zeners chapter) and I presume it's the reason why D2 has lower breakdown voltage than D1, which is a plain BE junction.

The metal connection to D2 cathode branches to the left and connects with structures in another isolation island. It appears to be connected to the bulk of the island and a P diffusion in it. The P diffusion has a comb-like N+ diffusion inside, forming a base-emitter diode. Reverse bias across this diode is no more than 5.9V (VREF minus two diode drops) and almost zero if we assume that D2 voltage is 5.7V. Therefore the chance that this diode forms some super-fancy large area zener appears rather slim. Furthermore, such zener would be in series with the wimpy D2 zener, so what's the point?

I think that the "comb" really is a compensation capacitor and nothing more. This is a bandgap chip, plain and simple. The datasheet schematic checks out 100% and it all makes sense.

Let's do the loop gain. Rising Q21 Vbe by 3mV we get 3µA extra current in Q21 and 1µA in Q22. That's 4µA total and 48mV extra across R15. So for a 51mV rise of Q21,Q22 base wrt ground we get 3µA rise in Q21 and 1µA in Q22 and therefore -2µA at the Q21,Q19 junction and a -2V shift of Q6 base voltage and the VREF node.

That's not a lot :-//
Yep, loop gain is only some 20~40dB. But current through Q5 is almost 0.5mA due to the resistive divider, so Q5 emitter output impedance is some 50~70Ω depending on temperature. Then, closed loop output impedance is a few ohms or maybe a bit under one ohm. I can believe it.
« Last Edit: March 02, 2020, 06:04:45 am by magic »
 
The following users thanked this post: Noopy

Online Noopy

  • Frequent Contributor
  • **
  • Posts: 858
  • Country: de
    • Richis-Lab
Re: LM723 die pictures
« Reply #73 on: March 03, 2020, 08:37:18 pm »
Thanks magic for your very interesting analysis!  :-+


I have a new LM723-variant!  ;D

Thanks to BravoV I was able to take a look at the Fairchild UA723:







=> https://www.richis-lab.de/LM723.htm
=> https://www.richis-lab.de/LM723_06.htm

 :popcorn:

Edit: I´m sorry for my german english.  ;D
« Last Edit: March 04, 2020, 04:22:25 am by Noopy »
 
The following users thanked this post: BravoV, Wolfgang

Offline BravoV

  • Super Contributor
  • ***
  • Posts: 7327
  • Country: 00
  • +++ ATH1
Re: LM723 die pictures
« Reply #74 on: March 04, 2020, 12:00:20 pm »
Thanks and credit should goes to you, Noopy, I really enjoy it.  :clap:  :-+

As I'm noob at interpreting from the die shot, so how this Fairchild UA723 differs from others ?


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf