Opamps - Die pictures

[Hydron]

Thanks David, that's a very useful doc!

The AD542 I've got my eye on is bootstrapped, so CMRR might not be quite as important as normal for an input buffer, but all of these options have far better specs in other areas too (and I guess I could look at some of the low voltage auto-zero amplifiers, even the 5V types if I drop the bootstrapped supply voltages down a bit).
I'm wondering if the audio types might be a good option for some stuff where offset is irrelevant (as long as it doesn't drift) but low noise would be nice, e.g. Vref buffers, or where offset is otherwise accounted for. Will need to compare for other downsides though.

Is there anything you're aware of that's anywhere near the price/performance ratio of the OPA14x types but in DIP? Or is it really just the OPA13x or TLE20x1 already in the doc (found these as well, but the OPA14x seem a lot nicer)? I normally avoid THT in anything I'm designing but in this case I'm messing with some old gear, so see some SOIC/VSSOP -> DIP adapter boards in my future Still better than paying OPA627 money though!

@Noopy - If I make the swap for a OPA14x I might have a spare AD542 coming available to post to you if it's of interest

[Noopy]

@Noopy - If I make the swap for a OPA14x I might have a spare AD542 coming available to post to you if it's of interest

Up to now I don´t have a AD542. So I would be happy to get one. 

[David Hess]

Thanks David, that's a very useful doc!

It is just what I came up with when looking for modern replacements for the AD542, but it applies more generally to precision high impedance buffering.

(and I guess I could look at some of the low voltage auto-zero amplifiers, even the 5V types if I drop the bootstrapped supply voltages down a bit).

Chopper stabilized and auto-zero parts have higher input bias current, and questionable input current noise.  They work for low impedance circuits, but use in high impedance circuits has historically been a problem.  There are modern parts which claim to work in high impedance circuits, but I have not verified this.

I'm wondering if the audio types might be a good option for some stuff where offset is irrelevant (as long as it doesn't drift) but low noise would be nice, e.g. Vref buffers, or where offset is otherwise accounted for. Will need to compare for other downsides though.

They should be a good lower cost solution.  Parts like the OPA1641 seem to be ideal for a low cost design.

Is there anything you're aware of that's anywhere near the price/performance ratio of the OPA14x types but in DIP? Or is it really just the OPA13x or TLE20x1 already in the doc (found these as well, but the OPA14x seem a lot nicer)? I normally avoid THT in anything I'm designing but in this case I'm messing with some old gear, so see some SOIC/VSSOP -> DIP adapter boards in my future Still better than paying OPA627 money though!

That is what I was looking for!  There is a column which shows the package type and I was originally looking for DIP replacements for the AD542.  As you identify, the OPA627 is a good default if nothing better can be found, but wow it is expensive.  The OPA132/OPA134 were possibilities except are not available or not available in DIP since I started that list.

Later I made a more comprehensive list to restock my parts cabinet with modern parts, and it revealed the TLE2071 which is a possibility, but not nearly as good at the OPA140 series.

What I finally decided is that the OPA140/OPA141/OPA145/OPA1641 performance was so much better that I would accept the inconvenience of a SOIC part.

[Hydron]

(and I guess I could look at some of the low voltage auto-zero amplifiers, even the 5V types if I drop the bootstrapped supply voltages down a bit).

Chopper stabilized and auto-zero parts have higher input bias current, and questionable input current noise.  They work for low impedance circuits, but use in high impedance circuits has historically been a problem.  There are modern parts which claim to work in high impedance circuits, but I have not verified this.

I had a quick look at my DMM6500 and it looks like a bootstrapped LMP2021 auto zero part is used as the input buffer, hence my thoughts about this. I was also wondering about the input noise aspect but if they're good enough for a 6.5 digit front end maybe worth looking at?

[Kleinstein]

In my DMM circuit I have a MCP6V66 with a bootstrapped supply at 2 positions: one as a buffer for the divider with some 500 K impedance and one at the ohms current source with 1.5 M as the largest resistor. The amplifier works reasonable OK, but the current noise (50 fA/sqrt(Hz) range) gets noticible. The input bias is a bit higher than with a OPA140 or similar, but still good.  With quite some of the AZ OP-amps the worst case input current is rather close to the supplies and for a point in the center the input current tends to be quite a bit better.  Like with many low bias parts the typical input current is often quite a bit lower than the test limits.
Noise wise the LMP2021 should be well OK for a 6 digit and even 8 digit DMM input. It is a bit odd type with higher noise at low gain. For the voltage input I use a AD8628 with bootstrapped supply (some 22 nV/sqrt(Hz) and a measured 40 fA/sqrt(Hz) current noise).

For a DIP part the OPA135/2135 can be an option, though an audio part with not so good offset. I consider this the predecessors to the OPA1641/2.

[David Hess]

I had a quick look at my DMM6500 and it looks like a bootstrapped LMP2021 auto zero part is used as the input buffer, hence my thoughts about this. I was also wondering about the input noise aspect but if they're good enough for a 6.5 digit front end maybe worth looking at?

The problem I see is that the LMP2021 has a current noise 437 times higher than an OP140, and it dominates the input noise of a 10 megohm system.  With a low source impedance this will not matter, but the lower limit of the source impedance depends on the series input protection of the multimeter.  This might be 500 kilohms, so the input current noise of the LMP2021 still dominates and the OP140 would produce less total input noise by a couple times.  Since they used the LMP2021 anyway, that level of noise must not be significant for 6.5 digits.

Of course the OP140 would require some external means of automatic zero correction, and inconveniently lacks any offset null capability.

[Andy B]

Hi all,

This is my first post on this forum. So if have any comments regarding what I should or shouldn't do in relationship to the post I would welcome it but keep it short so it doesn't clutter this thread.

What this post is about is a component analysis of a LM358 die which is a dual op-amp. It's my first attempt so I might be wrong so feel free to correct me. Even though I think I figured it out I still have queries for which I would welcome input.

For this post I'm just looking at one side of the op-amp and not the current bias (although I have also figured that out and will post later).

The die shot I'm using  is one of Richi's (who started this thread) LM358 die shots and according to his index (https://www.richis-lab.de/Opamp.htm) is a National Semiconductor LM358. It's from a Western Digital Cavier 22500 HDD.

The link to the original article and image is https://www.richis-lab.de/HDD_WD_Caviar_22500.htm#LM358.

Richi actually didn't post it to this forum as far as I can see so a just added this post after the last message in this thread.

The reason I chose this over other LM358/LM124/LM324 die shots is because of it's clarity that allows the doped regions to be determined. Big thumbs up to Richi for this brilliant shot because I was struggling using other die shots.

Note that I mentioned LM124/LM324 opamps as well as these are the quad version of the LM358.

I've scaled Richi's original image by a factor of 2 below to allow for a quick comparison with the image at the end with the components outlined.



Credit: Richi  @ https://www.richis-lab.de/

The die presented here uses a five collector transistor current mirror (Some LM358 designs use a three collector transistor current mirror).



The best schematic I can find for the LM358 that uses the five collector transistor current mirror is below.



Its from a LM358 Fairchild datasheet.
https://web.archive.org/web/20171003174216/https://www.fairchildsemi.com/datasheets/LM/LM358.pdf

It shows all components except for a missing "diode" (more about that later) and a bunch of components represented by the current source on the left of the schematic.

The current source is actually representative of the "common bias" used by both opamps.

For the analysis the big mystery to me was figuring out where the output stage components Q18, Q21, and R2 were. I figured out the others but was left with the following space to figure out.



On close examination I figured out the Q18 was actually a small lateral NPN transistor which I didn't notice at first sight as shown in the picture below. This transistor along with R2 acts as a current limiter that steals base current from Q19 so it only needs to be small.



So that left R2 and Q21 to be discovered.

R2  caries the source current for the output stage. On some schematics I have seen it is given as 25 ohms. On closer examination it seems R2 is actually two parallel resistors each in it's own P well (green) between the fingers shown.



That left Q21 which is the PNP output sink transistor shown own the schematic. On closer examination of the doped areas in the area I determined that this transistor is the outer part of this region and is almost two separate transistors. I marked the PNP doped regions on the following picture to make it easier to see. Note the top base region (Q21.1) seems thicker than the bottom base region (Q21.2). This isn't the case on other LM358/LM124/LM324 die shots I have seen and I have no clue why it is the case here.



On examining the die picture I determined there was an additional component not mentioned in the schematic. Its between the emitter of Q11 and the base of Q15 and runs to Vee.



If you look closely you can see its actually a NPN transistor with an unconnected base.

In National Semiconductor's LM358 datasheet simplified schematic this is shown as a resistor between Q11 and Q12 (different transistor labelling).
https://web.archive.org/web/20240723125259/http://web.mit.edu/6.115/www/document/LM158.pdf

A possible explanation of this is given by the following quote  in regards to LM124

"R1 proved to be a floating base NPN transistor, perhaps to be used as an over temperature and/or an over voltage shut down device to protect the circuits output [6]"

"Variations in SET pulse shapes in the LM124A and LM111"  Paywall: https://ieeexplore.ieee.org/document/1045536

[6] is a reference to Patent US4011470 CIRCUIT UTILIZING OPEN-BASE TRANSSTOR AS LEAKAGE BYPASS DEVICE : William Folsom Davis, Thomas Marinus Frederiksen
https://patents.google.com/patent/US4011470A/en

Given one of the patent assignees is "Thomas Marinus Frederiksen" who was the designer of the LM324/LM358 its a good bet that this is exactly what this is.

In the Texas Instrument application note "Application Design Guidelines for LM324 and LM358 Devices" it mentions there are two diodes in the op-amp. Given I cant find any other component that suits I assuming that this is the diode as it's somewhat acting like a reversed diode so I tagged it with "D1".

Finally one component which I'm sure is R1 is an IC resistor type I have never seen before. It could be a version of a pinched resistor but it almost looks like somebody soldered a resistor in! Could somebody enlighten me on this.



So here is the final result. Not as pretty as some component pictures I seen here but it does the job.



Cheers
Andy

[magic]

Yeah, that's what it looks like.

R1 should be a pinch resistor and it looks like one, not sure what your doubt is.
Q18 is a vertical transistor, probably a typo.

Sinking overcurrent protection is funny, because it looks like Q15 is protected by Q12 limiting its base current and Q21 is protected by having crap β.

Attached is one of the better LM358 images out there, the website which hosted it went down a few years ago. It's a slightly different variant, input emitter followers are biased differently and those output stage components aren't combined. And yes, it looks like the dummy transistor is simply used for its collector leakage, not entirely sure what's the deal with that.

[Noopy]

I agree with you, Andy B and I agree with magic. Everything fine! 

Unfortunately that´s a very old picture. I don´t have the die in my archive so we won´t get a better picture. (Until I decap another LM358.)


Regarding the circuit:

Yes, it seems this strange transistor with its open base seems to be such a leackage current compensation like they describe it in the patent. Interesting!

Yes, Q21 contains more or less two transistors. 

[Andy B]

Big thanks Magic for the reply.

"Q18 was actually a small lateral NPN transistor" is a boo boo. It is as you say a vertical NPN transistor.

"R1 should be a pinch resistor and it looks like one, not sure what your doubt is".  It's probably the die shot gives it much more of a 3D look than all other shots of pinch resistors I have seen which had me wondering.

"Sinking overcurrent protection is funny, because it looks like Q15 is protected by Q12 limiting its base current and Q21 is protected by having crap β."

Was wondering if there was any sinking overcurrent protection. This is a good explanation.

In regards to the  die image  you posted its a ST Microelectronics LM358 which uses a 3 collector transistor current mirror. The resistor R1 here is a pinch resistor which looks the way I'm used to on die shots.

"And yes, it looks like the dummy transistor is simply used for its collector leakage, not entirely sure what's the deal with that". The patent goes into great detail but its a little over my head.

Cheers
Andy

[Andy B]

Thanks Noopy for the reply.

"Unfortunately that´s a very old picture. I don´t have the die in my archive so we won´t get a better picture. (Until I decap another LM358.)".  It was good enough for me to figure this out so that's fine by me 

Cheers
Andy

[Noopy]

The "open base transistor" sinks current so that Q13 sinks less current. I assume otherwise Q13 would sink too much current at high temperatures.

For all the others: I was able to send Andy B a higher resolution picture of the LM358. The quality is not a lot better but at least the resolution is higher.
There was a time when small pictures were much appreciated. 
I will update my website as soon as I find the time to do it...

[magic]

It's probably the die shot gives it much more of a 3D look than all other shots of pinch resistors I have seen which had me wondering.

That stuff isn't flat so that's normal. Microscope images may look flat because they are often illuminated at 90° angle to the surface, but Noopy used a different technique with more angled light.

ST Microelectronics LM358 which uses a 3 collector transistor current mirror

I counted four

It's common practice to run "NPN base" P diffusion over isolations, and here this diffusion extends into the input stage bias PNP to collect and discard the current which would normally go into the remaining two collectors which are absent.

[Andy B]

Attached is one of the better LM358 images out there, the website which hosted it went down a few years ago. It's a slightly different variant, input emitter followers are biased differently and those output stage components aren't combined.

Referring to the picture of the ST LM358 that Magic posted above.

When is a LM358 not a LM358? Well you could answer a Chinese op-amp clone (though it's more likely a cheap Chinese op amp will be a cloned LM358) but the answer I'm after is when it's a ST LM358! ST's LM358 is actually a dual version of their single TS321 op-amp!

Here's another excellent die shot of the ST LM358 (rotated and reduced to suit the discussion below)


Credit: Marmontel at https://www.siliconpr0n.org/

Article and HD version:https://www.siliconpr0n.org/archive/doku.php?id=marmontel:stmicro:lm358

Here's a excellent die shot of the TS431 that Zeptobars did.


Credit: Zeptobars https://zeptobars.com

HD Version: https://zeptobars.com/en/read/ST-TS321-SOT23-opamp-LM358A-LM324

And here's a annotated version from the LTspice forum (provenance unknown) that uses the Zeptobars die shot.



And here's the schematic from the LTspice forum (provenance unknown)



The input stage of the TS321 is similar to Mororola's version (now ON Semiconductor) as shown on the schematic below.  Note the bias transistors with multiple collectors below are split into single transistors in the TS321 schematic above but are labelled accordingly.


Source: https://www.onsemi.com/pdf/datasheet/lm358-d.pdf

The following changes in the output stage can be observed (including as you observed the output components aren't combined):

1) Q18 (Q23 here) in the LM358 schematic is moved to it's own tank.

2) Q21 (Q26 here) has been moved to before the output over current resistor and is the large transistor on the top of the TS321 die and has it's own tank.

3) R2 (R5 here) now has it's own tank.

In regards to "input emitter followers are biased differently" I think you may be referring to that one collectors of Q18 and Q20 in the above Motorola as in TS321 schematics is shorted to the base of each transistor. The explanation of this and the original variation where they are shorted to Vee is given in the patent link below for which Frederiksen notably gave credit to in a paper just before the LM324 was released.

"Transconductance reduction using multiple collector pnp transistors in an operational amplifier" assigned to R Russell & J Solomon
https://patents.google.com/patent/US3801923A/en

Cheers
Andy

[David Hess]

In regards to "input emitter followers are biased differently" I think you may be referring to that one collectors of Q18 and Q20 in the above Motorola as in TS321 schematics is shorted to the base of each transistor. The explanation of this and the original variation where they are shorted to Vee is given in the patent link below for which Frederiksen notably gave credit to in a paper just before the LM324 was released.

"Transconductance reduction using multiple collector pnp transistors in an operational amplifier" assigned to R Russell & J Solomon
https://patents.google.com/patent/US3801923A/en

The old National Application note A discusses what is going on with transconductance reduction and 324/358 designs toward the end.  The significance is that transconductance reduction allows the compensation capacitance to be smaller, so while a 741 type of operational amplifier has a relatively huge 20 picofarad integrated capacitor, the 324/358 only requires 5 picofarads.  This makes the 324/358 cheaper because it requires less area, and allows duals and quads to fit within the package.

The smaller compensation capacitance also allows lower quiescent current to achieve a given bandwidth and slew rate.

In the 1970s Fairchild had their own improved 324/358 parts with class-AB output stages, and a single part as well with offset null capability, but I do not know what happened with them.

[Andy B]

The old National Application note A discusses what is going on with transconductance reduction and 324/358 designs toward the end.

It's on PDF page 16 under the section "A. Transconductance Reduction" and Fig 28 shows the two different methods to reduce transconductance using split collectors that were used by the two different LM358s above.

Thanks for that David!

Cheers
Andy

[Andy B]

Big thanks to Noopy who sent me a higher resolution version of his LM358 die shot.

Here it is!


Image provided by Noopy @ https://www.richis-lab.de/


And here's the updated annotated version!



Bias circuit annotation finished but have to create a circuit diagram to go with it. Will post once done.

Cheers
Andy

[David Hess]

The old National Application note A discusses what is going on with transconductance reduction and 324/358 designs toward the end.

It's on PDF page 16 under the section "A. Transconductance Reduction" and Fig 28 shows the two different methods to reduce transconductance using split collectors that were used by the two different LM358s above.

There are a bunch of ways to implement transconductance reduction, but I got the idea from studying various published schematics that the manufacturers kept some of them secret, or published misleading schematics.  The National Semiconductor application note was very helpful.

Usually the details do not matter so simplified schematics are good enough, but not always.  For instance the difference in input bias current cancellation circuits between early PMI and early Linear Technology parts reveals why the PMI parts had higher distortion, and I guess they also explain why the Linear Technology parts have correlated input current noise so benefit from matched source impedance even when they have input bias current cancellation.

[Andy B]

Here's the annotated die for the National Semiconductor LM358 complete with the common bias components.



And here's the schematic for the common bias circuit.



Derived from Figure 20: "Schematic of the LM124 [7]" in the following linked document. Redrawn with corrections and additions by me.

"Variations in SET pulse shapes in the LM124A and LM111"
https://ieeexplore.ieee.org/document/1045536 (paywall)
[7] Y. Boughassoul to be published NSREC 2002

I couldn't find  J1 (N channel JFET)  initially but on chasing down the components I figured out it's a long straight N channel pinched by P regions that runs from Q104 to Vcc. Correct me if I'm wrong.

Final count of components is:

Epi-Fet        1
Diodes         2
Resistors      7
Transistors    51
Capacitors     2

Through coincidence or not it matches the component count below extracted from a Texas Instruments (TI) document.
Note it is unclear whether this component count is for the TI LM358 or for the National Semiconductor (which TI took over) LM358 or both



Figure 1-1. Device Schematic from Data Sheet.
"Application Design Guidelines for LM324 and LM358 Devices"
https://www.ti.com/lit/an/sloa277b/sloa277b.pdf

Cheers
Andy

[magic]

I only don't know the logic behind this circuit. The goal was supposedly to stabilize input bias current over temperature (IIRC) but it's not entirely obvious why it should happen here. ST and ON use a different circuit and Chinese clones routinely use a PTAT generator.

It's on PDF page 16 under the section "A. Transconductance Reduction" and Fig 28 shows the two different methods to reduce transconductance using split collectors that were used by the two different LM358s above.

They all use the Fig 27 scheme. Something similar to Fig 28 I have only ever seen on the published schematic of LM12 power amplifier.

Through coincidence or not it matches the component count below extracted from a Texas Instruments (TI) document.
Note it is unclear whether this component count is for the TI LM358 or for the National Semiconductor (which TI took over) LM358 or both

That's a schematic from TI datasheet and those "component count" tables were a TI thing.
Zeptobars has TI LM358, it's same topology. One difference is that they seem to use an emitter diffusion over isolation area instead of the dummy transistor. I guess that's what the one solitary "diode" per channel is.

[Andy B]

It's on PDF page 16 under the section "A. Transconductance Reduction" and Fig 28 shows the two different methods to reduce transconductance using split collectors that were used by the two different LM358s above.

They all use the Fig 27 scheme. Something similar to Fig 28 I have only ever seen on the published schematic of LM12 power amplifier.

I made a boo boo here as on a sleepy look thought figure 28 showed the collector of one transistor was tied back to the base as done on the variation for the ON/Motorola/ST version (outlined in the patent) rather than to Vee as done for the National Semiconductor LM358. Sorry about that and thanks for correcting me.

That's a schematic from TI datasheet and those "component count" tables were a TI thing.
Zeptobars has TI LM358, it's same topology. One difference is that they seem to use an emitter diffusion over isolation area instead of the dummy transistor. I guess that's what the one solitary "diode" per channel is.

Big thanks for this snippet of info. Certainly solves the mystery for me of the missing diode. I will however leave the open-base transistor designation as D1 on the annotated die shot for the time being as it means the two are consistent as it largely acts as a reversed diode unless the consensus is otherwise (I have seen it elsewhere labelled as QR1 from a radiation SET analysis and as a resistor on some datasheets).

Cheers
Andy

[Noopy]

Oh, I didn´t realize there were new posts here. 

I have to put two OPA140 pictures in between:




I have taken a darkfield picture of the tuned resistors. Now we can see the tuning marks a lot better. It´s interesting how blurry they are compared to the old designs. The "lines" in the background seem to have nothing to do with the resistor. Perhaps that is something that is needed for the laser tuning process. 




Removing the metal layers reveals that a not inconsiderable proportion of the surface is filled with dummy structures. The circuit appears to be significantly less complex than one might expect at first glance at the modern process.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/92x08XL.jpg (17MB)


https://www.richis-lab.de/Opamp88.htm#part2

 

[magic]

So here's the TI logo.

And I guess the input JFETs are confirmed now. BTW, the annoying thing about covering JFETs with metal has apparently been going on for a while at TI. I found TLE2062 on siliconpr0n - similar thing.

I have taken a darkfield picture of the tuned resistors. Now we can see the tuning marks a lot better. It´s interesting how blurry they are compared to the old designs. The "lines" in the background seem to have nothing to do with the resistor. Perhaps that is something that is needed for the laser tuning process. 

Maybe a matter of magnification? Note that these laser lines are really thin actually.
No idea what are those dense lines but it looks funny when it all falls apart and they flap around in the breeze
And why is the silicon under resistors connected to something. Is it supposed to be R||C maybe?

The circuit appears to be significantly less complex than one might expect at first glance at the modern process.

There is no CMOS rubbish and many of the BJTs look like they are just large common centroid pairs, but there is still over 100 of them.

Maybe it would be possible to figure out this chip by focusing at single metal layers at high magnification, but it would take a little longer than LM358

[Noopy]

I have taken a darkfield picture of the tuned resistors. Now we can see the tuning marks a lot better. It´s interesting how blurry they are compared to the old designs. The "lines" in the background seem to have nothing to do with the resistor. Perhaps that is something that is needed for the laser tuning process. 

Maybe a matter of magnification? Note that these laser lines are really thin actually.
No idea what are those dense lines but it looks funny when it all falls apart and they flap around in the breeze
And why is the silicon under resistors connected to something. Is it supposed to be R||C maybe?

Hm, the edges are blurry and not 100% straight. I don´t think that is a magnification topic.
Are you sure the silicon under the resistors is connected to something? These contacts at the edges look more like they contact this blurry blue layer. In my view...
It could be a RC thing, that is possible.

The circuit appears to be significantly less complex than one might expect at first glance at the modern process.

There is no CMOS rubbish and many of the BJTs look like they are just large common centroid pairs, but there is still over 100 of them.

Maybe it would be possible to figure out this chip by focusing at single metal layers at high magnification, but it would take a little longer than LM358

Unfortunately the dummy structures make it hard to identify more than the uppermost layer.

[Andy B]

This is some more analysis  in relationship to my annotated version of Richi's die shot of the National Semiconductor LM358 and the mysterious Q21/Q18/R2 combo component as I showed in this diagram



I decided to draw up what I thought a front side section of this component would look like without the Q18 transistor.

I've drawn in also the resultant internal components



The connection "R" is the R2 resistor connection to Q20's emitter.

Note there is two diodes to the base connection from the R connection.

Whilst these diodes seem odd they don't conduct when Q21 is conducting (output current sink mode) as there's no voltage across R2.

And they don't conduct when R2 is conducting (output current source mode) as Q21's base is at a higher voltage than R.

Effectively this acts as a resistor (R2) in output current source mode and as a transistor Q21 in current sink mode.

Pretty ingenuous I say especially when you add the Q18 transistor into the midst of this so hats off to Tom Frederiksen the designer of this op-amp!

Feel free to correct me if I'm wrong or if there is something else worth mentioning.

Cheers
Andy