Opamps - Die pictures

[Noopy]

Today I have an old LM360 for you:





With the same die you can build a LM361. 

https://www.richis-lab.de/Opamp10.htm

 

[Noopy]

Today I have a fake NE5534 for you:

https://www.richis-lab.de/Opamp11.htm

1,40€ for ten of these bugs including shipping. 




Tried to fake a Texas Instruments Logo? I don´t know. 




The number 659 doesn´t really fit. It seems to be a RC4558. 

 

[magic]

You were lucky
I got LM358 and so did some Turkish poster here the other day.

If you are looking for recycled authentic NE5532/4 from China it may be hard. I tend to look for auctions with positive feedback and preferably including buyer's photos of the delivered ICs. Prefer auctions which show unblurred manufacturer logo. But it's still lottery - I once ordered from an auction with a real photograph of ON Semi NE5534 and got a mix of recycled chips from unknown manufacturers with fake NXP branding (NXP never made those chips). At least they weren't completely fake.

edit
Wait, why all the pads look like they had been bonded?
Did they connect the second channel's bonding pads to the compensation/balance pins?
Or is it actually a normally bonded dual opamp, not pin-compatible with single opamps and there will be smoke if somebody tries to use it?

[Noopy]

I searched for the cheapest NE5534 to find a fake.

I assume it's a RC4558 with the pinout of a RC4558. Probably they recycled RC4558 and changed them to NE5534 to make more money.
Normal (NE5534) connection will probably give you magic smoke...

[magic]

The shape of the packages looks Chinese and I found this exact die in a few different fake opamps. It's smaller than Raytheon or TI dice. It's some Chinese clone of RC4558 with fake markings.

Smarter fakers modify the pinout to be compatible with single opamps.

[Noopy]

The shape of the packages looks Chinese and I found this exact die in a few different fake opamps. It's smaller than Raytheon or TI dice. It's some Chinese clone of RC4558 with fake markings.

That would explain why there are the numbers 659 which don't match with a RC4558.

[magic]

Some Chinese opamps:
https://www.eevblog.com/forum/beginners/opamp-input-offsets-working-in-the-opposite-direction-to-what-i-expect/25/
https://www.eevblog.com/forum/projects/whats-inside-the-cheapest-and-fakest-jellybean-opamps/
Have you really not seen those threads yet?

Yes, they sometimes incude some numbers and logos, nobody knows what they mean.

[Noopy]

Of course I have seen those threads! 
But that was some time ago and I have forgotten you already had found the numbers 659.

Nevertheless I´m not sure whether these 659-dies are conterfeits.
I know the datasheet describes the RC4558 to be bigger but zeptobars already found one smaller than that (bigger than this one here). Perhaps they did another die shrink and the opamp found here is no fake-RC4558 but a real one?

[Noopy]

I have something for you AMD built just before they started up their 7nm-fabrication. 







https://www.richis-lab.de/Opamp12.htm


Looks quite similar to the Silicon General LM310 (https://www.richis-lab.de/Opamp08.htm) but has some differences. 

[David Hess]

I have something for you AMD built just before they started up their 7nm-fabrication. 

Few people remember that AMD second sourced linear ICs.  I remember having to remove them as a supplier because too many parts had popcorn noise which is a processing problem.  I heard cussing over the reliability of their UVEPROMs also.

[Noopy]

I took pictures of a OPA676 (thanks to dzseki).
That´s a very interesting opamp! It can go up to 185MHz with a slewrate of 350V/µs and it has two differential inputs which you can switch as you want.
Even more interesting: The OPA676 is integrated on an universal die. Something like an analog gatearray.






The two metal layers are designed by Burr-Brown. The "analog gatearray" is supplied by VTC:






More pictures here:

https://www.richis-lab.de/Opamp13.htm

 

[Noopy]

Today I have a LF355 for you. This one was manufactured 1977. 




Although the LF355 is the slowest opamp of the LFx5x-family, the capacitors are not as big as possible. I assume 1977 TI wasn´t sure how big they needed the capacitors to get stable operation and because of that they made the possible capacitor area bigger... 


https://www.richis-lab.de/Opamp14.htm


 

[magic]

Damn, thank you, man. I now know what was inside my fake AD797 from AliExpress

I couldn't figure out how those input transistors work. JFET

But wait, is this a genuine LF355?

[Noopy]

At least it was an opamp!

[magic]

Did you get this chip from eBay?

There is no LF-anything in the 1984 Texas Instruments linear databook. There is a lot of second source LM parts, second sources of half a dozen other manufacturers, and there are TI's own TL07x JFET opamps and many other TL and TLC devices, but I can't see a single National JFET chip of any sort.

Their current datasheet SNOSBH0D dates to year 2000, FWIW.

[Noopy]

Hm, you are right, that´s strange...

I got it from Ebay and the printing is very modern and clean. That is somehow suspicious with such an old part.

But it seemed plausible. I have three different National Semiconductor LF355 here. All three have the same design with two different revisions. They look quite similar to the "TI LF355" but are not the same. It seemed quite plausible that both companys had built one.

 

[magic]

Actually my chip is slightly different but I'm pretty sure it's the same LF15x circuit. Maybe one of the faster versions because the capacitors are smaller.

And there are two capacitors, I think the one near pin 4 is the compensation capacitor indicated on the schematic and the one near pin 1 is something else. One plate is connected to ground.

The large JFETs on the left appear to be the input devices, pins 1 and 5 go to another pair of structures which probably are JEFTs and I have no idea what are the transistors in the middle. Perhaps JFET constant current sinks?

I'm not sure why the connections to input JFET gates are swapped. On my chip, the IN+ JFET is near the IN+ pad and the IN- JFET is near the IN- pad.

[Noopy]

The second capacitor is connected to the non-inverting output of the differential stage. There is a more detailed schematic in a National datasheet showing this capacitor.


I now have pictures of the other LF355 (National Semiconductor). First one was built 1982:










The second one was built 1988:






That´s odd:



The 1982-LF355 mask revisions were modified often.



The 1988-LF355 shows only A-revisions. 
There is also a C at the bottom of the die. The older LF355 shows an A. 


Still there is the question whether the Texas-LF355 is a fake or not. 

[David Hess]

Although the LF355 is the slowest opamp of the LFx5x-family, the capacitors are not as big as possible. I assume 1977 TI wasn´t sure how big they needed the capacitors to get stable operation and because of that they made the possible capacitor area bigger... 

I am sure they knew exactly how large to make the capacitors.

JFETs have lower transconductance at the same current than bipolar transistors so a smaller compensation capacitor is required yielding a higher slew rate.  Bipolar parts get the same advantage by using transconductance reduction which is why the 741 compensation capacitor is several times larger than later 741 replacements like the MC1458 which use transconductance reduction for exactly this reason.  Transconductance reduction is also what made the 324 so economical; its compensation capacitor is tiny.

So early JFET parts had an inherent size, and therefor cost, advantage over early bipolar parts because they had smaller compensation capacitors.

[Noopy]

But that doesn't explain why the capacitor area is bigger than actually necessary.
If they had known exactly how big the capacitance had to be, they would have integrated the right size and saved the area as in the National Semiconductor LF355.
I'm pretty sure they didn't use the same die for a bipolar opamp...

[magic]

JFETs have lower transconductance at the same current than bipolar transistors so a smaller compensation capacitor is required yielding a higher slew rate. Bipolar parts get the same advantage by using transconductance reduction

If you mean that LM324 trick of discarding 75% of the input stage current to ground, then no, not exactly the same. Mind that LM324 is the worst opamp in the world in terms of slew rate, besides ultra low power stuff.

IMO "transconductance reduction" is a big misnomer. In practice, it's just reduction of the input stage current as seen by the VAS, with all the usual consequences: lower gain, higher noise, lower slew rate. I don't know what was the supposed advantage of that over simply making these transistors 4x smaller and running them at 25% current. I can guess that maybe they couldn't make them small enough and the additional grounded collector and additional bias were necessary to clear the base of stored charge acceptably fast.

which is why the 741 compensation capacitor is several times larger than later 741 replacements like the MC1458 which use transconductance reduction for exactly this reason.

Not sure if they do, certainly not Raytheon. They conveniently provided schematics of most of their analog parts and specified 25pF on both versions. Their die photographs don't indicate significant difference in capacitor area either. The RC1458 seems more efficiently packed with less wasted space, though, and its die is only 50% larger.

So early JFET parts had an inherent size, and therefor cost, advantage over early bipolar parts because they had smaller compensation capacitors.

Well, for the record, this LF155 die is huge because of all those silly JFETs whose function could be replicated with 5x smaller bipolars But there are much better JFET opamps out there, like TL072, which packs two channels on about the same area IIRC.

I lost my LF155, maybe Noopy could post exact dimensions?

[Noopy]

I lost my LF155, maybe Noopy could post exact dimensions?

~ 1,87mm x 1,06mm

 

[David Hess]

I don't know what was the supposed advantage of that over simply making these transistors 4x smaller and running them at 25% current.

As explained on page 19 of National Semiconductor application note A, reducing the current to reduce the transconductance also reduces phase margin from the mirror pole and the tail pole, so in that case the compensation capacitance must be *increased* to reduce bandwidth maintaining stability.

https://web.ece.ucsb.edu/Faculty/rodwell/Classes/ece2c/resources/an-a.pdf

which is why the 741 compensation capacitor is several times larger than later 741 replacements like the MC1458 which use transconductance reduction for exactly this reason.

Not sure if they do, certainly not Raytheon. They conveniently provided schematics of most of their analog parts and specified 25pF on both versions. Their die photographs don't indicate significant difference in capacitor area either. The RC1458 seems more efficiently packed with less wasted space, though, and its die is only 50% larger.

Schematics are usually simplified to not show the transconductance reduction, including the Raytheon RC1458 datasheet I just checked, and I suspect the 741 schematic and values were used instead.  Could a process difference explain the capacitor area you saw?  What really matters is the difference so a 741 on the same process should be compared.

I thought I saw an MC1458 schematic which showed a much lower value of compensation capacitor but now I cannot find it.  Hmm, maybe I was thinking of what sure looks like the MC1458 schematic shown on page 20 of the application note linked above which indicates 5 picofarads instead of the customary 30 picofarads.

[magic]

So there are three schemes described in this appnote: figure 27, 28a and 28b.

They admit that 28a has a problem with increased noise and 28b may be difficult to fabricate accurately at high "reduction ratio".

Figure 27 could in theory be implemented sneakily in lateral PNP input stages, by increasing parasitic collection by the substrate (which normally is undesirable and efforts are made to prevent it), so you could look at a die and never know that a deliberately introduced, significant substrate collector is there.

But the figure 28 schemes are impossible to realize without additional surface collectors and metal traces hooking them up to the mirror and/or ground. So if a chip exists which uses an "advanced" scheme, we could find it, tear it down and see it. So far I haven't seen anything like that. Not in several 358s, not in the RC4558 from Zeptobars and in Chinese RC4558, not in NJM2068 (a Japanese 4558 on steroids), not in the numerous voltage references posted in "metrology". Nor in this LF155 or TL072, for that matter. If these schemes are used, they probably aren't that very common.

This leaves us with the figure 27 scheme, which is hard to disprove by eyeballing because of the aforementioned possibility of a hidden substrate collector. But we can look at its noise implications. Protest if you think I'm wrong, but I'm quite convinced that noise performance of such input stage is simply equivalent to a normal stage running on n-times reduced bias. I will ignore mirror contribution (imagine that it's sufficiently degenerated or whatever) and look at the LTP.

If I got my math right, transconductance of a mirror loaded LTP equals transconductance of each individual transistor. Noise of an undegenerated BJT happens to be equivalent to the Johnson noise of half its intrinsic emitter resistance (which doesn't have real Johnson noise, obviously), and therefore noise of an LTP conveniently equals the "Johnson" noise of 1/gm. And 1/gm happens to be the reactance of Cc at unity gain frequency, so our math is surprisingly easy.

Take a normal 741 with Cc=25~30pF and GBW=1MHz. That's some 5.5~6kΩ impedance and therefore a hair under 10nV/rtHz LTP noise. Multiply by 1.4 because of the NPN emitter followers and we are at 14nV/rtHz. A real 741 has a hair over 20nV/rtHz IIRC.

Now take the "improved" 741 with 5pF. That's 32kΩ and 22nV/rtHz, even before the 1.4x factor. It simply cannot meet the original spec.

Curiously, Raytheon specifies RC1458 noise similarly to 741, but Motorola's MC1458 density plot shows 40nV/rtHz. Hmm... that puppy may need a teardown.

[Noopy]

A little bit more modern: LF411





Sorry, have no size for this one.




It seems that the upper five testpads are used to adjust the absolute value of the offset while the lower three testpads change the polarity of the value. Interesting...
By the way: That´s an interesting transistor type!




The LF411 has four cross connected JFETs at the input.
Nevertheless the offset of the LF411 is a bit higher (7µV/°C typ) than the LF355 (5µV/°C typ)! 
I assume the higher integration of the LF411 leads to more temperature depending drift.