Opamps - Die pictures

[Noopy]

I totally agree with you! 

It seems everything has to be switched to digital, even the analog parts...

[Noopy]

The OPA140 from Texas Instruments is a precision opamp with a very low noise figure. The supply voltage may be between 4,5V and 36V. The OPA140 typically draws 1,8mA. The output can drive between 36mA and -30mA. The cut-off frequency is 11MHz. The slew rate is typically specified as 20V/µs. The input current remains below 10pA, over the entire operating temperature range below 3nA. The datasheet specifies +/-30µV as a typical offset voltage with a maximum temperature drift of +/-1µV/°C.




The datasheet contains a simplified circuit diagram of the opamp. A classic differential amplifier with J-FET transistors is located at the inputs. This is followed by the driver and finally a push-pull output stage whose output potential can be driven up to the supply potentials.




A polyimide layer protects the die.




The dimensions of the die are 1,4mm x 0,8mm. The high degree of integration makes it difficult to recognize individual function blocks. The wide lines in the right-hand area show where the output stage is located. In addition to the output, the bondpads for the power supply are also located there, so that the relatively high currents do not have to travel long distances. Three bondpads were only contacted during production. It is possible that the offset voltage was adjusted via these. In any case, no traces of laser adjustment can be seen.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/92x03XL.jpg (12MB)




At one point test structures of the different layers are shown.




The numbers 5225 are shown in the lower right-hand area. This could be an internal project designation. The meaning of the geometry above it remains unclear.


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

 

[iMo]

Are you sure it is the die of the 140? 
After seeing this, it is clear to me the REF80 is just the 4 zeners and 3 opamps (2x output buffers on the left, 1x oven on the right)..

[Noopy]

You can build Opamps quite complex. 
Anyway, the OPA140 has some nice specs. 

[magic]

Damn, this is pretty hardcore. I hoped it would be more like LM4562.
Is it three layers of metal?

I wonder if the metal in the center is just capacitors, or if it covers the input JFETs.
I have a feeling that nothing else on this die looks like an input stage.

Are you sure it is the die of the 140?

Good question. There is no TI logo, but it's definitely a modern chip and I'm not sure who else is making similar opamps besides TI.
It surely isn't a relabeled 358.

[iMo]

The finest older opamps looked like a baroque fine art, hand crafted with upmost care and love, having its personality and "soul"..
This modern one - like an oil refinery blueprints..
Their layouters are banging their off the shelf silicon libraries from left to right and the autorouter finishes it up..

[David Hess]

Anyway, the OPA140 has some nice specs. 

The OPA140 is an outstanding replacement for many previous precision low input bias current operational amplifiers.  The only place it is not an improvement is input offset voltage and input offset voltage drift in comparison to parts like the LT1012, but these are still very good.

[Noopy]

Is it three layers of metal?

I can see three... Yes, I assume that are three metal layers.

I wonder if the metal in the center is just capacitors, or if it covers the input JFETs.
I have a feeling that nothing else on this die looks like an input stage.

I´m not sure. The placement of the vias is interesting. That indicates an input stage I would say.
In the left lower corner there are big structures too. And at the edge there is a big resistor line. Perhaps that´s the current mirror. For the input stage it seems a little too of center.

Are you sure it is the die of the 140?

Good question. There is no TI logo, but it's definitely a modern chip and I'm not sure who else is making similar opamps besides TI.
It surely isn't a relabeled 358.

I have a OPA2140 that looks quite similar (coming soon). There are the common dummy structures TI uses. And for a fake I would use a cheaper opamp.
But you are right there is no TI logo. On the other hand if I remember correctly I have seen TI designs without the TI logo.
I´m pretty sure it is a OPA140.

The finest older opamps looked like a baroque fine art, hand crafted with upmost care and love, having its personality and "soul"..
This modern one - like an oil refinery blueprints..

That is absolutely correct! 

[David Hess]

When I looked into it, the only thing I really found out is that the OPA140 is built on a dielectrically isolated complementary BiFET process, and I am not sure where I saw that.  I think it was in a Texas Instruments selection guide.

[magic]

It has to be this or something similar to it.



BiCom3HV was supposedly used for OPA211, OPA207 and OPA827, so OPA140 is quite possible too. Thin dielectric isolation is used to pack the transistors at maximum density, it looks like this chips could fit about 20 transistors on an area of a bondpad, which used to be only 4 in LM4562 and 1 in LM358.

Space is further saved by routing the metal in several layers over the transistors rather than in gaps between them. This unfortunately doesn't make it easy to "read" the chip. They say it's 1μ technology with three layers, which I suppose costs peanuts these days considering that cheap 8 bit MCUs are built similarly.

So that's how you have your high performance opamp in SOT23. And SiGe transistors, which have improved bandwidth to power ratio, so it's low power too. Looking at the SOT23 pinout, it appears that the die is mounted upside down in this package. The bottom pad on Noopy's image is V- and the top right is V+, you can figure out the rest.

[magic]

The coating is polybenzoxazole rather than polyimide, whatever it means.

And the metal in the center is definitely the input stage. If the input stage works like in OPA627 (source followers, shared drain) then the vertical columns are:
1. source of J2
2. gate of J2
3. drain of J1 and J2
4. gate of J1
5. source of J1
repeated 3 times,

where J1 is the JFET at the top of the left column.

Similar connections in the right column.

The four smaller transistors on the right are also a common centroid pair. Maybe it is indeed a similar input stage to OPA627.

[Noopy]

The coating is polybenzoxazole rather than polyimide, whatever it means.

Polybenzoxazole seems to be very "similar" to polyimide:

https://onlinelibrary.wiley.com/doi/abs/10.1002/pol.20230659

 

[David Hess]

BiCom3HV was supposedly used for OPA211, OPA207 and OPA827, so OPA140 is quite possible too. Thin dielectric isolation is used to pack the transistors at maximum density, it looks like this chips could fit about 20 transistors on an area of a bondpad, which used to be only 4 in LM4562 and 1 in LM358.

What I have about the OP140's process is shown below where the selection guide (slyp795.pdf) discusses a "Difet" process, which I remember from Burr-Brown.  Other sources say it is also a complementary bipolar process which is what I would expect.

Space is further saved by routing the metal in several layers over the transistors rather than in gaps between them. This unfortunately doesn't make it easy to "read" the chip. They say it's 1μ technology with three layers, which I suppose costs peanuts these days considering that cheap 8 bit MCUs are built similarly.

I thought cheap microcontrollers were at 45 nanometers because until recently that was the cheapest process that supports Flash memory.

[Haenk]

The numbers 5225 are shown in the lower right-hand area. This could be an internal project designation. The meaning of the geometry above it remains unclear.

A crossed-out 5235 maybe (i.e. switching between two designs by exchanging one layer mask)? Just an idea...

[D Straney]

Glad to see some analysis of this one, I've always been curious about the modern "high-transistor-count-looking" op-amps, like these two:
https://zeptobars.com/en/read/Microchip-MCP6V27-opamp-auto-zero-input-offset
https://zeptobars.com/en/read/Microchip-MCP6024-CMOS-R2R-opamp-trimmed
...and exactly what all those extra transistors are used for? (besides the obvious answer of the auto-zero function, in the first link)

Makes sense that at least some of the apparent visual complexity is just because multiple metal layers isn't a special high-cost thing now, so the economic tradeoffs point towards more metal layers & smaller die size, even if transistor count is exactly the same.

[Noopy]

The numbers 5225 are shown in the lower right-hand area. This could be an internal project designation. The meaning of the geometry above it remains unclear.

A crossed-out 5235 maybe (i.e. switching between two designs by exchanging one layer mask)? Just an idea...

That´s an interesting interpretation! You could be right. 

...and exactly what all those extra transistors are used for? (besides the obvious answer of the auto-zero function, in the first link)

Makes sense that at least some of the apparent visual complexity is just because multiple metal layers isn't a special high-cost thing now, so the economic tradeoffs point towards more metal layers & smaller die size, even if transistor count is exactly the same.

Hard to say... 

[magic]

What I have about the OP140's process is shown below where the selection guide (slyp795.pdf) discusses a "Difet" process, which I remember from Burr-Brown.  Other sources say it is also a complementary bipolar process which is what I would expect.

Yeah, an old B-B trademark. TI loves those for marketing wank. But goes without saying, this new chip doesn't look much like OPA627

I found that BiCom3HV is the keyword to search for whatever limited information is available about modern TI bipolar tech.

I thought cheap microcontrollers were at 45 nanometers because until recently that was the cheapest process that supports Flash memory.

Well, maybe. Point was, making a few metal layers at 1μ is no big deal. Unfortunately for us in this thread, I guess.

Glad to see some analysis of this one, I've always been curious about the modern "high-transistor-count-looking" op-amps, like these two:
https://zeptobars.com/en/read/Microchip-MCP6V27-opamp-auto-zero-input-offset
https://zeptobars.com/en/read/Microchip-MCP6024-CMOS-R2R-opamp-trimmed
...and exactly what all those extra transistors are used for? (besides the obvious answer of the auto-zero function, in the first link)

Makes sense that at least some of the apparent visual complexity is just because multiple metal layers isn't a special high-cost thing now, so the economic tradeoffs point towards more metal layers & smaller die size, even if transistor count is exactly the same.

For MCP6024, I would bet on the top half being some digital offset adjustment and only the lower half actual opamps. The arrays of small squares are likely common centroid input pairs similar to LMC662, one P-ch and one N-ch for each opamp. This looks like a much simpler circuit, with only two layers of metal I think, but the image is too low resolution to see what's going on. CMOS is also generally harder than bipolar, at least IMO.

[David Hess]

What I have about the OP140's process is shown below where the selection guide (slyp795.pdf) discusses a "Difet" process, which I remember from Burr-Brown.  Other sources say it is also a complementary bipolar process which is what I would expect.

Yeah, an old B-B trademark. TI loves those for marketing wank. But goes without saying, this new chip doesn't look much like OPA627

I included the OPA627 when I was looking for replacements for the old AD542 precision JFET operational amplifier, used as a high impedance buffer in early digital multimeters.  The thing which struck me most about the OPA627 is its high cost at $35 compared to $3.50 for an OPA140 which is better in every respect except speed.

[Noopy]

Now let´s take a look into the OPA2140.






The dimensions of the die are 1,5mm x 1,5mm. It is therefore approximately twice as large as the die of the OPA140.

High resolution: https://www.richis-lab.de/images/Opamp/93x03XL.jpg (22MB)




A direct comparison shows that the individual opamps of the OPA2140 are very similar to the OPA140. Only the shifting of the supply connections resulted in minor changes.




Here you can see the familiar test structures. In the upper area, the contacts to the input transistors can be seen in more detail.




On the OPA2140 the designation 5226 is shown. On the left, the number 5236 could be crossed out three times. The letter B could indicate a second revision.


https://www.richis-lab.de/Opamp89.htm

 

[Hydron]

Ooh thanks, I was literally looking at the AD542 - OPA140 input buffer swap that was just mentioned by David, great to see a die shot.

OPA145 would also be an interesting one to compare to - not quite as good specs as the 140, but close in many ways and half the speed/price (lower speed might be a good thing in some cases).

[Noopy]

I have a OPA145 here but not opened yet. I will bring it forward in the queue. 

[Kleinstein]

Chances are the OPA145 will look very much like the OPA140, just with slightly lower current used for the output stage and the bit slower compensation.

The OPA145 also has lower supply current and thus less heat, which can help with thermal effects. On the downside expect more cross over error from the output stage.

[magic]

no traces of laser adjustment can be seen

I'm not entirely sure about it. I think it was lasered.

Similar marks are visible on the 2140 too, but in different places and in different places on different channels.

[Noopy]

Interesting! You could be right. That's a strange structure. Thin lines but wide area laser marks... 

[David Hess]

Ooh thanks, I was literally looking at the AD542 - OPA140 input buffer swap that was just mentioned by David, great to see a die shot.

There are a bunch of acceptable replacements for the AD542, which I think was the first really precision integrated JFET input operational amplifier with a CMRR of 76dB minimum.  The primary limitations in a follower circuit for a high impedance input are flicker noise, common mode rejection, input bias current, and input offset drift.  If present, the automatic zero loop relaxes the input offset drift requirement, but may require reasonable speed and settling time.  The OPA140 is about the best part available, especially considering its relatively low cost.  The OPA1641 audio version is almost as good for an even lower cost.

OPA145 would also be an interesting one to compare to - not quite as good specs as the 140, but close in many ways and half the speed/price (lower speed might be a good thing in some cases).

As far as I can tell, there are 4 related parts:

OPA140   Precision, low noise, and lowest input bias current, graded and trimmed
OPA141   Raw OPA140 without trimming or grading
OPA145   Low power OPA140, so slower and slightly higher noise, see below
OPA1641  Audio, maybe they grade this one for low noise but nothing else

Chances are the OPA145 will look very much like the OPA140, just with slightly lower current used for the output stage and the bit slower compensation.

The OPA145 also has lower supply current and thus less heat, which can help with thermal effects. On the downside expect more cross over error from the output stage.

The OPA145 also has about half the differential and common mode input capacitance, so I suspect the input JFET pair was scaled in size to keep the same current density.