Opamps - Die pictures

[magic]

I will have to compare this against two Signetics branded NE5532 that I ought to have somewhere. IIRC they looked the same.
Later Philips made a different version which can be seen at Zeptobars. The new die was smaller and didn't have offset/comp pads.

Old NE5532 wouldn't fit into SO8 so the SMD version was wide SO16. Sadly, they didn't bond offset/comp to the extra pins.

Those numbers like "6663A" were different on my two dice but I couldn't see any other difference. I'm not sure if they are die versions or maybe location of the die on the wafer.

The transistor T16 seems to prevent saturation effects. If the voltage at the collector of T17 drops too far, T16 becomes conductive and consequently reduces the modulation of T13 and thus also of T17.

That's also the role of the second emitter of T10 - preventing saturation of T12. In normal operation its Vbe is zero.

In the upper part of the first differential amplifier there is a significant difference to the schematic. Between the collector resistors R1/R2 and the positive supply two series connected diodes are integrated (DR). The purpose of these diodes remains unclear. These diodes are not integrated in the NE5534.

As far as I understand, NE5532 reduces its gain by running the input stage at lower current than NE5534. This allows unity gain stability without increasing compensation caps and die area. Other consequences are lower bias current and worse voltage noise.

These diodes provide additional voltage drop to establish proper bias of the second stage, whose operating current depends on common mode voltage at T8,T9 bases.

The resistor R8 in the start-up circuit is a JFET according to the structures. The NE5532 datasheet from Texas Instruments contains a circuit diagram that is almost completely similar to the circuit diagram from Signetics. However, a JFET is actually drawn in place of the resistor R8.

Also the diodes are included.

[exe]

Thank you very much for the explanation, that's right on time: I was recently researching on decompensated opamps and NE5534 was one of the few. There are even less with external compensation.

I somewhat like NE5532/NE5534. They, however, have quite some input current (200nA typ, I'll measure it and report). It also looks like their open-loop gain isn't that high (onsemi reports 50000, which is a shame  ). But they are cheap. However, it looks like LM4562 is a better op amp, as it offers more slew rate, GBW, open loop gain, input bias current and doesn't cost a fortune (hello, OPA827).

The difference in datasheets amuses me. The one from TI has just three plots plus generic info about opamps. Onsemi's has more plots, but no application information. The one from Signetics is much more useful. Like, it discusses how to make 5534 unity-gain stable. And it's not just about whacking in a 22pf compensation cap, but designing a proper lead-lag compensator to preserve slew rate.

Links to datasheets:
- Signetics: http://www.elektronikjk.com/elementy_czynne/IC/SE5532A.pdf
- Onsemi: https://www.onsemi.com/pdf/datasheet/ne5532-d.pdf
- TI: https://www.ti.com/lit/ds/symlink/ne5532.pdf

Some entertaining reading for sesquipedalian librocubicularists:
- http://nwavguy.blogspot.com/2011/08/op-amp-measurements.html
- https://www.nanovolt.ch/resources/ic_opamps/pdf/opamp_distortion.pdf

[Noopy]

Those numbers like "6663A" were different on my two dice but I couldn't see any other difference. I'm not sure if they are die versions or maybe location of the die on the wafer.

It seems you are right. My old NE555 all show "1000" (https://www.richis-lab.de/555_6.htm). In the first view it looks like it´s a design name. The first few are all named 1000. But the 1973 one and the 1976 one are different!
But what is the number telling us then? I doesn´t look like the location on the wafer to me.

The transistor T16 seems to prevent saturation effects. If the voltage at the collector of T17 drops too far, T16 becomes conductive and consequently reduces the modulation of T13 and thus also of T17.

That's also the role of the second emitter of T10 - preventing saturation of T12. In normal operation its Vbe is zero.

You are right, thanks! 

In the upper part of the first differential amplifier there is a significant difference to the schematic. Between the collector resistors R1/R2 and the positive supply two series connected diodes are integrated (DR). The purpose of these diodes remains unclear. These diodes are not integrated in the NE5534.

As far as I understand, NE5532 reduces its gain by running the input stage at lower current than NE5534. This allows unity gain stability without increasing compensation caps and die area. Other consequences are lower bias current and worse voltage noise.

These diodes provide additional voltage drop to establish proper bias of the second stage, whose operating current depends on common mode voltage at T8,T9 bases.

I agree! Unfortunately I didn´t find an obvious option to reduce the current.

The resistor R8 in the start-up circuit is a JFET according to the structures. The NE5532 datasheet from Texas Instruments contains a circuit diagram that is almost completely similar to the circuit diagram from Signetics. However, a JFET is actually drawn in place of the resistor R8.

Also the diodes are included.

Interesting! It seems I had an old datasheet.
I will add that information.
Thank you for your support! 

[Noopy]

I´m not sure about that capacitor! 

V+ is connected with the buried collector (red). Optical that could be the base too but electrical that would make no sense.

The metal layer is connected with something in the upper area (cyan). Looking at the edges that has to be the collector too. Of course that doesn´t make sense. 

The metal layer is connected to something through "windows". I assume the first window is in the emitter area. I assume the second window is the base layer. The innermost square is the via through the oxide. All in all the metal layer contacts the collector area. 

=> Everything is connected to the collector, nice! 

I´m not sure whether there is a emitter layer but without one the structure doesn´t make more sense. 

Any ideas?

[magic]

I agree! Unfortunately I didn´t find an obvious option to reduce the current.

R3 value should be higher than in NE5534.
It looks like NE5532 and NE5533 were designed together to use the same silicon with different metal. Those resistor strips would be connected differently in NE5533 and the original NE5534 was likely different altogether.

Old Raytheon databooks show resistor values and this resistor is different between RC5534 and RC5532.
They also show no diodes above R1/R2, but I'm pretty sure it's an error and RC5532 had them too.

Also the diodes are included.

Interesting! It seems I had an old datasheet.

I suspect you looked at TI NE5534 datasheet, which has no diodes of course.

I´m not sure about that capacitor! 

On TI from Zeptobars it looks like this:

metal layer: signal, connected through the trace coming from top left
a blue layer under the metal, likely a special dielectric
violet emitter layer: VCC, connected at the first DR diode anode
pink base layer: signal, connected through the "windows", which are holes in dielectric and emitter layers
pink-brown collector layer: VCC, connected through the emitter layer at the first DR diode and maybe in other places

TI NE5532 schematic shows almost exactly this connection, but the collector is shown shorted to the base. That's obviously an error, it's shorted to the emitter.

edit
AFAIK similar capacitors were used in OP27 and derived chips, maybe in OP07 too.
This is Linear's schematic symbol for it:

[magic]

Those numbers like "6663A" were different on my two dice but I couldn't see any other difference. I'm not sure if they are die versions or maybe location of the die on the wafer.

It seems you are right. My old NE555 all show "1000" (https://www.richis-lab.de/555_6.htm). In the first view it looks like it´s a design name. The first few are all named 1000. But the 1973 one and the 1976 one are different!
But what is the number telling us then? I doesn´t look like the location on the wafer to me.

Plot twist: one of my dice is 5663A and I think that yours is also 5663A, not 6663A. The first digit looks different than the two 6 in the middle, the top left corner is sharper.

But my other die is 3992A

[David Hess]

AFAIK similar capacitors were used in OP27 and derived chips, maybe in OP07 too.
This is Linear's schematic symbol for it:

That has come up before.  To me it suggests a "charge storage" transistor structure, although I only know of examples which are PNP.

[Noopy]

R3 value should be higher than in NE5534.
It looks like NE5532 and NE5533 were designed together to use the same silicon with different metal. Those resistor strips would be connected differently in NE5533 and the original NE5534 was likely different altogether.

I agree. R3 of NE5532 should be twice as high as R3 of the NE5534. I didn´t see a 1:2 option but this R3 with its value R+1/3R can be stripped down to 2/3R which is half of the NE5532 value. A little strange that there is no 2R : R option but it seems they needed a R3 with a little more resistance...

I´m not sure about that capacitor! 

On TI from Zeptobars it looks like this:

metal layer: signal, connected through the trace coming from top left
a blue layer under the metal, likely a special dielectric
violet emitter layer: VCC, connected at the first DR diode anode
pink base layer: signal, connected through the "windows", which are holes in dielectric and emitter layers
pink-brown collector layer: VCC, connected through the emitter layer at the first DR diode and maybe in other places

TI NE5532 schematic shows almost exactly this connection, but the collector is shown shorted to the base. That's obviously an error, it's shorted to the emitter.

edit
AFAIK similar capacitors were used in OP27 and derived chips, maybe in OP07 too.
This is Linear's schematic symbol for it:

Ah, there is a thin dielectricum! I missed that, thought there is just the normal oxide layer. Now that makes much more sense.
And now I think the TI schematic is correct (talking about my pictures):
V+ connects the capacitor through the emitter doping. Strange for me but makes sense in the big picture.
With the new informations the "windows" under the metal layer are cut into the dielectricum and in the emitter area. That means the metal layer contacts the base area.
At the upper edge the metal layer contacts the collector. That means base and collector are connected and the pn capacitor is between base and emitter.
 

[sansan]

In the upper part of the first differential amplifier there is a significant difference to the schematic. Between the collector resistors R1/R2 and the positive supply two series connected diodes are integrated (DR). The purpose of these diodes remains unclear. These diodes are not integrated in the NE5534.

Maybe this?

From NJM5532DD datasheet

[magic]

This is unrelated.

The parasitic diodes between Vin and V+ are the BC junction of the input transistor (Q1 or Q2) and the junction between the diffused collector resistor (R1 or R2) and its isolation island. They are shown as PNPs because the P substrate can collect minority carriers injected from the P element into its corresponding N isolation island. The resistor represents substrate resistance. The parasitic NPN's base is the substrate, collector is the Q1/Q2 isolation island and its emitter must be some other nearby isolation island connected to ground through a pair of transistors. My bet is on C2, which is close to the input pair in Signetics, TI and Philips and solidly grounded to the V- pin through BE junctions of the VAS transistors.

The way they draw this parasitic circuit looks like it may be capable of latchup.

[Noopy]

I have some updates for the AMD AM685. The AMD Linear and Interface Data Book contains some interesting information.




The quality of the schematic is a little better and here we have the link between the resistors R4 and R21 that is missing in the datasheet.




The transistors are described in detail in the Linear and Interface Data Book. In the input stage, the transistors have to behave as equally as possible. In addition, a high amplification factor and a high cutoff frequency are necessary.

To get high switching speeds, one has tried to keep the parasitic capacitances and the resistances low. These two goals require partially counterproductive measures. The epi collector layer has been made very thin. This reduces the size of the lateral contact areas with the substrate and thus the collector-substrate capacitance. Furthermore, the epi-layer was only slightly doped, which reduces the base-collector and the collector-substrate capacitances. At the same time unfortunately, this increases the collector resistance. The so-called n+ sinker and the low-lying collector feed line ensure that the collector is connected with the lowest possible resistance. Since this highly doped area is only locally below the active area, it does not increase the collector-substrate capacitance too much.

To reduce the base-collector capacitance, the base region was just lightly doped too. Since this increases the base resistance, two base contacts were integrated to the right and left of the emitter. Highly doped regions within the low-doped active base layer further reduce base resistance.

According to the Linear and Interface Data Book, the emitter is 25,4µm x 6,35µm. The tolerance of these structures is 0,25µm. The size is a compromise between good high frequency characteristics, which require a small emitter, and the need to create transistors with as similar characteristics as possible, which is more difficult with smaller structures due to tolerances.

At the right end, the structure features a Schottky diode created by direct metal contact with the weakly n-doped collector region. Integrating the diode into the transistor reduces the area required for the circuit.

Above the sectional view, transistor Q4 is shown. According to the schematic, this is a normal transistor. The die shows that actually three emitters were integrated. The remaining areas can also be seen clearly. The n+ sinker, which contacts the buried collector supply line, is very wide in the above transistor and thus offers a certain variability in contacting and line routing.




The Linear and Interface Data Book also shows the metal layer of the AM685. You can see that the design has been revised once. The most striking feature is the smiley in the lower right corner.




If you superimpose the printed metal layer and the die available here, you can see that areas within the circuit have been redesigned too.




The more extensive connection of the substrate can be seen very clearly. Even though only the metal layer is shown for the first revision, it can be said that the contacting of the substrate only took place next to the bondpad (cyan). In the present design, the distribution of the V- potential has been made wider and connected to the substrate over a large area at three additional locations (red). This ensures that free charge carriers in the substrate are drained quickly and do not negatively influence the circuit.




An interesting difference from the schematic and КP597CA1 is the absence of diodes D1 and D2. The transistors Q3/Q4 contain the areas and contacts for the diodes, but they are not cross-connected. The metal layer of the first revision shows that there the two diodes were still integrated in the circuit.

The diodes D1/D2 suppress saturation effects and thus accelerate the switching of the comparator. However, in the Linear and Interface Data Book AMD points out that it meant a certain effort to manufacture the Schottky diodes in the necessary quality. After all, in the input stage the two branches have to behave as equally as possible. In addition, the diodes bring parasitic capacitances into the system. Perhaps the transistors were fast enough even out of saturation mode that it was more advantageous to do without the diodes.






Some of the resistors have enlarged contact areas on one side, which make it possible to vary the effective resistance values by moving the vias (red).
In comparison to the first revision of the metal layer, it can be seen that here the resistor R9 has been shorted (blue). This changes the bias of the output stage.


https://www.richis-lab.de/Opamp65.htm

 

[David Hess]

The diodes D1/D2 suppress saturation effects and thus accelerate the switching of the comparator. However, in the Linear and Interface Data Book AMD points out that it meant a certain effort to manufacture the Schottky diodes in the necessary quality. After all, in the input stage the two branches have to behave as equally as possible. In addition, the diodes bring parasitic capacitances into the system. Perhaps the transistors were fast enough even out of saturation mode that it was more advantageous to do without the diodes.

If a diode structure is used, then its leakage (even at zero volts) also reduces the gain of a high impedance output from a differential pair so transistor junctions are used instead, but I do not think that would matter here with a 300 ohm load and even schottky diodes could be used.

It is not usually shown in the schematics, but comparators have schottky diodes used as baker clamps on selected transistors to prevent saturation.  Operational amplifiers lack these and other anti-saturation measures leading to long recovery time.

[Noopy]

You can find hardly any information about the Signetics NE1037. Linear Technology mentions the NE1037 in the "Linear Databook Supplement" from 1988 and refers to the LT1037 as an alternative. As will be shown in a moment, this is indeed an LT1037 that Signetics seems to have adopted. The LT1037 is advertised as a precision opamp that is very fast (60MHz) while offering low noise (4.5nV/√Hz, 10Hz) and low offset voltage (<110µV).






The dimensions of the die are 2,5mm x 2,1mm. It is the typical topology with the input stage integrated at the left edge centered around the horizontal axis. With the output stage centered on the right edge, this guarantees that the power dissipation there affects the two branches of the input stage as symmetrically as possible.

The capacitors take up a lot of surface area on the die. Particularly noticeable is the upper capacitor, of which only a fraction of the area is used. One can assume that this design is used for at least one other opamp, and there a much larger capacitor area is incorporated into the circuit. In Linear Technology's case, this is almost certainly the LT1007, which is listed with the LT1037 on the same datasheet. The LT1037 is stable just at amplification factors greater than 5, but offers a cutoff frequency of 60MHz and a slewrate of 15V/µs. The LT1007 can also be used as a voltage follower, the cutoff frequency is 8MHz and the maximum slewrate is 2,5V/µs.




The bottom edge shows that Linear Technology's design dates back to 1983.




Next to the Linear Technology logo are the numbers 37, which most likely stands for LT1037. The LT1007 simply uses a different metal layer. I don´t know that the I can tell us.






ZE and GE could be abbreviations of the developers. Next to the letters GE, a smiley has made it onto the die.




The two input transistors are doubled and cross-connected so that thermal gradients affect both branches of the opamp as equally as possible. To the right, the following elements are also arranged symmetrically around the center axis.




The structure in the lower left corner is unusual. It seems to be a capacitor like in the NE5532 (https://www.richis-lab.de/Opamp66.htm), which additionally uses the capacitance of the base-emitter junction.

At the lower edge of the picture you can see the diodes which limit the voltage between the inputs of the opamp.




In the upper left corner there are some small resistors which can be configured via several testpads and fuses. They allow to adjust the offset voltage of the opamp during manufacturing. The design of these resistors in the supply of the input amplifier strongly reminds of the LT1012 (https://www.richis-lab.de/Opamp56.htm).




The LT1037 offers two pins to further adjust the offset voltage. This leaves only one bondad, which at first glance cannot be assigned to any function. The detail shows that the bondpad in the upper right corner just supplies the powerstage of the opamp. The bondpad next to it seems to be isolated from it and feeds the positive supply potential to the rest of the circuit.


https://www.richis-lab.de/Opamp67.htm

 

[magic]

Interesting find. The die has LT trademarks all over it but the package and markings are typical Signetics.
I guess Signetics packaged the chip and Linear designed the masks, but not sure who fabbed the die.

LT1007/LT1037 are Linear's improved substitutes for OP27/OP37. The 37 are decompensated, as you noted, hence much smaller capacitors.
GE is probably George Erdi, who originally designed these PMI parts (also OP07) and later moved to Linear and designed chips for them.
His name may also appear on LT1001, which was Linear's answer to OP07.

The capacitor likely uses the same structure previously discussed in the context of NE5534.
Also LT1028/LT1128/LT1115, where it looks visually similar to this one. LT1115 is on Zeptobars.

BTW, I think I previously said that OP27/37 also used such capacitor. I was wrong, your OP27 teardown shows only ordinary MOS caps.

[Noopy]

From this component the labelling has been removed. As we will see, it is a TLC074 from Texas Instruments. The opamp family TLC07x is the successors of the TL07x series. Most of the specifications have been improved. The bandwidth is 10MHz, with a slew rate of up to 16V/µs. Expect a maximum offset voltage of 1mV, typically drifting at 1,2µV/°C. Typical of more modern designs, the supply voltage range has been lowered. While the TL07x opamps allow 7 - 36V, 4,5 - 16V is specified for the TLC07x variants. The outputs can drive more than 50mA.




The TLC07x family offers several variants, including housings with improved thermal conductivity. The TLC074 contains four opamps. The TLC075 variant additionally offers the possibility to switch off the operational amplifiers in pairs.






The dimensions of the die are 2,4mm x 1,4mm. According to the datasheet, it was manufactured with the LBC3 BiCMOS process.

The areas of the four opamps are clearly visible. On closer inspection, one can identify where the bondpads for the supply voltage, the inputs and the outputs are placed. There are two unused bondpads in the right-hand area.






The designation TLC075 is found on the die. That is plausible so far. The same designs are obviously used for the TLC074 and the TLC075, only the shutdown inputs of the TLC074 are not led to the outside. The circuit dates back to 1999, which is consistent with the first revision of the datasheet.




The TLC074 at hand was obviously massively overloaded. This is most clearly visible in the opamp in the upper right corner. In the large, regular structures, two areas are blackened. But all other circuit parts are also discoloured over a large area.


https://www.richis-lab.de/Opamp68.htm

 

[Noopy]

The labelling of this component has been ground off. As will become apparent, it is the dual opamp TLC2262 from Texas Instruments. The datasheet specifies a supply voltage range of +/-2,2V to +/-8V, whereby single supply is also possible. The current consumption is typically 400µA. The TLC2262 offers a rail-to-rail output. Absolute Maximum Ratings specify 50mA as the output current. Typically, the currents are much lower. If one approaches the supply potentials, the maximum possible current also drops sharply.

The relatively low noise of 40nV/√Hz (10Hz) or 12nV/√Hz (1kHz) is emphasised. The offset voltage is typically 300µV with a temperature drift of 2µV/°C. At room temperature, a bias current of 0,5pA can be expected. The low current consumption of the opamp is noticeable in the speed. The bandwidth is only 0,71MHz. The maximum possible slewrate is typically 0,55V/µs.




The datasheet contains a circuit diagram that shows a very clear circuit. The circuit section Q14-Q17 on the far right generates a reference voltage that ensures that Q3 and Q6 behave like current sources. The differential amplifier Q1/Q4 at the input works against the current mirror Q2/Q5.

The output of the differential amplifier directly controls the lowside transistor Q13 at the output. R5/C1 limits the frequency response and thus stabilises the circuit. Q11 represents a current limitation with R1.

The connection of the highsidetransistor Q12 and the transistors Q7-Q10 seems illogical, but it becomes clearer in the following if you look at the circuits of the related opamps.




In addition to the TLC2262, Texas Instruments also offers the TLC2252 and the TLC2272. The TLC2252 typically consumes only 70µA, but also offers a slewrate of just 0,12V/µs and a bandwidth of 0,2MHz. The TLC2272, on the other hand, draws 2,2mA and thus enables a slewrate of 3,6V/µs and a bandwidth of 2,18MHz. The TLC2262 represents the compromise of both types: 400µA / 0,55V/µs / 0,82MHz. With increasing current consumption, the noise of the opamp is also reduced.

The datasheets of all three opamps contain circuit diagrams that only differ in the wiring of the transistors Q9 and Q12. In the TLC2272, transistors Q12/Q13 form a push-pull output stage. Q12 is controlled from the differential amplifier at the input, whose signal is transmitted to the output stage via transistors Q7-Q10. Q9 and Q12 form a current mirror here.

In the TLC2262, transistor Q9 is isolated from the rest of the circuit. Since the control of Q12 lacks any pull-up structure, Q12 behaves like a current source in the TLC2262. Consequently, the output stage operates in class A mode. The maximum possible current change at the output is reduced accordingly, but the current consumption of transistor Q9 is saved.

In the TLC2252, Q9 is integrated into the circuit again. The resistor R6 has been added. This results in the same function as in the TLC2272, but with a reduced current. The combination of the more efficient class B operation of the output stage with the reduced current through Q9 obviously results in a minimal current consumption. Now, however, the highsidetransistor must be charged with the low current of Q9, which significantly reduces the speed of the opamp.

The advantage of these modifications is that one just has to make small changes to the circuit to be able to represent three different opamps.






The dimensions of the die are 2,3mm x 1,6mm. It is manufactured using the so-called "Advanced LinCMOS" process. A more detailed analysis of the circuit is difficult due to the small structure sizes and the two metal layers.

The dichotomy of the die is clearly visible. The inputs with their extensive protective structures can also be clearly identified. The large structures in the middle of the die appear to be the transistors of the input stage. On the right and left edges there are three testpads each, which are clearly used to adjust resistors. Above them, the offset voltage is most likely adjusted via resistors R3/R4.

The function of the central testpad on the upper edge remains open. Since it apparently influences both opamps, it could be that the bias is adjusted over it.




The die is marked with TLC2262. In principle, one cannot be sure in such a case whether it is not just a basic variant that can also be configured as TLC2252 or TLC2272. Most likely, however, it is a TLC2262 and the variants TLC2252 and TLC2272 are represented by variations of the metal layer, which then can be given their own designations. It hardly seems conceivable that the one testpad is sufficient to switch between the three different circuit variants.




The design dates from 1998. In the lower left corner at the output bondpad, the output stage can be seen. The highside transistor is surprisingly small.


https://www.richis-lab.de/Opamp69.htm

 

[magic]

I don't believe this schematic.

The moment Q10 turns on to any degree for any reason, it grounds Q12 gate and then the gate is left charged to ground potential.
Might as well wire the gate to ground permanently.

[Noopy]

I agree with you but assume it is done that way to make it as easy as possible to change between the three opamps.
Just some minor changes around Q9 and you get the one or the other opamp...

[Noopy]

The marking of this component has been ground off (again  ). As will be shown, it is a TLE2072 from Texas Instruments. The TLE207x family is listed under the sounding name "Excalibur Low-Noise High-Speed JFET-Input Operational Amplifier". The datasheet presents them as an upgrade to the TL07x opamps.

The TL2072 achieves a slewrate of typically 40V/µs and a bandwidth of 10MHz. The circuit is rated at supply voltages between +/-5V and +/-19V and then typically draws 2,9mA. Due to the Excalibur manufacturing process, the voltage noise is only 48nV/√Hz (10Hz) and 12nV/√Hz (1kHz) respectively. If this characteristic value is less critical, an opamp from the TL208x series can be used. The offset voltage of the better bin with index A is typically 0,7mV and drifts with 2,4µV/°C. A variant exists that is approved for an operating temperature range of -55°C to 125°C.

The TLE2072 contains two opamps. The TLE2071 and TLE2074 offer one and four operational amplifiers, respectively.




The TLE2072 is a BiFET opamp. It therefore combines JFET transistors with classic bipolar transistors. The schematic in the datasheet looks very complex, but a large part of the circuit is only used to generate constant current sources (blue).

In the center we see the input amplifier (yellow). Interesting are the two JFETs Q15 and Q19. They ensure that an overdrive of an input does not have too negative an effect on the behavior. If the potential at one of the inputs drops, then the associated transistor (Q13 or Q20) allows more current to flow through its path. However, if the potential drops to the point where the junction of the transistor becomes conductive, then current flows across it and a phase reversal occurs at the output of the opamp. Transistors Q15 and Q19 ensure that in such an operating state a similar current also flows from the other branch and the basic function of the differential amplifier is not completely cancelled out.

Transistor Q16 compensates saturation effects that can occur if the power amplifier's overcurrent protection drains current from the differential amplifier. At the bottom of the differential amplifier is a current mirror whose frequency response has been extensively limited with capacitors C2 and C4. At this point, the TLE2071 offers the possibility to externally adjust the offset voltage. Capacitor C6, which limits the frequency response of the complete opamp, is fed back from the output to the input amplifier.

The differential amplifier is followed by the voltage amplifier stage (purple). The voltage amplification is done by Q26. Q22 is a buffer amplifier.

The transistors Q24/Q25 (gray), connected as diodes, generate a certain voltage drop and thus a certain quiescent current through the complementary output amplifier (red). In the case of the highsidetransistor, the overcurrent protection (green) directly sinks the base current. In the case of the lowside transistor, the circuit reduces the output current of the input amplifier.




The datasheet documents the number of components used in the variants with one, two and four amplifiers. The numbers reveal that the two opamps in the TLE2072 share the bias circuit. In the TLE2074, on the other hand, the TLE2072 was merely doubled up.






The dimensions of the die are 2,3 x 2,0mm. The symmetry of the circuit can clearly be seen, as well as the deviating circuit part in the right area, which contains the common bias circuit.




In the datasheet of the TLE208x the metal layers of the opamps are shown. These pictures confirm that the TLE207x and the TLE208x contain the same die and just have been binned.




The design obviously dates from 1994. Some of the masks used in the manufacturing process can be seen on the top edge in the scribe line.




The EX 2072B designation suggests that the circuit is a second revision.




The layout of the circuit parts is rather unusual. The input transistors are located in the center. To the left is the current mirror of the input stage with the large capacitors integrated there. To the right, directly next to the input transistors, the two large current sources are integrated. Further to the right is the output stage.






The input transistors are doubled and cross-connected so that thermal gradients do not have too strong an effect. Directly to the left of the input transistors are two elements that resemble capacitors. In fact, they are the JFETs that become conductive when the potentials at the inputs get too negative.




Two testpads are integrated on the left edge. They allow to adjust the emitter resistors of the current mirror and thus the offset voltage of the input stage via Zener fuses.


https://www.richis-lab.de/Opamp70.htm

 

[Kleinstein]

The TLE207x have a strange effect when the output is driven hard to the negative side. Under this conditions the supply current goes up quite a bit (up to some 20 mA). At least this strange behavior is shown in the data-sheet.
So one should avoid using this type when the neg. side saturation can happen regularly for a longer time.

[exe]

I also noticed a strange thing: on Figure 14 they plot "INPUT BIAS CURRENT vs TOTAL SUPPLY VOLTAGE". I don't get it, after 20V of total supply voltage input bias goes to the roof?

[T3sl4co1l]

A typical JFET effect, hot carriers in the channel increase gate leakage.  A cascode could alleviate that, at some expense of headroom near the negative rail.

Tim

[magic]

Unusually, the schematic shows compensation (C6) being connected to the output stage rather than Q26 collector. This includes those emitter followers in the feedback loop of the second stage. High frequency performance ought to be better, but there is higher risk of instability due to phase shit of the output stage.

The TLE207x have a strange effect when the output is driven hard to the negative side. Under this conditions the supply current goes up quite a bit (up to some 20 mA).

It's not clear if anything seriously limits how much base current Q22 can feed into Q26. In TL072, a diode forward biases and allows Q26 to steal base current from Q22.

[Noopy]

Thanks for all the input, very interesting! 

Unusually, the schematic shows compensation (C6) being connected to the output stage rather than Q26 collector. This includes those emitter followers in the feedback loop of the second stage. High frequency performance ought to be better, but there is higher risk of instability due to phase shit of the output stage.

It´s amazing how small C6 is and how big the other capacitors are.

[Noopy]

Nice!