Opamps - Die pictures

[D Straney]

Ok, it's "mystery op-amp time" again.  I made a pass through this whole thread rather than relying on my fallible memory this time, to make sure this one hadn't already been discussed   (Was going to take a pass through Zeptobars too, but it seems to be down at the moment???)

These come from a Burr-Brown 3650HG (optical) isolation amplifier, which has 3 identical copies of this op-amp as its only active circuitry.  They seem to fall into a mid-range complexity: there's no laser-trimmed resistors or zener-zap adjustments or chopper circuitry, but it's also not a minimal op-amp like the 709.

The input diff pair lives at the top, and I think they're cascoded with the circular transistors below them, if I'm reading it correctly.  There's also some connections to possibly weird-shaped transistors that I'm not sure about: see both ends of the horizontal trace just above the capacitor.

[magic]

NPN emitter followers driving a pair of common base PNPs was state of the art 1960s circuit design.
Collector current of the NPNs mirrored into the bases of the PNPs means it's some 741 variant.
A later one with improved thermal symmetry. I don't know who made it.

[AnalogTodd]

The input diff pair lives at the top, and I think they're cascoded with the circular transistors below them, if I'm reading it correctly.  There's also some connections to possibly weird-shaped transistors that I'm not sure about: see both ends of the horizontal trace just above the capacitor.

Those connections actually look like they are what we used to use to cross under single layer metal on an old bipolar process. There's contact to a light pink diffusion that sits on top of the tub isolation and runs out to the ground (or V-) line. It's usually a N+ layer doped on top of the P-type isolation and since it is at the same potential (ground or V-) there's not a concern of hitting breakdown or worrying about leakage.

[Noopy]

Ok, it's "mystery op-amp time" again.

I have seen these auxiliary structures and this invers name with revision. Somewhere... But I can´t find the part.





The OPA4134 documented here was purchased from AliExpress. At first glance, the package and the marking appear unremarkable. Upon closer inspection, however, the texture of the indentation at pin 1 seems unusually smooth. Added to this is the shape of the letter B in the logo. On the original, the curves of the letter are all the same length.




It is immediately apparent that this is not an OPA4134. The die has a edge length of just 1,1mm.

There are four opamps on the die. The manufacturer is unclear. Only the character strings 20 or 02 and 324 are visible in the center of the die. A nice detail is the + and ground symbols on the power supply bondpads.

This image is also available in higher resolution: https://www.richis-lab.de/images/Opamp/a27x02XL.jpg (14MB)




It is clearly an LM324. That explains the number sequence 324.


https://www.richis-lab.de/OpampA22.htm

 

[Noopy]

One more LM2904! 

This one was purchased from Mouser as the Texas Instruments LM2904N. As already mentioned the “N” refers to the LM2904 variant from National Semiconductor. Texas Instruments acquired National Semiconductor and continues to offer their variant. The device actually still bears the National Semiconductor logo.




The die measures 1,2mm x 0,9mm. The letters “UK” in the upper right corner indicate that it was developed or manufactured in the United Kingdom. The marking on the left edge indicates that this LM2904 is also based on the LM158 design. It appears to be Revision D. The circuit has not changed compared to Revision C (https://www.richis-lab.de/HDD_WD_Caviar_22500.htm#LM358). The geometries of the individual elements differ slightly, and in some cases, their placement is minimally different.

This image is also available in higher resolution: https://www.richis-lab.de/images/Opamp/a28x02XL.jpg (13MB)




In the center of the die is a National Semiconductor logo. To the left and right of it are the revision numbers of five masks.






The circuit corresponds to the circuit in the Texas Instruments LM2904.




Since not everybody has recogniced it in the first place: Transistors Q2 and Q4 divert a large portion of their emitter current to Vcc-. On the die, this is done within one transistor structure. A small collector is connected to the current mirror (yellow). The remainder of the emitter consists of a p-doped region connected to the isolation frame connected to Vcc-. A large portion of the emitter current is dissipated in this region (purple).


https://www.richis-lab.de/OpampA23.htm

 

[magic]

This one was purchased from Mouser as the Texas Instruments LM2904N. As already mentioned the “N” refers to the LM2904 variant from National Semiconductor. Texas Instruments acquired National Semiconductor and continues to offer their variant. The device actually still bears the National Semiconductor logo.

Nice to see that they still haven't replaced it with CMOS
Pin 1 marking style is TI, so they package them together with their own chips now.

Are you sure that UK means UK? National was a US company.
I have also seen UK on LM336.
And I have seen NSUK on LM4562, LM3915.

[Noopy]

Are you sure that UK means UK? National was a US company.
I have also seen UK on LM336.
And I have seen NSUK on LM4562, LM3915.

I'm not 100% sure but National Semiconductor produced quite some analog circuits in UK:
https://www.glassdoor.co.uk/Overview/Working-at-National-Semiconductor-UK-EI_IE36966.11,36.htm

[mkmk!]

Hello,
just for curiosity, i opened a ne5532 from the new ones (2026) and the die looks identical to TL6113 mentioned above. Sorry i don t post pictures, they are not clear.

[Noopy]

The part documented here is manufactured by Fairchild and is designated µ3A770931. The package is known as a flat pack and is typically used in military and aerospace applications.




The designation µ3A770931 or U3A770931 cannot be easily identified. The “Fairchild Semiconductor Integrated Circuit Data Catalog” from 1970 lists a U3F7709311 and a U3F7709312 and explains how the designations are structured. The package family 3 lists several flat packages. However, a 3A variant is not mentioned there. 7709 indicates that it is a µA709. The operating temperature range 31 extends from -55°C to +125°C.






The edge length is 1,4mm. The design differs somewhat from the 1987 design. However, the circuit is essentially the same.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/a29x06XL.jpg (18MB)


https://www.richis-lab.de/OpampA24.htm

 

[exe]

Was the metal layer mask hand-drawn? Wait, the whole chip is probably hand-drawn?

[Noopy]

As far as I know they used a special tape to "draw" the geometries they needed. Back then that was a very manual process.

[AnalogTodd]

As far as I know they used a special tape to "draw" the geometries they needed. Back then that was a very manual process.

Rubylith layout process. VERY tedious, I never had to work with it though I have seen old rubylith stencils in my years of doing IC design and layout. They were kept in case things needed to be updated, at which point they would actually digitize the layout and then do the re-work. I started when we still used color keys to check layouts between release from our layout group and prior to mask generation. Watched us go from there to 'fractured' data, then to just digital JDV (job deck view) on screen.

More info: https://en.wikipedia.org/wiki/Rubylith

[Noopy]

The MUSES02 is a dual opamp optimized for high-end audio applications. It was developed by the Japanese company New Japan Radio. It is now part of Nisshinbo Micro Devices, a joint venture between New Japan Radio and Ricoh Electronic Devices. Digi-Key offers the MUSES02 at a unit price of 56€.

The operating voltage range extends from ±3,5V to ±16V. The typical current consumption is 8mA. The datasheet prominently highlights the noise voltage density of 4,5nV/√Hz and the voltage gain of 110 dB. The maximum voltage rise time is typically 5 V/µs. The bandwidth is specified at 5,8MHz. The manufacturer guarantees isolation with an attenuation of 150dB between the two integrated opamps. Changes in common-mode voltage and supply voltage are typically attenuated by 110dB.




The two opamps are located on two separate dies. This explains the low crosstalk from one side to the other.




The die measures 2,5mm x 1,2mm. The process does not appear to be particularly advanced. The structures are relatively large, and there are no special elements. Large structures are definitely an advantage in the field of precision circuits. For example, larger elements inherently generate less noise.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/a30x03XL.jpg (36MB)




The design appears to date from 2008. The letter B could stand for a second revision, but it could also have a different meaning. The symbol, which resembles the number 7, is often found on components from Nisshinbo Micro Devices. Its meaning remains unclear.




If you take a closer look at the circuit, you’ll notice that a great deal of effort went into the power supply. Both the negative and positive potentials are routed via two bondwires each to different bondpads. In addition, the potentials have been distributed in a star configuration from these bondpads so that the individual circuit blocks interfere with each other as little as possible.






The integrated circuit is not overly complex, but it does have a few unique features. Two PNP transistors, embedded in a classic differential amplifier, serve as the input stage. The traces from the bondpads to the input transistors are designed to be of equal length and thus have equal resistance. The transistor pairs Q1/Q2 and Q4/Q5 have been placed roughly in the center and in a crisscross pattern. Both measures reduce the influence of thermal gradients on the input stage.

Transistor Q6 acts as a buffer stage that reduces the unbalanced load on the differential amplifier caused by the subsequent amplifier stage. Transistor Q7 provides the voltage gain of the opamp. The voltage amplifier stage and the output stage use the second negative potential V2-, so that their dynamic currents do not affect the potential V1-. The voltage amplifier is powered by a constant current supplied via transistor Q11. Since the operating current is constant, the potential V1+ can be used for this purpose.

The Capacitor C2 reduces the effect of the base-collector capacitance of transistor Q7. This so-called Miller capacitance is voltage-dependent and can distort the signal accordingly. (I know most of you know the topic but sometimes there are newbis in the room.) C2 also limits the frequency response of the opamp. The RC network R7/C3 provides additional local negative feedback. Transistor Q8 protects the output stage from overload. If Q7 is driven to the point where its collector potential drops significantly, Q8 turns on and draws base current from Q6, which ultimately reduces the drive level of the chain all the way to Q9.

The Q12/R8/Q13 block generates a specific voltage drop that ensures both output stage transistors always allow a certain amount of quiescent current to flow in the crossover region. This reduces crossover distortion at the signal’s zero-crossing point. The output stage is configured in a complementary configuration. The two paths are combined via two low-impedance resistors.

The control of the two current sources, Q3 and Q11, is unusual. It is not achieved via a common bias potential. Each current source has its own current mirror, which gets its reference current from its own current sink. Clearly, great care was taken to isolate the current sources from one another as much as possible. In contrast, the control of current sources Q16 and Q17 is designed to be relatively simple. The reference voltage on which the currents are based is generated by the two base-emitter paths of transistors Q18 and Q19. Their operating current is set solely via resistor R11. There is a significant influence of voltage and temperature on the voltage drop across transistors Q18 and Q19 and thus on the operating currents.

The MUSES02 contains several variable and optional elements. The operating current of the differential amplifier can be adjusted via the three elements of resistor R1. The capacitance of C1 can also be adjusted via the metal layer. It limits the frequency response of the differential amplifier. Similarly, the frequency response of the opamp can be influenced via the metal surface of capacitors C2 and C3. A second, unused resistor is connected in parallel with R8. Both strips have a widening in the center that could be used as a contact area. This can be used to vary the quiescent current of the output stage. To the right of transistor Q13 are two unused transistors. Perhaps these transistors were intended to allow for a different configuration of the quiescent current setting. The reference currents of the current sources can also be adjusted extensively. Resistor R11 has two optional contacts. Above transistor Q18 is another transistor that could be used to increase the reference voltage by one additional base-emitter voltage drop. Less obvious are the extended contacts on resistors R9 and R10. These can also be used to slightly vary the currents in the differential amplifier and in the voltage amplification stage.




The substrate of the MUSES02 is p doped. It is accordingly connected to the negative supply potential (black). The two elongated contacts have been deliberately placed next to the output stage and next to the voltage amplifier. The transistors and resistors are located in n-type wells and are thus isolated from one another. The differential section of the differential amplifier has an additional, wider frame (yellow). Additional p-doping has apparently been introduced there. A similar frame is found in the bias section. There, the frame is additionally connected to the less-loaded negative supply potential (blue). It remains unclear why the frame of the differential amplifier was not connected in the same way. Nevertheless, the frame has a balancing effect on the local substrate potential.


https://www.richis-lab.de/OpampA25.htm

 

[Noopy]

Dedicated to deviantlogic (https://www.eevblog.com/forum/projects/different-die-pictures/msg6326362/#msg6326362)






The RCA CA3080 is a transconductance amplifier, or OTA (Operational Transconductance Amplifier). Extensive information on the CA3080 can be found in Application Note AN6668. The simplified circuit diagram shown above is also included there. Unlike a conventional opamp, the transconductance amplifier has a current source at its output. The gain can be adjusted via an additional current interface. This interface can be used, among other things, to modulate the gain or to disable the signal path.

The supply voltage may range from ±2V to ±15V. Noteworthy is the high maximum slew rate of 50V/µs. The datasheet specifies the bandwidth as 2MHz. For an opamp, the input resistance is relatively low at typically 26 kΩ.






The CA3080 circuit is not overly complex, but it appears confusing at first glance. The block diagram, which is also shown in Application Note AN6668, is much easier to understand. At the input is a differential amplifier that operates with the controllable current sink W. On the collector side of the differential amplifier is the current mirror Y. However, this current mirror does not serve the differential amplifier exclusively. The current mirror Y and the differential amplifier are cross-connected with the current mirrors Z and X on the output side.

The current mirrors are not immediately recognizable in the circuit, as they are slightly more complex, improved versions. In the case of current mirror X, an additional transistor has been integrated that also serves as the output transistor. In addition Darlington transistors have been used in the upper current mirrors Y and Z. Diodes D2 and D4 increase their switching speed.




The die measures 1,0mm x 1,1mm. 6389 is the internal part number of the device. The circuit corresponds to the schematic.


https://www.richis-lab.de/OpampA26.htm

 

[mawyatt]

The PNP mirrors are Darlington versions of the popular Wilson current mirror, this Darlington use is due to the low Beta of the PNPs, which limits mirror accuracy.

Best

[magic]

It's a bit strange, because Wilson's accuracy comes from the fact that Q3 and Q4 base currents enter the output branch and Q6 diverts one base current back to the input branch. This works when Q6 has the same base current as Q3 and Q4, in which case adding Q5 breaks the scheme and only reduces accuracy.

In practice, Q6 emitter current is higher due to those base currents, so its base current is higher too. Q5 could compensate for that, but only if its β is ridiculously low due to running at low bias (much less than unity) or all of those PNPs have very low β. For example, if they all had equal and constant β independent of collector current, you can calculate this would work for β = 0.618

Maybe it's more important that both differential branches pass through identical mirrors, so their errors somewhat cancel out. Then the limiting factor is matching and NPN mirror accuracy, which has to be much better. But this only fixes offset, some gain error would still remain. One of the selling points of these amplifiers was that their gain can be linearly controlled by varying I_ABC.

[MT4S301]

Found this in intersil application note 6668:

Transistor Q3 and diode D1 (shown in Figure 4) comprise the current mirror “W” of Figure 2. Similarly, transistors Q7, Q8 and Q9 and diode D5 of Figure 4 comprise the generic current mirror “Z” of Figure 2. Darlington-connected transistors are employed in mirrors “Y” and “Z” to reduce the voltage sensitivity of the mirror, by the increase of the mirror output impedance. Transistors Q10, Q11, and diode D6 of Figure 4 comprise the current-mirror “X” of Figure 2. Diodes D2 and D4 are connected across the base-emitter junctions of Q5 and Q8, respectively, to improve the circuit speed. The amplifier output signal is derived from the collectors of the “Z” and “X” current-mirror of Figure 2, providing a push-pull Class A output stage that produces full differential gm. This circuit description applies to both the CA3080 and CA3080A.

I do not understand since how does darlington config increase transistor output resistance? Is there something unique to lateral pnps?

[magic]

That potentially makes sense, I can imagine two mechanisms:

1. Base width modulation in Q6 varies its β and therefore base current; Q5 sends this base current to the output. Probably the main reason for doing it.
2. Base width modulation in Q6 varies its Vbe and therefore Q4 Vce; increasing Vce by 0.6V maybe raises its output resistance by operating further from saturation.

As for mirror error due to sending both Q4 and D3 base currents to the output branch, I guess they just considered it acceptable. Probably reasonably good PNPs were used with β of at least a few tens, so gain error was just a few % and less than thermal drift anyway.

[mawyatt]

Suspect using a Darlington Wilson cell keeps the main PNP current source Vce at ~2 Vbes which helps at higher currents because these PNPs have a high Rc and they also get help with the low Beta.

Best

[Noopy]

Had some problem with the forum the last days ("400 Bad Request"). Deleted cookies, now it seems to work again... 






Unisonic Technologies also sold a CA3080.
It seems the circuit was designed with other circuits in mind. There are quite some unused transistors.


https://www.richis-lab.de/OpampA27.htm

 

[RoGeorge]

Yes, it was some DDoS attack in the last couple of days (and probably still is), now the forum has in place extra measures against bots.

As always, big thank you for all the pics! 

It happens that I've just bought two CA3080 yesterday, made by intersil, never had one of these before. 
Letting a link to the AN6668, just for the docs.  That AN also discusses the choices for the current-mirrors.
https://www.renesas.com/en/document/apn/an6668-applications-ca3080-and-ca3080a-high-performance-operational-transconductance-amplifiers

[Noopy]

The MUSES8820 is a dual opamp optimized for high-end audio applications. Its specifications are identical to those of the MUSES02. The only difference is that the isolation between the two channels is 10dB lower in the MUSES8820. In addition to the SO-8 version documented here, the chip is also available in a DIP-8 package. At €5,13 for a purchase of 2.000 units, the MUSES8820 is significantly less expensive than the MUSES02.




It turns out that the circuit corresponds exactly to the circuit in the MUSES02. In the MUSES8820, the two opamps have simply been integrated onto a single die. The die measures 2,4mm x 2,3mm. This is slightly less silicon area than the MUSES02 occupies in total. In addition to the familiar multiple power supply connections, a strip of the metal layer has been inserted between the two opamps to serve as a shield. This strip was connected via its own bondwire and, given its location, is most likely connected to the negative power supply potential.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/a33x02XL.jpg (65MB)




The circuit layout of the MUSES8820 (below) corresponds to that of the MUSES02 (above). However, there are a few minor differences. The bondpad in the lower right corner has been moved down slightly (yellow). To achieve this, one of the optional transistors had to be omitted. At first glance, one might think this optimized the bond process. However, it is highly questionable whether the minimal shift of this contact provides any significant advantage.

The optional metal areas for the three capacitors have been omitted (red). The total area has been reduced accordingly. Overall, some components have been repositioned to save a little space. In this context, an optional transistor was also removed in the area of the bias current source (blue), and the substrate interconnects were shortened slightly (black). At the end of the metal layer, however, there would still be a strip available to make contact with the substrate (green).


https://www.richis-lab.de/OpampA28.htm

 

[Noopy]

I took another look at the second die, and that's when I realized that the two opamps aren't quite the same after all. The design was mirrored! That makes bonding easier. However, the effort required for manufacturing and handling the wafers is significantly higher as a result:




https://www.richis-lab.de/OpampA25.htm#update

 

[exe]

The design was mirrored! That makes bonding easier. However, the effort required for manufacturing and handling the wafers is significantly higher as a result

That explains the price). From datasheet:

advanced circuitry and layout, unique material and assembled technology by skilled-craftwork

Well, at least there is some truth in this. I'm still not sure I want to pay for this.

[Noopy]

Well, at least there is some truth in this. I'm still not sure I want to pay for this.

I'd say the circuit is decent, but not outstanding and not worth so much money. 




The MUSES01 is the other dual operational amplifier optimized for the high-end audio sector. Digikey offers the MUSES02 at a unit price of €39 for a purchase of 2,000 units.

The operating voltage range extends from ±9V to ±16V. The typical current consumption is 8,5mA. Unlike the MUSES02, the MUSES01 features a JFET input stage. Accordingly, the noise voltage density is noticeably higher at 9.5nV/√Hz. In contrast, the bias current is nearly three orders of magnitude smaller at 200pA. The maximum slew rate is typically 12V/µs, which is significantly higher than that of the MUSES02. The bandwidth, however, is noticeably lower at 3MHz. The manufacturer guarantees isolation between the two integrated operational amplifiers with an attenuation of 150 dB. Changes in common-mode voltage and supply voltage are typically attenuated by 75dB and 83dB, respectively, which is significantly worse than with the MUSES02.




As with the MUSES02, the two opamps are located on two separate dies. This explains the low crosstalk from one side to the other.




The circuit is coated with a protective polyimide layer.




As with the MUSES02, the design has been mirrored so that the bondpads can be properly connected to their corresponding pins.




The die measures 2,7mm x 1,3mm. Even without a more detailed analysis of the circuit, it is already clear that the sensitive circuit components and the output stage are arranged around a horizontal axis of symmetry.

This image is also available in a higher resolution: https://www.richis-lab.de/images/Opamp/a34x04XL.jpg (43MB)




The chip bears the string “NJM2799,” a type designation typical of Nisshinbo Micro Devices. The string “MLK14” cannot be identified. It could be related to a developer or to the manufacturing process.




Unlike the MUSES02, the MUSES01 has only one bondpad for each supply voltage. Nevertheless, great effort was made to ensure the best possible power supply. Many circuit components use dedicated traces that radiate in a star pattern to the corresponding bondpad. The negative potential is connected to the substrate all the way around the perimeter. The output stage area (on the right) is shielded with an additional contact strip. A similar strip can be found in the lower left area. There, it was likely the protective structures of the inputs that were being shielded.






The input of the MUSES01 consists of a differential amplifier with four JFETs (yellow). These input transistors are cross-connected to minimize the influence of temperature gradients. Protective diodes connecting to the two supply voltages are located directly at the input. The differential amplifier is flanked by the current source Q1 and a special current mirror (gray).

The current mirror is special in several respects. These transistors, too, have been duplicated and cross-connected. Like the input transistors and the output stage, the transistor block is located in a straight line in the center of the die. Transistor Q4 (dark green) supplies the necessary base current for the current mirror. However, it has been expanded with a current sink, so that it behaves similarly to the voltage amplifier stage (light green). This ensures that the differential amplifier is loaded as symmetrically as possible. Capacitor C3 limits the frequency response. Diodes D1–D4 limit the voltage across the resistors. This is likely intended to suppress saturation effects.

The differential amplifier is followed by a voltage amplification stage (light green). Transistor Q3 operates with Q10 as the driver stage. The actual voltage amplification takes place in transistor Q9. The compensation capacitor C2 bypasses the amplifier stage Q7/Q9. Additionally, the frequency response is limited by the RC circuit R1/C1. Transistor Q2 likely serves as overload protection for the output stage, as in the MUSES02.

The voltage amplifier stage also serves as the driver stage for the output stage. Three transistors (purple) generate a defined voltage drop, thereby ensuring a certain quiescent current in the output stage. The output stage itself consists of an NPN and a PNP transistor, each equipped with an emitter resistor.

The bias currents (blue) are based on a reference circuit that is slightly more complex than in the MUSES02. However, here as well, two separate current sinks control two distinct regions. The current source for the output stage driver is controlled independently of the rest of the opamp. In the input stage, the two current sources Q1 and Q13 have been expanded to include transistors Q14–Q16.




In the traces between the input bondpads and the input transistors, there are two short strips that look like cross-connections. In fact, however, a conscious decision was made to use base doping (purple) there, which has a higher resistance than the collector connections (cyan). Apparently, the design intended to introduce a certain amount of resistance into the gate line. This could be a noise-reduction measure, such as the one explicitly advertised for the Texas Instruments LM2904B (https://www.richis-lab.de/Opamp94.htm).


https://www.richis-lab.de/OpampA29.htm