How hard is it a make a transistor opamp like LM358 ? Chances of working ?

[lordvader88]

I don't have any certified brand name transistors (mine could all knock-offs I suppose) I do have Beta testers tho to try and match a few transistors. So I'm wondering how easy/hard it is to get an opamp working with the +20 transistors like in 1/2 of an LM358.



I'd like to try for the fun of it, at least in stages. Or is it to much like trying to balance a pencil on its tip and probably won't work at all

[Vtile]

Here is a more simple one the uA709.

http://www.ee.nsysu.edu.tw/lab/F6027/LM709%20Operational%20Amplifiers.pdf

[Zero999]

It will work, but the voltage and bias current offsets will be worse than with the IC design. Matched pairs could be used for the current mirror and input stage, which should give comparable performance to the IC.

I've done a similar thing myself, for educational purposes, but I used a much simpler design, than any IC op-amp. It worked and produced an OK op-amp, but nothing which could match an IC, purchased at a much lower cost.

[Kleinstein]

An exact copy could be tricky as there are no multi-collector transistors available. One would need to use 2 or 3 matched ones to replace one of those multi emitter transistors.

The main part to look for might be thermal coupling at some current mirrors. It might need extra emitter transistors to get good accuracy.

I don't know about the 358, but the 741 uses quite unusual PNPs with pretty poor parameters (low beta and slow) however with a high permissible base emitter voltage. There may not be a direct replacement discrete transistor available.

Otherwise most of the OPs internal circuits should work - discrete audio amplifiers in the 1970's-1980's where not not that much different.
There should be simpler circuit with maybe 10 transistors.

[glarsson]

You have to redesign the current sources as they are constructed from multi-collector transistors where the sizes of the collectors set the current. See for example Q19. You will not find discrete of-the-shelf  transistors with these characteristics.

[Ian.M]

You aren't going to get there with discrete transistors - they wont have the close thermal coupling required for their characteristics to track accurately so some parts of the circuit will go out of balance even if you start with well-matched parts.  Also I see a lot of multiple collector transistors in that schematic.  If the ratio of junction areas between collectors isn't unity or a small integer, they will be impossible to reproduce using discretes.   You *may* be able to do something with carefully selected monolithic transistor arrays, but if you really want to build a retro OPAMP, why not take a look at the generation of discrete OPAMPs built with silicon BJTs from the early 60's, prior to the launch of Bob Widlar's uA709 in 1965 and the subsequent commercial dominance of the monolithic IC OPAMP?

See https://www.eevblog.com/forum/projects/doing-opamp-educational-stuff-suggestions-wanted/ where I linked to the GAP/R P65 OPAMP (six transistor potted module) schematic and specs.

[Lee Leduc]

Discrete 741 Kit.
https://www.evilmadscientist.com/2014/the-xl741/

There are many sites doing this. Google Images has a lot of examples. Click on an image and visit the page.
https://www.google.com/search?q=discrete+op+amp&rlz=1CAASUA_enUS723US724&tbm=isch&tbo=u&source=univ&sa=X&ved=0ahUKEwj6kMSPh-LYAhVCQKwKHQr9BWAQsAQIRQ&biw=1745&bih=855#imgrc=_

Example: http://www.partsconnexion.com/opamp_burson.html

[danadak]

You copuld always consider fooling arounbd with a vacum tube opamp.


https://www.google.com/search?q=vacuum+tube+op+amp&rlz=1C1CHBF_enUS748US749&tbm=isch&source=iu&ictx=1&fir=ZJ_Caoy6lhS9WM%253A%252CB-sF-xa4NP4hVM%252C_&usg=__A6KM8za4-jpZduNCQBd5-NSJ3J4%3D&sa=X&ved=0ahUKEwjt_f20ieLYAhVJKqwKHQRvC2QQ9QEIMDAB#imgrc=ZJ_Caoy6lhS9WM:


Classic Philbrick design -




Regards, Dana.

[lordvader88]

Ok I'll try an easier/older op-amp, and I never realized what those double collector things are.

I think I have dual triode tubes, but no transformer for them right now. Maybe I can try a JFET setup, I'll have a look in the future.

[danadak]

Datasheet for KW-2 -


http://www.philbrickarchive.org/k2-w_refurbished.pdf


and general archive -


http://www.philbrickarchive.org/



Note you could try it with some of the lower voltage tubes, the space charge tubes.


http://www.junkbox.com/electronics/lowvoltagetubes.shtml



Regards, Dana,

[Zero999]

Here's a design, similar to the one I used. I say similar, because I drew it from memory. Later, I'll update this post with a better description.

[T3sl4co1l]

You have three problems:
- Matching.  Thermally, and by geometry.  You can select transistors with same Vbe at the same current (not hFE, that basically doesn't matter -- in fact all the PNPs in that design were hFE of 5-10), but you can't guarantee those Vbe's will stay matched as different transistors heat up due to current flow.
- Geometry.  Many of those transistors had different sizes.  A typical example is a bandgap voltage reference (which wasn't used here, I think), where a ratio of transistor is used to create a stable voltage output.  Now, the application here is mainly about having transistors sized appropriately for their place in the amplifier: you use small widths for the input transistors, and small bias currents; and larger ones towards the output.  This isn't as big of a deal, because you can operate a given transistor at much lower currents than its size suggests (e.g., a 2N4401 is rated for 600mA, but it works just fine at 10uA), albeit with way more capacitance than you would have otherwise (which slows down the amplifier a whole lot).
- Ratings.  Most of those transistors are pretty normal (the NPNs are something like Vceo = 40V, Vebo = 7V, hFE ~ 100), but the PNPs are actually an ingenious hack.  On the upside, Vebo = Vceo, which means using PNPs for the input stage gets you the full input voltage range (cool!).  In short, 2N3906s would only give you a ~15V input range on this circuit, not the full 30 or 40V.

For sure, there's a lot that can be simplified, at the expense of supply voltage range, distortion, and such.  The typical discrete op-amp circuit has:
- Input diff pair: 2 transistors
- Volt amp stage (VAS): 1 transistor
- Follower: 2 transistors (diode biased, no current limiting)
This can be enhanced with current sources in various places:
- Input diff pair "tail": increases gain and input voltage range, reduces DC offset
- Diff pair collector load (current mirror): increases gain significantly
- VAS load: increases gain, output voltage range, output bias stability

Further, there are options for biasing (Vbe multiplier to set output stage bias to something other than a multiple of Vbe's from stacking diodes), current limiting, and alternative topologies to improve speed, input and output voltage ranges, and so on.

Spotting which options they used in the above circuit is a fun exercise for the student...

Tim

[rstofer]

Maybe w2aew's video on the long tail pair differential amplifier will help:

[danadak]

Sources of matched transistors -


https://www.digikey.com/catalog/en/partgroup/npn-transistor-arrays/15434?utm_adgroup=General&gclid=EAIaIQobChMIg_el5P_i2AIVwouzCh1PGQvpEAAYASAAEgJ7TfD_BwE


https://www.diodes.com/products/discrete/bipolar-transistors/matched-pairs/


http://www.ti.com/lit/an/snoa626b/snoa626b.pdf


And google "current mirror ic", some solutions as well.


Regards, Dana.

[Zero999]

As promised, here's an overview of the components used in my circuit.



Q1 & Q2: both emitter followers, forming the output stage. R1 and R2 improve the thermal stability and prevent thermal runaway.

Q3 & Q4: Current mirror. This converts the differential output of the long tailed pair (see below) to a single ended output. It acts as an inverting amplifier. When Q5's collector current increases,  Q4 will turn on more, pulling its collector voltage up.

Q5 & Q6: long tailed pair. This was described in the video, previously, so there's not much point in going into to much detail. Obviously the transistors need to be fairly well  V_BE matched. A matched, high Hfe is good, as it keeps bias currents balanced and low.
https://www.eevblog.com/forum/beginners/how-hard-is-it-a-make-a-transistor-opamp-like-lm358-chances-of-working/msg1403984/#msg1403984

Q7, D1 & D2 and R3 & R5 form a constant current sink for Q5 and Q6. According to the simulation, the current is about 55µA.

Q8, D1 & D2 and R4 & R5 form a 630µA constant current sink for Q9, a common emitter amplifier.

Q10 is an emitter follower, which buffers the output of the input stage (Q3 to Q6), allowing it to drive the common emitter amplifier: Q9.

D3 to D5 produce a voltage drop of about 1.9V, which bias Q1 and Q2 on. The simulation says the current through R1 and R2 is 2.65mA, which makes sense because the base-emitter drop of Q1 and Q2 will be a bit higher than the two of D3 to D5, leaving 0.53V across R1 and R2. Try removing one or two of D3 and D5, and see the current consumption go down, but crossover distortion increase.

C1 is a phase compensation capacitor. Try changing its value. Higher values will slow it down, reducing the bandwidth and slew rate, but make it more stable and less likely to ring and oscillate. Lower values will speed it up, but make ringing and oscillation more likely.

I've just realised I've used different transistors in the simulation, to the ones I used to build the real circuit. I think I used BC548C for the NPNs and BC558 for PNPs.

[Kleinstein]

The circuit Hero999 showed is the basic discrete amplifier. It would profit from a base to emitter resistor at Q9 to give Q10 a little more current to work with. The shown output stage with 3 diodes and rather large resistors R1,R2 is one option with quite some standing current for low distortion. It would be also possible to use only 2 diode and smaller values for R1,R2. The alternative would be an VBE multiplier set to a voltage in between.

There is headroom for small resistors (up to a value close to R3) at the emitters of Q3 and Q4 to reduce the demand in coupling / matching  these two.

The circuit is limited to a differential input voltage of about 5 V. So one might consider protection if needed.

[paulca]

I did similar for a 555 Timer.  I got it to work, took me about 4 multi hour sessions though.  It was only about 18 transistors.  Quite a lot of fun.

[IanMacdonald]

Many of the techniques used in ICs are not applicable to discrete circuits. Although the reverse is true as well, and that is part of the reason why the ICs are so insanely complex. The minimalist opamp is just three transistors (Long tailed pair and class A amp) and works reasonably well so long as you don't need the inputs or output to go too close to the supply rails. 

Increase that to 5 and you can give it much more current driving ability. (You can use a complementary power stage, which can't easily be done in an IC)

Other major improvement is a couple for current sources in place of feed resistors - Works better when inputs are near to supply rails. Plus, a Vbe multiplier for more accurate output stage bias. Then, maybe make your class A amp cascode for better HF performance.

Think that makes ten in all for a fairly kickass design. Still a LOT simpler than the IC variant with its need for convoluted designs to avoid PNP transistors, etc.

[Audioguru]

Why did you choose the lousy old LM358 dual opamp to copy? It has almost the worst spec's of any opamp because it is "low power" then it produces noise and crossover distortion and has a poor high frequency slew rate. One of its features is that its inputs work all the way down to its negative supply so it does not need a negative supply.

Whenever I see a circuit using it I use an MC3317x low power opamp instead that has the same input voltages but no crossover distortion and a slew rate to 35kHz.

[David Hess]

Many of the techniques used in ICs are not applicable to discrete circuits. Although the reverse is true as well, and that is part of the reason why the ICs are so insanely complex.

Exactly.

A discrete implementation of an integrated operational amplifier will take advantage of discrete design elements instead of integrated ones.  The big differences are availability of matched transistors, precision resistors, and weird circuit elements like super gain and effectively chopper bipolar transistors which have a high base-emitter breakdown voltage.

Discrete current mirrors which rely on good resistor tolerance replace integrated current mirrors which rely on transistor matching and emitter ratios.

Low offset voltage differential pairs will require monolithic pairs or arrays from suppliers like Linear Systems and they also have super beta parts.  Analog Devices appears to be discontinuing their matched transistor products.  The last time I checked, Diodes Inc. and NXP had some surface mount matched pairs also.

Some "chopper" transistors are made which could be used in a high differential input voltage design but none of them are matched monolithic pairs.  Alternatively a pair of diode connected monolithic transistors could be used to replace the PNP cascodes in a 741 or 301A design or for a single supply PNP input design; I have seen some discrete designs which do this.

Discrete input bias current cancellation in a discrete design can take advantage of high value precision resistors.

Discrete designs have some advantages.  The lack of thermal feedback from the output stage to the input stage increases precision but you can get most of this advantage in an integrated part by limiting the output loading.

Why did you choose the lousy old LM358 dual opamp to copy? It has almost the worst spec's of any opamp because it is "low power" then it produces noise and crossover distortion and has a poor high frequency slew rate. One of its features is that its inputs work all the way down to its negative supply so it does not need a negative supply.

The difficult part of the 324/358 to copy is the high differential input voltage range without chopper type transistors.  The crossover distortion issue can be solved by replacing the class-b output stage with a class-ab output stage like Fairchild did in their 324/358 equivalents.

If you took a 301A and swapped all of the transistor types, then it becomes a single supply input operational amplifier.

[Lee Leduc]

I did similar for a 555 Timer.  I got it to work, took me about 4 multi hour sessions though.  It was only about 18 transistors.  Quite a lot of fun.

Discrete 555 Timer kit from Evil Mad Scientist Laboratories is a faithful and functional transistor-scale replica of the classic NE555

https://shop.evilmadscientist.com/productsmenu/652

[paulca]

I did similar for a 555 Timer.  I got it to work, took me about 4 multi hour sessions though.  It was only about 18 transistors.  Quite a lot of fun.

Discrete 555 Timer kit from Evil Mad Scientist Laboratories is a faithful and functional transistor-scale replica of the classic NE555

https://shop.evilmadscientist.com/productsmenu/652

Seen Dave building it.  Mine was copying an instructable.  I used 2n3904s and 2n3906s and it took up two breadboards.

I then simplified it down, replacing the comparators with actual comparators and then the latch with a set of NOR gates, leaving only the discharge and output driver transistors.

[Zero999]

The circuit Hero999 showed is the basic discrete amplifier. It would profit from a base to emitter resistor at Q9 to give Q10 a little more current to work with.

Why do recommend a base-emitter resistor on Q9? The only reason I can think of is speed.  The downside is it will increase Q10's base current, which will load the previous stage down more, resulting in a reduction in gain and a greater current imbalance in the differential pair, which will increase the offset, voltage and bias, current slightly.

The shown output stage with 3 diodes and rather large resistors R1,R2 is one option with quite some standing current for low distortion. It would be also possible to use only 2 diode and smaller values for R1,R2. The alternative would be an VBE multiplier set to a voltage in between.

I agree, especially about the V_BE multiplier, although I wanted to limit it to 10 transistors. Two diodes would give some crossover distortion, as I've found V_BE is higher, than the voltage drop of a diode, such as the 1N4148, but it would probably be acceptable and certainly better than the LM358.

There is headroom for small resistors (up to a value close to R3) at the emitters of Q3 and Q4 to reduce the demand in coupling / matching  these two.

Yes, some emitter resistors on Q6 and Q6 may also help too.

The circuit is limited to a differential input voltage of about 5 V. So one might consider protection if needed.

Yes, series or back-to-back diodes would do that.

The output could also benefit from over-current protection.

[jmelson]

I don't have any certified brand name transistors (mine could all knock-offs I suppose) I do have Beta testers tho to try and match a few transistors. So I'm wondering how easy/hard it is to get an opamp working with the +20 transistors like in 1/2 of an LM358.



I'd like to try for the fun of it, at least in stages. Or is it to much like trying to balance a pencil on its tip and probably won't work at all

Where are you going to get those dual- and triple-collector transistors?  Look at Q19 in your schematic.

Jon

[T3sl4co1l]

Where are you going to get those dual- and triple-collector transistors?  Look at Q19 in your schematic.

That's IC shorthand for "B and E wired in parallel".

Similarly if you see multiple emitters (but you'll never see multiple bases), though those can actually be constructed as a single device, which means you get some hFE between emitters (in a half-inverted configuration, so to speak), whereas you get no hFE from discrete devices wired in that way.  This is probably never used as a feature.

Tim

[Zero999]

Yes, it's effectively three transistors with the bases and emitters connected together. In real life it's probably just one transistor, but with multiple collectors.

[David Hess]

Yes, it's effectively three transistors with the bases and emitters connected together. In real life it's probably just one transistor, but with multiple collectors.

But usually schematics do not show the ratio of emitter areas so some reverse engineering to determine the operating point is needed.

[edavid]

Where are you going to get those dual- and triple-collector transistors?  Look at Q19 in your schematic.

That's IC shorthand for "B and E wired in parallel".

Similarly if you see multiple emitters (but you'll never see multiple bases), though those can actually be constructed as a single device, which means you get some hFE between emitters (in a half-inverted configuration, so to speak), whereas you get no hFE from discrete devices wired in that way.  This is probably never used as a feature.

No, they are lateral PNPs, so they really are single devices with multiple collector contacts.  See page 1-23 of Hans Camenzind's book: http://www.designinganalogchips.com/

[T3sl4co1l]

No, they are lateral PNPs, so they really are single devices with multiple collector contacts.  See page 1-23 of Hans Camenzind's book: http://www.designinganalogchips.com/

Well, that's basically what I said about emitters, using a pile of emitters, isn't it?

Tim

[Zero999]

Yes, it's effectively three transistors with the bases and emitters connected together. In real life it's probably just one transistor, but with multiple collectors.

But usually schematics do not show the ratio of emitter areas so some reverse engineering to determine the operating point is needed.

The simplified schematics on the other data sheets are quite helpful.
http://www.st.com/content/ccc/resource/technical/document/datasheet/61/46/87/01/98/ed/44/c5/CD00000464.pdf/files/CD00000464.pdf/jcr:content/translations/en.CD00000464.pdf
http://www.ti.com/lit/ds/symlink/lm358.pdf

[amspire]

A thing that has always annoyed me a little is you see components of cheap IC's that are almost impossible to get as discrete devices. Things like super-beta transistors.

In the case of the LM358, those input PNP's have a reverse base emitter breakdown voltage over 30V. This is nice because it is one of the few opamps that can have to the inputs connected permanently to, say, a 12v or 24v voltage even when the opamp power is off.

I don't remember ever seeing a junction transistor with a breakdown voltage more then 8V. Sure the 30V breakdown transistor probably has much lower gain, but it would still be good to have it available.

[edavid]

A thing that has always annoyed me a little is you see components of cheap IC's that are almost impossible to get as discrete devices. Things like super-beta transistors.

Buy a bag of 2N5961/2N5962/2N5963 while they are still around.

In the case of the LM358, those input PNP's have a reverse base emitter breakdown voltage over 30V. This is nice because it is one of the few opamps that can have to the inputs connected permanently to, say, a 12v or 24v voltage even when the opamp power is off.

I don't remember ever seeing a junction transistor with a breakdown voltage more then 8V. Sure the 30V breakdown transistor probably has much lower gain, but it would still be good to have it available.

You could use a regular PNP in inverted mode.

[amspire]

In the case of the LM358, those input PNP's have a reverse base emitter breakdown voltage over 30V. This is nice because it is one of the few opamps that can have to the inputs connected permanently to, say, a 12v or 24v voltage even when the opamp power is off.

I don't remember ever seeing a junction transistor with a breakdown voltage more then 8V. Sure the 30V breakdown transistor probably has much lower gain, but it would still be good to have it available.

You could use a regular PNP in inverted mode.

I did think of that, but you need 30v breakdown for both junctions.

[David Hess]

I don't remember ever seeing a junction transistor with a breakdown voltage more then 8V. Sure the 30V breakdown transistor probably has much lower gain, but it would still be good to have it available.

I linked an entire list of currently available parts like this earlier but most are surface mount now.  They used to be known as "chopper" transistors but they were also used in audio muting applications.

I don't remember ever seeing a junction transistor with a breakdown voltage more then 8V. Sure the 30V breakdown transistor probably has much lower gain, but it would still be good to have it available.

You could use a regular PNP in inverted mode.

This will only work for low voltage where the transistor would not need to be inverted anyway.  This swaps breakdown of the base-emitter junction at high differential input voltages for breakdown of the base-collector junction at high common mode voltages.  If your operational amplifier is only going to run on a 5 volt supply voltage, then nothing needs to be done.

What will work is adding a diodes in series with the emitters which is likely needed in one form or another anyway for a single supply input but this sacrifices gain and the offset voltage of the diodes add to the input offset voltage.  Matched diodes or matched diode connected transistors will ameliorate the problems with precision.  I have seen integrated and discrete operational amplifier designs which did this.

[David Hess]

Yes, it's effectively three transistors with the bases and emitters connected together. In real life it's probably just one transistor, but with multiple collectors.

But usually schematics do not show the ratio of emitter areas so some reverse engineering to determine the operating point is needed.

The simplified schematics on the other data sheets are quite helpful.

Oh, they are very helpful.  And in the case of the LM324 and many other parts, complete schematics are available.  But even the complete schematics do not usually show emitter ratios.

https://www.onsemi.com/pub/Collateral/LM324-D.PDF

Note that the above example shows the value of the 5pF compensation capacitor with goes along with Gm reduction of the differential input which is also shown; the later allows the former.  Implementing Gm reduction in a discrete copy of the integrated design without compromising performance will be very difficult because now you need *four* matched PNP transistors.

It is trivial to do by adding emitter degeneration resistors however this compromises offset voltage and noise.  Video amplifiers including the old LM318 did this and audio power amplifiers do this for increased slew rate.  Without Gm reduction, the compensation capacitance would be about 30pF making the chip larger and more expensive because the compensation capacitor takes up a lot of space.

[Richard Crowley]

In pro retro audio circles, there are a several discrete component op-amp circuits still quite popular and sourced by multiple vendors.  And some of them available as kits.

[RandallMcRee]

It seems to me that this thread has wandered off from the OP's original intent.

First, Mr. OP you should not be trying to build a discrete opamp using a schematic for an integrated circuit opamp. The strengths of integrated circuits are not the same strengths available to you. Unless you are one of those students who is designing a circuit for some CMOS fab. But, no, this is 'Beginners'.

Therefore, you should simply google "discrete opamp" and follow your nose. There are some good designs out there. There is a sort of competition to make the best discrete opamp using the fewest transistors. Some of them seem to be quite good.

Samuel Groner has some good low-noise designs, for example, e.g. http://www.nanovolt.ch/resources/discrete_opamps/

[David Hess]

In pro retro audio circles, there are a several discrete component op-amp circuits still quite popular and sourced by multiple vendors.  And some of them available as kits.

My preference would be to use an integrated operational amplifier with a discrete output stage and maybe a discrete input stage if necessary.  In precision applications, self heating from the output stage is what ultimately limits performance.

First, Mr. OP you should not be trying to build a discrete opamp using a schematic for an integrated circuit opamp. The strengths of integrated circuits are not the same strengths available to you. Unless you are one of those students who is designing a circuit for some CMOS fab. But, no, this is 'Beginners'.

Therefore, you should simply google "discrete opamp" and follow your nose. There are some good designs out there. There is a sort of competition to make the best discrete opamp using the fewest transistors. Some of them seem to be quite good.

Most discrete audio power amplifiers use the traditional 3 stage operational amplifier topology so a lot can be learned from them.  As they get more complex, they include the discrete equivalents of what more advanced integrated operational amplifiers use.

[Zero999]

In pro retro audio circles, there are a several discrete component op-amp circuits still quite popular and sourced by multiple vendors.  And some of them available as kits.

They're mostly audiophool products, aimed at silly people who blindly replace op-amp ICs in pieces of equipment, in the hope it will magically improve the sound quality, when in reality, it's more likely to bugger it up.

It seems to me that this thread has wandered off from the OP's original intent.

Fair point. Awhile ago I did an LTSpice simulation of the LM358, made with discrete transistors, in the hope of predicting whether the power consumption increases or decreases, when the output saturates to the positive rail. Thread linked below:
https://www.eevblog.com/forum/beginners/unused-opamps-within-a-multi-opamp-package/msg1074284/#msg1074284

I've modified it to work with discrete transistors. Note that this will still not behave exactly like the LM358. The input bias currents will be much lower, as the Hfe of the BC557 is higher than the transistors used inside the IC. The maximum input voltage will also be limited to 10V. If you  need more, add diodes such as the 1N4148, in series with the inputs.

[schmitt trigger]

In these days where a diy project usually means some form of Arduino-based circuit, I applaud the individuals who have the desire to tackle a discrete component project.

So I say: of course give it a try!
In the end, with all the money you will spend, you would have purchased a cutting edge premium opamp and still have some spare change, yet most likely its performance will be lower than the lowliest opamp.

However, the learning process will be priceless.

[David Hess]

In the end, with all the money you will spend, you would have purchased a cutting edge premium opamp and still have some spare change, yet most likely its performance will be lower than the lowliest opamp.

The only places where performance cannot be better or at least close is where matched transistors are required and even there, hybrid and integrated duals, quads, and arrays can make up for this and where construction limits high frequency performance.  The trade offs are a little different but in most cases, better performance is possible for selected characteristics.

One big advantage of a discrete design is that you can add functionality like clamping, compensation, and high voltage operation easily.  Usually though it is better to add discrete parts to an existing integrated operational amplifier.  For instance the input and output stages can be replaced or bootstrapping can be used for high voltage operation or to improve common mode and power supply rejection.

[iampoor]

In pro retro audio circles, there are a several discrete component op-amp circuits still quite popular and sourced by multiple vendors.  And some of them available as kits.

They're mostly audiophool products, aimed at silly people who blindly replace op-amp ICs in pieces of equipment, in the hope it will magically improve the sound quality, when in reality, it's more likely to bugger it up.

In pro audio retro circles (I like that phrase! ) They are used for mic preamps, and other circuits where the non linear characteristics, and sonic effects are desirable. They are also used for driving low impedance loads, or output transformers. The distortion characteristics of some discrete opamps are desirable. This is very different from someone ripping apart a modern piece of gear, and putting in discrete opamps willy-nilly.

@OP Trying building an API 2520, or gar1731 opamp. Very easy circuits to build, and can be built for well under 10$ in parts.

[Wimberleytech]

I built this long ago with the CD4007.  It is a great learning tool.  Vary the bias current and see the effect on open loop gain, GBW, slew rate, etc.  Vary the capacitor to play around with compensation.

I just sketched this out this morning and did not build it to make sure I got all the connections correct.

[Zero999]

How matched are the transistors in the CD4007?

I've read about using them in an op-amp before:
https://wiki.analog.com/university/courses/alm1k/alm-lab-ota
http://sites.bu.edu/engcourses/files/2016/08/mosfet-differential-amplifier.pdf

[David Hess]

How matched are the transistors in the CD4007?

Since they are monolithic and have identical geometries, they are pretty good; I would expect offset voltages of 10s of millivolts which may not seem very good but MOSFETs do not match as well as JFETs which do not match as well as bipolar transistors.  The p-channel MOSFETs do not closely match the n-channel MOSFETs however.

[Wimberleytech]

How matched are the transistors in the CD4007?

Since they are monolithic and have identical geometries, they are pretty good; I would expect offset voltages of 10s of millivolts which may not seem very good but MOSFETs do not match as well as JFETs which do not match as well as bipolar transistors.  The p-channel MOSFETs do not closely match the n-channel MOSFETs however.

Two critical parameters, oxide thickness and carrier mobility, will match well on a common die.  Even with these identical process parameters, matching (at the next level) depends on geometry.  Looking at the metal layer die image, it appears that no special care was taken to match.  The structures are serpentine (not best for matching), and not drawn identically (also adverse to matching).  Obviously no attempt to common centroid (why would they?)...etc.  Since the 4000 series are metal-gate CMOS, they will not give the matching performance of self-aligned Si-gate CMOS.

I designed an integrated pacemaker driver once and the company wanted to see a breadboard working (back when simulations were nascent).  So, I had to actually breadboard the amplifiers etc.  I did it using 4007s.  Wire wrapped no less.

Maybe today, I will do a test and see how well these match.  I suspect that 10s of millivolts is about right--prolly on the low side 10-20 mV though.

Oh, one more thing.  Since a design does not have independent control of W/L ratios, you cannot get an optimal first-order design, so there will be some systematic offset in the 7-transistor diff amp architecture.  With some gain in the diff stage, it is probably swamped by the diff pair mismatch though.

---here is an update on matching
I tested the nch devices using 135uA drain current in a diode configuration (VDS = VGS) and got the following VDS measurements:
1.588
1.600
1.594

12mV maximum spread for ONE (and only one) device.
---

[Wimberleytech]

How matched are the transistors in the CD4007?

I've read about using them in an op-amp before:
https://wiki.analog.com/university/courses/alm1k/alm-lab-ota
http://sites.bu.edu/engcourses/files/2016/08/mosfet-differential-amplifier.pdf

These are good links.  I had seen the first one, but not the second.

[Richard Crowley]

My understanding is that those original (and extinct) matched-pair devices (LM394, etc.) were designed with 50 adjacent monolithic transistors (for each "transistor") all connected in parallel to average-out the characteristics. And putting them within microns of each other on the same silicon substrate makes them as tightly integrated as possible.