TL07x, TL08x eBay fakery

[John B]

This is probably a "well duh" moment, but is still worth documenting. It would seem that eBay sourced TL op amps have a radically different architecture to what they're supposed to. So, not even factory rejects, but probably some other op amp with a TL label on it.

This follows on from a comment here: https://www.eevblog.com/forum/beginners/adding-filters-to-class-ab-audio-amp/
that TL op amps shouldn't exhibit crossover distortion. Below is a comparison of an eBay sourced "TL072CP" and "TL082CP", vs an LCSC sourced TL072CP. Each op amp is powered at +/- 5V, +/- 15V, with a 10k load to ground. All configured as a non inverting buffer.


eBay "TL082CP" @ +/- 5V. Note crossover distortion around -1V.




eBay "TL082CP" @ +/- 15V. Note crossover distortion around -1.5V.




eBay "TL072CP" @ +/- 5V. Note crossover distortion around -0.8V.




eBay "TL072CP" @ +/- 15V. Note crossover distortion around -1.6V.




LCSC TL072CP @ +/- 5V. No crossover distortion, but note the distortion/non linearity at the bottom of the sinewave.




LCSC TL072CP @ +/- 15V. All G




eBay "TL082CP" and "TL072CP" on the left, LCSC TL072CP on right. Apart from differences in labelling and package shape, the legs on the LCSC item are made of a stronger metal than the eBay ones, which are softer.

[Audioguru again]

Your oscilloscope or your signal generator is digital with a limited number of steps that cause the steps on the waveform. The top and bottom are supposed to have a gradual curve that the product cannot produce. The sides of the waveform also look torn apart with reduced resolution.

[John B]

Here's a closeup of the crossover distortion on the eBay "TL072CP" +/-15V:

More pics to follow.

[John B]

Here's the lower part of the waveform on the LCSC sourced TL072CP. This experiment isn't operating inside the recommended supply range and common mode input range, so I'm not claiming it's amiss, but I just didn't expect this behaviour. This distortion isn't at the top of the waveform. Yellow is op amp output and green is signal generator.

+/-6V



+/-5.4V



+/-5V



+/-4.8V. Finally into clipping.

[floobydust]

Hard to tell if it's an imposter or a reject of an original part. You can measure quiescent current and see if the output stage is under biased, but slew rate would tell lots too.
There are some generic op-amps out of china that get used for counterfeits.

I would send them to Zeptobars in Moscow, as a fake/genuine comparison, with the beautiful die shots they do.
There is an ST TL072 die-shot and I see no laser trims. OP265 fake was a NE5532 inside.

https://www.eevblog.com/forum/reviews/chinese-fake-op-amps/

[Zero999]

The one with crossover distortion is definitely faulty/fake, but the distortion on the negative side is probably within specification. The worst case output swing is specified at ±12V, with a bipolar 15V supply, with is 3V away from either rail, so the distortion is not surprising. The output stage of an op-amp is normal asymmetrical so the fact clipping occurs on the negative rail first is also unsurprising.
http://www.ti.com/lit/ds/symlink/tl074.pdf

[magic]

The clipping behavior is a bit surprising. I would expect it to flatten and then reverse and shoot to the positive rail as the input approaches negative rail, not to stick towards the negative rail before hard clipping.

It was unity gain?

As for the fakes, sure, all popular opamps on AliBay these days are fakes because the sellers ran out of harvested ones long ago.
More obscure ICs can sometimes be found genuine used

Decap them and see what's inside. Probably some LM358. Check bias currents.

They really don't bother faking those parts accurately. Almost all NE5532 are actually some unidentified Chinese RC4558 (see Zeptobars), nobody cares about wrong polarity of bias current and lower bandwidth. Similarly, I have bought some precision JFET opamps which turned out to be NE5532. The latter totally makes sense, those things are made for the audiophool market for upgrading NE5532s so nobody is going to notice that something is off.

The worst case output swing is specified at ±12V, with a bipolar 15V supply, with is 3V away from either rail, so the distortion is not surprising. The output stage of an op-amp is normal asymmetrical so the fact clipping occurs on the negative rail first is also unsurprising.

No, it is surprising. It has to be fault of the input stage, any funkiness in the output characteristics feedback would correct. Besides, if anything, it would be soft clipping rather than the opposite of it.

It's the input stage running out of CMIR, gain of 2x would fix it and give hard clipping at 1~1.5V above negative rail, by my estimation.

And it still is surprising. Normally what happens is that pulling the inputs to VEE shuts down the input stage and causes the output to jump up, pulled by VAS collector load CCS.

[Zero999]

I think you're right about it being the input stage, but still wouldn't completely rule out the output stage. Negative feedback helps reduce distortion, but it doesn't completely eliminate it, otherwise there wouldn't be that nasty crossover distortion.

Yes, the TL072 does go into phase inversion when the input is taken out of the CMMR, but it wouldn't surprise me if there's an appreciable level of distortion before then. Another way to rule out the output stage is to change it to an inverting configuration.

The ones with crossover distortion might be the LM358. Build a unity gain buffer, with the input connected to a potential divider made from two 10M resistors. If it's the LM358, the output voltage will be a couple of hundred or more mV greater than half the supply voltage.

[David Hess]

The clipping behavior is a bit surprising. I would expect it to flatten and then reverse and shoot to the positive rail as the input approaches negative rail, not to stick towards the negative rail before hard clipping.

Phase inversion is not immediate.  It requires the input common mode range to be grossly violated.

[magic]

Yes, "flatten and then reverse"
That's the behavior I remember from similar RC4558 type BJT input amplifiers.

There is an ST TL072 die-shot and I see no laser trims.

Beauty, I like the simplicity in comparison with some other parts. Seems to match TI datasheet schematic perfectly, by the way. So assuming TI is the same, who has any idea what could be pulling this output towards negative rail?

This one made me laugh:
https://zeptobars.com/en/read/Ti-JFET-TL051CDRA-TL051

Extremely similar to BB OPA134PA.

You don't say

Who else thinks Burr Brown was totally very likely to etch "1995TI" on their die back then?

[John B]

A few more quick observations:

The fake eBay ones draw very little current at idle with no load on the output, <1mA. The real LCSC one draws around 3-4mA, which is consistent with the datasheet. These are rough values off my power supply.

I also tried to determine the bias currents on the fake TL072. I put a 10M resistor on the input and connected it to ground. I measured about 27mV, so also taking into account the multimeter impedance 27mV/5M = ~5nA? But interestingly the current is flowing out of the input. So it doesn't even sound like a JFET input, or an LM358 either.

[magic]

The fake eBay ones draw very little current at idle with no load on the output, <1mA. The real LCSC one draws around 3-4mA, which is consistent with the datasheet.

Beware that idle current is usually specified with the input tied to ground. It may change if the input is left floating, hits one or the other rail and causes shutdown or saturation of some internal stages.

I also tried to determine the bias currents on the fake TL072. I put a 10M resistor on the input and connected it to ground. I measured about 27mV, so also taking into account the multimeter impedance 27mV/5M = ~5nA? But interestingly the current is flowing out of the input. So it doesn't even sound like a JFET input, or an LM358 either.

I measure from output to ground, it's simpler that way. Offset voltage needs to be subtracted.
Current flowing out is consistent with 358, though 5nA is below spec. At any rate, it doesn't seem to be JFET input.

[John B]

The fake eBay ones draw very little current at idle with no load on the output, <1mA. The real LCSC one draws around 3-4mA, which is consistent with the datasheet.

Beware that idle current is usually specified with the input tied to ground. It may change if the input is left floating, hits one or the other rail and causes shutdown or saturation of some internal stages.

Yep, inputs were tied to ground. Even with the above images I made sure that the other opamp was tied to ground in a buffer configuration to ensure that no amp is saturating. I think it was TI app note which stated that one opamp saturating could result in distortion on other amps on the same chip, so I've always kept that in mind as best practice.

[Nominal Animal]

I wonder if it would make sense to create a microcontroller/DSP circuit with a 24-bit ADC+DAC, just to characterize opamps at unity gain with different wave forms down to DC behaviour?

(There are now microcontrollers like Teensy 4.0 with native 480 Mbit/s USB support and I2S interfaces that would allow easy use of audio ADCs and DACs for this.  Say, generate and record 100,000 samples, and do all the processing on a computer.  One channel would be the generated signal as captured by the ADC, and the other the opamp output in unity gain configuration.)

Sounds like a long-term hobby project to me.. 

[David Hess]

Below is a comparison of an eBay sourced "TL072CP" and "TL082CP", vs an LCSC sourced TL072CP. Each op amp is powered at +/- 5V, +/- 15V, with a 10k load to ground. All configured as a non inverting buffer.

It is interesting that the crossover distortion occurs so far below ground;  I would have expected it to be much closer to ground.  The symmetry between the positive and negative edges indicates no capacitive loading which is consistent with the description of your test.

Is the distortion different in inverting mode where there is no common mode voltage?  Maybe it is not coming from the output stage.

[magic]

Also try 5k load and see if crossover artifacts move 50% closer to ground
Outputs pre-biased with 80~150µA to VEE is within possibility, 358 does it too although it's supposed to be 50µA.

[John B]

Also try 5k load and see if crossover artifacts move 50% closer to ground
Outputs pre-biased with 80~150µA to VEE is within possibility, 358 does it too although it's supposed to be 50µA.

Yes it does move the crossover distortion closer to ground. I also put a 910R load on, which moved the distortion just under ground.

Also, the TLOL72 clocks in a whopping 0.55v/us slew rate.

Below is a comparison of an eBay sourced "TL072CP" and "TL082CP", vs an LCSC sourced TL072CP. Each op amp is powered at +/- 5V, +/- 15V, with a 10k load to ground. All configured as a non inverting buffer.

It is interesting that the crossover distortion occurs so far below ground;  I would have expected it to be much closer to ground.  The symmetry between the positive and negative edges indicates no capacitive loading which is consistent with the description of your test.

Is the distortion different in inverting mode where there is no common mode voltage?  Maybe it is not coming from the output stage.

To clarify, you want a unity gain inverting configuration, with the input connected to ground, or left floating?

Link to falstad diagram

[magic]

0.5V/µs is typical of 358. Also a bit under 1mA supply current. I guess it is indeed some kind of 358-like design, maybe with slightly improved specs (less input bias, more output bias).

I wonder if somebody specifically designs those things for faking other parts or if they are just off-the-shelf Chinese jellybeans with changed markings. Probably the latter, but who knows. Fakes of digital chips like DS18B20 or PL2303 certainly need some die customizations even if derived from off-the-shelf MCU designs. Somebody actually invests in it

[David Hess]

I wonder if it would make sense to create a microcontroller/DSP circuit with a 24-bit ADC+DAC, just to characterize opamps at unity gain with different wave forms down to DC behaviour?

It would not take that much resolution to quantify DC behavior.  In the past it was often done using a triangle wave generator and oscilloscope in XY mode and this was sufficient even for precision parts down to picoamps and microvolts.

Outputs pre-biased with 80~150µA to VEE is within possibility, 358 does it too although it's supposed to be 50µA.

The 324/358 are among the few or only with that feature.  Contemporary single supply parts with class-AB emitter follower output stages did not bother and required an external pull down unless their schematics are wrong which makes me wonder if that 50 microamp internal pull-down was patented.

It is interesting that the crossover distortion occurs so far below ground;  I would have expected it to be much closer to ground.  The symmetry between the positive and negative edges indicates no capacitive loading which is consistent with the description of your test.

Is the distortion different in inverting mode where there is no common mode voltage?  Maybe it is not coming from the output stage.

To clarify, you want a unity gain inverting configuration, with the input connected to ground, or left floating?

I mean ground the non-inverting input and drive the input resistor as shown in your example.  Instead of exercising the operational amplifier in a gain of +1 configuration (voltage follower), exercise it in a gain of -1 configuration where the common mode voltage is constant.

0.5V/µs is typical of 358. Also a bit under 1mA supply current. I guess it is indeed some kind of 358-like design, maybe with slightly improved specs (less input bias, more output bias).

I wonder if somebody specifically designs those things for faking other parts or if they are just off-the-shelf Chinese jellybeans with changed markings. Probably the latter, but who knows. Fakes of digital chips like DS18B20 or PL2303 certainly need some die customizations even if derived from off-the-shelf MCU designs. Somebody actually invests in it

More likely they are parts which failed testing or are completely different parts which were remarked.  The TL07x series is the low cost jelly bean part for audio and low input bias current.