CMOS Rail to Rail Op Amps

[WGTP]

2262, 2272, etc.  Do we "know" that they clip more softly than an old standard like the 4558 or LM308 or are we speculating they do?  Are there clipping characteristics similiar to the 4048 or 4069 type chips?  Is this an urban myth?  Thanks for the discussion.   

[gez]

Don't know about those chips, but with the CMOS ones I use clipping isn't soft.  Although the output is CMOS the rest of the chips isn't so clipping in the other stages can be just as harsh as in any op-amp.  Also, as they're multi stage, gain is as high as any other op-amp - a lot of the reason why Logic chips soft clip is because gain is so pitiful (they make poor amps).

Interesting thing happens with CMOS chips in comparator fuzz type circuits if both inputs are biased at same voltage.  The output tries to bias at half supply and actually acts as an amp towards the tail end of notes, rather than as a comparator.  End result is that the 'slope' of the square wave alters with amplitude of input signal so you get some 'timbral' dynamics.

[MartyMart]

2262, 2272, etc.  Do we "know" that they clip more softly than an old standard like the 4558 or LM308 or are we speculating they do?  Are there clipping characteristics similiar to the 4048 or 4069 type chips?  Is this an urban myth?  Thanks for the discussion.   

All I know is that when using the specified TLC2262 in my XXL clone, it made quite a difference from a TL072
in the same slot.
I'm not sure that I'd call it "soft" but perhaps "rounded" !
The XXL is still quite a spikey distortion ... not harsh though ..

.... waffle ... waffle  descriptions of sound ... waffle waffle ... MM !

[Jay Doyle]

To my ears they have a different tone but not a softer one.

I think that there is a fundamental difference between an opamp and a CMOS inverter. The opamp is made to be clean and we are slamming it into the rails for distortion, whereas the CMOS is meant to slam into the rails and we are adapting that to an analog situation trying to make it work like an amplifier, not a switch. I don't think you can compare the two.

Jay Doyle

[Sir H C]

Most rail-to-rail op-amps can get within 10-20 mV of the rails before they can no longer track the input.  As such, that is a very sharp cutoff, so harsh clipping.

[stm]

Most rail-to-rail op-amps can get within 10-20 mV of the rails before they can no longer track the input.  As such, that is a very sharp cutoff, so harsh clipping.

Not necessarily.  Most rail-to-rail opamps have VERY weak current capabilities near either supply rail.  As such, if some loading is placed at the output the saturation and clipping waveforms should round off.  For instance, consider loading the output of such opamp with a 1k resistor in series with a 10uF cap to ground.

[gez]

Most rail-to-rail opamps have VERY weak current capabilities near either supply rail.  As such, if some loading is placed at the output the saturation and clipping waveforms should round off.  For instance, consider loading the output of such opamp with a 1k resistor in series with a 10uF cap to ground.

Interesting!  What does it sound like though?

[puretube]

gimme them inverters...

[Jay Doyle]

For instance, consider loading the output of such opamp with a 1k resistor in series with a 10uF cap to ground.

Have you done this?

How did it sound?

How do you have any signal left to hear what is going on in that situation?

What you mention may be a hazard to the opamp because when you cut power the charge in the cap could rush into the output of the opamp and fry it. I think.

[Sir H C]

For instance, consider loading the output of such opamp with a 1k resistor in series with a 10uF cap to ground.

Have you done this?

How did it sound?

How do you have any signal left to hear what is going on in that situation?

What you mention may be a hazard to the opamp because when you cut power the charge in the cap could rush into the output of the opamp and fry it. I think.

Nah, you have a 1k current limiting resistor and the output devices have inherent in them very large diodes that should be able to handle the current.

[Paul Perry (Frostwave)]

If anyone is trying to make some distortion fx by loading the output of a CMOS rail to rail op amp, here's 2 things: 1. there isn't much output range over which the sagging occurs, and 2. the effect probably won't be symmetrical (that could be GOOD, though!)

[gez]

If anyone is trying to make some distortion fx by loading the output of a CMOS rail to rail op amp, here's 2 things: 1. there isn't much output range over which the sagging occurs, and 2. the effect probably won't be symmetrical (that could be GOOD, though!)

Why wouldn't it be symmetrical Paul?  Not being funny - genuine question - just not sure why that would be.

[WGTP]

Wasn't there another post here about assymetry?

Thanks for the info.  I need to try this alternative method, that I guess could be combined with the other standard clipping stuff.   

[Peter Snowberg]

Sadly, Sir H C's post about the difference in fabrication required for NMOS and PMOS was lost in the move.

Let me see if I can repeat part of it....

Thanks to the difference in function (holes vs. electrons and all that) and the doping required to fabricate both transistor types on the same die, the process favors creating one type of MOS transistor over the other. In CMOS logic, the PMOS devices are generally two to four times larger (physically) than the NMOS ones. The differences in size and performance is going to introduce some asymmetry.

[Sir H C]

And what I forgot to mention, when doing the doping, they do a "threshold shift" which causes the PMOS threshold to go down (up?) while the NMOS goes the other way.  So it is only when they get it just right that they will match.

Also from general IC design stuff, you assume a variance of at least 50% for device strength.  Some smaller geometries seem to do better but that gets offset by other errors.

[stm]

Looking at the TLC2262 schematic you can see the two output transistors are Q12 and Q13.  The lower transisor has a series resistor (R1) which is not in the upper side of the output.  Also, the stages that drive the output transistors are not close to be symmetrical and/or complementary.



In addition, take a look at the values labeled VOL and VOH in the datasheet.  You will see these values are specified for different current levels, depending on the side of the supply you are clipping into.

For instance, the high output voltages are specified for currents in the order of 100uA (20uA, 100uA and 400uA), while the negative or lower output voltage is specified for currents near 1ma (50uA, 500uA, 1mA and 4ma).  This indicates how close to the supply rails you can get depending on the loading at the output.

[Sir H C]

Okay, R1 and Q11 are current limiting.  If the voltage across the resistor gets too high, it pulls the gate of Q13 down to shut if off.  The four transistors on the right with the diode (odd usually see with a resistor there) is the bias generation.  Doesn't show the start-up usually needed for this.  For the drive, this is a common two stage design, Q9 is just weird, not sure what that is trying to do.  The cool concept for the PMOS drive is actually something akin to a class AB circuit where you use Q7 to give a similar dirve current as is going to the NMOS output device then with Q6 you have a current source that via Q8 shunts off some the current over this to be the actual drive for PMOS device Q12.  Now Q9 makes sense, it should be diode connected like Q8.  Cool little circuit, there are better class AB rail-to-rail architectures out there.

This is a rail-to-rail output design but not rail to rail input.  I would guess about a volt from the top rail.

[gez]

Well, took some time off from soldering and quickly breadboarded a circuit.

Set up the first stage of an ICL7621 for variable gain and the second stage as a non-inverting buffer.  Had the load set up to switch between either output.

1. No soft clipping (dives straight in) Edit: tested with a scope by the way.

2. There is asymmetry, but not how I imagined it would be.  The output ends up being half-wave rectified: positive output cycles only, so both Paul and HC were right about the p-channel being able to source more current.  Had I made the load lighter so that -ve cycles weren't loaded to the point that they were obliterated, I doubt clipping would have been soft, and even if it were the current sourcing/sinking mismatch would mean that +ve output cycles are always going to hard clip when reaching the +ve rail.

3.  Sounds bad.  There's a slight octave effect around the 12th fret, as you'd expect, other than that it sounds like a woolly Ge diode clipper...probably slightly softer than a normal opamp driven into clipping but that's about it.

Perhaps other chips might be more successful at doing this?  Nice idea, shame it didn't pan out.

[WGTP]

Thanks for the report.  Probably a silly question, but looking at all those ?MOSFETS? in the lay outs, would those all be clipping at once when driven hard?  Like in a Rat or Dist+?  Like a multi-stage distortion???

[Jay Doyle]

Thanks for the report.  Probably a silly question, but looking at all those ?MOSFETS? in the lay outs, would those all be clipping at once when driven hard?  Like in a Rat or Dist+?  Like a multi-stage distortion???

Most of those MOSFETs you are seeing in that schematic are for biasing purposes, current sources or sinks, actually. Only a couple actually do the amplification.