Auto-zero technique questions

[AMSA84]

Hi guys,

I read about the auto-zero technique in razzavi's book (page 474-475) and I was wondering if we can make this kind of amplifier using differential input to single output amplifier using this technique?

Regards.

 

[dick_freebird]

Well, I've made and tested an auto-zeroed comparator
(differential to single) but whether it follows Razavi's
style I couldn't say. Probably not, it was a kooky part.

But early on in autozero amplifiers, fully differential was
not so popular as today (back then it was 5V CMOS and
higher). You probably could do well looking at JSSC papers
from the '90s for topology cues.

 

[AMSA84]

Thanks for the reply.

I was looking to razzavi's book, I don't know if you have it (tell me, to post the circuit), but basically is a fully differential folded cascode plus an additional differential stage, that connects directly to the side where you have the current bias to the output stage and which will add or subtract to that current the offset voltage.

I was wondering if there there sis any version of this but for single ended output. If there isn't we would need an additional stage.

- - - Updated - - -

EDIT:

By the way, if I wanted to measure the system stability, how should I do it?

 

[Warpspeed]

Hi guys,

I read about the auto-zero technique in razzavi's book (page 474-475) and I was wondering if we can make this kind of amplifier using differential input to single output amplifier using this technique?

Regards.

Not familiar with that particular book, but chopper stabilised amplifiers have been around for a very long time and are readily available as fully contained single op amps with self correcting offset.
Here is a typical device from Intersil:
https://www.intersil.com/content/dam/Intersil/documents/icl7/icl7650s.pdf

 

[FvM]

None of the recent "chopper stabilized" or "zero drift" OPs is using a simple auto-zero topology with chopped signal path like that in Razavi's book. First integrated "commutating autozero" OP ICL760x did nearly 40 years ago, next generation ICL7650 (more than 30 years ago) changed to "chopper stabilized" topology with an uninterrupted main signal path and all later developments followed it up to now.

So why would you implement the historical topology? Single or differentially ended is surely not the problem.

 

[AMSA84]

This is something that is confusing me. You have chopper stabilised and auto zeroing amplifier and then you have one that joins the two functions, correct? The chopper stabilised that you are referring is the chopper stabilised? Where you chop the input and output separately?

 

[FvM]

There may be different names for the same thing. An essential point is if the signal path from input to output is continuous and the offset correction signal only added or if the signal itself is chopped, as with Razavi's example design. In the latter case you observe strong intermodulation with the chopper frequency, thus this topology is avoided by most chopper stabilized amplifiers.

 

[AMSA84]

Thanks for the reply. I will give a tough on this.

Regarding the autozeroing, where you use a capacitor, do you know how to get the compromise between speed and size of the capacitor? Basically how to calculate the capacitor for a given frequency?

 

[Warpspeed]

Read through the fairly comprehensive application notes for the chip.
The chopping frequency is pretty much fixed, and fitting the external capacitor size and type recommended, will provide the specified performance in the data sheet.

 

[FvM]

Early chopper stabilized OPs like ICL7650 used fairly large external 0.1 µF auto zero capacitors. Modern types have internal capacitors. I presume you are designing an IC with built-in capacitors.

Chopper frequency will be selected according to MOSFET noise corner, capacitor values as a trade-off between chip area and charge injection and leakage induced offset error.

 

[AMSA84]

None of the recent "chopper stabilized" or "zero drift" OPs is using a simple auto-zero topology with chopped signal path like that in Razavi's book. First integrated "commutating autozero" OP ICL760x did nearly 40 years ago, next generation ICL7650 (more than 30 years ago) changed to "chopper stabilized" topology with an uninterrupted main signal path and all later developments followed it up to now.

So why would you implement the historical topology? Single or differentially ended is surely not the problem.

There may be different names for the same thing. An essential point is if the signal path from input to output is continuous and the offset correction signal only added or if the signal itself is chopped, as with Razavi's example design. In the latter case you observe strong intermodulation with the chopper frequency, thus this topology is avoided by most chopper-stabilized amplifiers.

@FvM

I don't know if what you are talking about is true or not because this is the first time I heard about this technique. I might have talked about it at the university but it was a long time ago. Who told you that the chopper technique is not being used anymore? You can see that technique on the CDS amplifiers.

Just one side not: The topology that I am referring to in my first post was just to check if it is possible to make an offset cancellation using a topology from diff. input to single ended because most of the things I found was always diff. to differential.

From your comment above about why I want to implement a historical topology, can you specify what historical topology are you referring to? I got confused with your comment, the first one in a quote.

From my understanding, there 3 kind of offset cancellation techniques: autozeroing, which is a sampling technique, chopped stabilization, which is a modulation technique and from what I read is considered to be a continuous time technique and finally CDS (correlated double sampling) which I think it makes use of both the techniques, the chopper and the autozero. Adding to those you have the ping-pong topology which is an alternative solution to the autozero technique where the amplifier is always available, so it's a continuous time technique.

Now, since you said that those techniques are old, what are the new techniques then? Can you develop a bit more your comment, please? Even if it's an old technique, we have seen that old kinds of stuff are being brought to the present nowadays.

Despite that, I think it's a good exercise to start learning the offset cancellation techniques unless you advise me another way to learn offset cancellation.

Allow me just to correct you. The topology in razavi's book it's not a chopper stabilized technique but an auto-zero technique. The chopper-stabilized technique makes use of 4 switches in the input and another 4 on the output depending on the topology you are using.

Regards.

 

[FvM]

An old circuit topology isn't necessarily bad. It's up to your decision if it's still useful.

I wanted to point out that recent chopper stabilized amplifiers are using a different topology. I suggest to pay attention to intermodulation problems of topologies that directly chop the signal path like the shown Razavi circuit. If the comparison of historical with modern chopper topologies confuses you, just ignore my posts.

 

[Warpspeed]

Yes the ancient dinosaur method was to chop the incoming dc signal into a true alternating ac signal (which could then be ac coupled through more than one high gain stage) then reconvert the output back to dc by synchronous switching at the output.

This was even sometimes done with vibrating mechanical contacts !

This was typical 1930's valve technology, often with transformer and capacitor interstage coupling and other types of quaint but long obsolete methods. They were called chopper amplifiers.

Times change, and these days we use chopper stabilised amplifiers, or auto zeroing amplifiers that have a continuous signal path.
However, the old terminology still persists. The technology has changed greatly but modern day stuff is still frequently referred to as a chopper amplifier or chopper stabilised amplifier. Wrong it may be, but that is the language still used.

 

[AMSA84]

An old circuit topology isn't necessarily bad. It's up to your decision if it's still useful.

I wanted to point out that recent chopper stabilized amplifiers are using a different topology. I suggest to pay attention to intermodulation problems of topologies that directly chop the signal path like the shown Razavi circuit. If the comparison of historical with modern chopper topologies confuses you, just ignore my posts.

And I thank you for trying to point that out. However, you keep saying that the topology in razavi's book is a chopper amplifier which is not true, otherwise, an amplifier using SC would be called chopper amplifiers. Why are you calling chopping stabilized amplifier to that topology in razavi's book?

From my understanding and according to the book "Dynamic Offset Compensated CMOS Amplifiers", the autozero amplifier uses capacitors to store the offset in one phase and subtract it on the next phase:


Then you have the chopper amplifier, which modulates the input to get rid of the offset, noise and amplify the signal then brought back to the baseband:


Finally, you have chopped auto-zero amplifier, which I think is the so called chopped stabilised amplifier, that makes use of both:


The topology that you have in razavi's book falls into the autozero amplifiers, so why are you calling it a chopper amplifier? The chopper amplifier doesn't make use of capacitors.

If the comparison of historical with modern chopper topologies confuses you, just ignore my posts.

You could be more specific on what you are referring to and give some examples. I would have thought.

Yes the ancient dinosaur method was to chop the incoming dc signal into a true alternating ac signal (which could then be ac coupled through more than one high gain stage) then reconvert the output back to dc by synchronous switching at the output.

This was even sometimes done with vibrating mechanical contacts !

This was typical 1930's valve technology, often with transformer and capacitor interstage coupling and other types of quaint but long obsolete methods. They were called chopper amplifiers.

Times change, and these days we use chopper stabilized amplifiers, or auto zeroing amplifiers that have a continuous signal path.
However, the old terminology still persists. The technology has changed greatly but modern day stuff is still frequently referred to as a chopper amplifier or chopper stabilised amplifier. Wrong it may be, but that is the language still used.

Thanks for your comment. One of the auto-zeroing amplifiers that you are talking about is, for example, the ping pong amplifier?

I could start learning the autozero technique by grabbing right away these advanced topologies but doesn't hurt to start with the historical one, I guess where the principle of operation is basically what razavi's have in his book for the autozero amplifier. Store the offset and add or subtract it to the main amplifier.

Regards.

 

[Warpspeed]

I have never heard of a ping pong amplifier until just now.

This is the venerable classic chopper amplifier, and before the age of silicon they used real mechanical switch contacts !
For some reason the image will not post

 

[Warpspeed]

Ah that worked.

 

[AMSA84]

Hi everyone,

I have a question regarding MOS switches. From what I know we can have two types of switches: single nmos/pmos or a transmission gate.

When using a single transistor as a switch, supposing in this case that is responsible for letting a capacitor charge or discharge, it is advisable to have a moscap with opposite phase to the transistor switch.

Question: Supposing that we have some sort of circuit, like the one I am discussing on this topic or maybe any type of SC circuits, we have to put that moscap with opposite phase to the transistor switch only where we have a switch plus a capacitor or we have to place it next to each transistor switch we have in the circuit?

From the design that I am doing, the switches sizes are small, otherwise, the offset waveform that I am getting will be affected significantly. For example, I found my switches sizes to be < 1um (with the minimum length of course).

Question: Is this normal?

Finally, if I wanted to be more accurate on my simulations, I should include the switches in the stability analysis of the amplifier that I am using, with the switches closed (fixed gate voltage applied) or not?

If that's a way to do it, then I have a comment. Since the switch has a "dynamic" resistance, that is, when it is switching the resistance varies, at what extent this approach is suitable?

Regards.

 

[Warpspeed]

Its all a question of unwanted charge injection via the gate capacitance.

That may or may not be an important issue, depending on switching frequency, and the relative switching and signal voltages, and the impedances involved.

 

[AMSA84]

So basically what you are saying is a person has to try each of the solutions and see what's the best result it gets?

 

[Warpspeed]

What I am saying is that there is not just one application only for a mos switch.

Is it switching on and off maybe once a day, or on and off at hundreds of Mhz ?
Is it switching microvolts, or several volts ?
Are the input and output impedances tens of ohms, or tens of megohms ?

The solution needs to meet the requirements of the application.
A bit of thought and planning should narrow it down.