The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Analog Design >> miller compensatone
https://designers-guide.org/forum/YaBB.pl?num=1307475244

Message started by vineeth on Jun 7^th, 2011, 12:34pm

Title: miller compensatone
Post by vineeth on Jun 7^th, 2011, 12:34pm

hey do anybody have an idea about the maximum value a miller compensation capacitance can take in 90nm tech...

Title: Re: miller compensatone
Post by loose-electron on Jun 7^th, 2011, 5:37pm

How much space you got to waste?
I think you need to give a much more specific question, way too vague

Title: Re: miller compensatone
Post by vineeth on Jun 7^th, 2011, 9:03pm

sorry fr that and thanks for the reply.....well my unity gain bandwidth is 10Mhz a gain of 60db and a phase margin greater than 65. my output load would be in pico range.
currently i am getting the specifications with a miller cap of 100p. i wana make sure this is not gona take a large space in 90nm tech.

Title: Re: miller compensatone
Post by raja.cedt on Jun 7^th, 2011, 9:36pm

first design open loop amplifier with specified dc gain. So you would get gm1 and gm2. Generally you have load cap spec from there your second pole would be gm2/cl, for 60 deg PM at least place UGB 4 times below 2nd pole. so now you wil get ugb and that is gm1/cc.

you can refer razaavi for this.
for all feedback and op amp compensations better follow this site.

http://www.ee.iitm.ac.in/~nagendra/videolectures/doku.php?do=login&id=200901vlsiconf%3Astart

Title: Re: miller compensatone
Post by vineeth on Jun 8^th, 2011, 2:51am

thanks raja..i was exactly following the same..inorder to kep my ugb well below second pole..i need to increase ma miller cap...actually ugb is one of ma spec(10 Mhz)...and i m able to achieve this using a miller cap of 100pf. pm achieved is around 71..i wana make sure 100pf is not a large value in 90nm tech.

Title: Re: miller compensatone
Post by vivkr on Jun 8^th, 2011, 4:21am

vineeth wrote on Jun 7^th, 2011, 9:03pm:

what a terrible waste of area! I would suggest that you go over the basics of miller compensation once so as to better understand how pole splitting works. as you do not mention either power consumption or noise as a reason, I suppose that those are not critical.

with that out of the way, the general rule is that the compensation cap is NEVER made larger than the output load cap in a good design. you have got gm1, gm2, Cc and Cl. Usually, Cl will be fixed by the application. And gm1 is tied to Cc. So you still have two parameters to play with: gm2 and Cc. Use both! Usually, gm2 is about 4x-5x larger than gm1 for decent phase margin. With that, you will also get a decent value for Cc. And don't forget to nullify that RHP zero using one of the various methods available for that purpose...

Vivek

Title: Re: miller compensatone
Post by raja.cedt on Jun 8^th, 2011, 9:24am

hi vineeth,
general rule of thumb cc=.25*cl.

thanks.

Title: Re: miller compensatone
Post by loose-electron on Jun 8^th, 2011, 10:24am

100pf is a ton of real estate inside a chip, that probably needs to be reduced in size.

Funny thing - I have not used pole splitting in many years.

I use a source follower to drive the feedback capacitor, which provides feedback while getting rid of the HF feedforward path. (No resistor anymore in the feedback path that way)

Also, for best swing, outputs are OTA, class AB, so that you can get rail to rail output.

Just my approach, others are viable, but the above has served well for me.

Title: Re: miller compensatone
Post by vivkr on Jun 10^th, 2011, 2:34am

loose-electron wrote on Jun 8^th, 2011, 10:24am:

i.e. Ahuja compensation. Yes! that's a good way to do it. However, I think the classical Miller compensation scheme is not so bad. At the very least, one is forced to confront all these nasty design issues ;) I actually cannot think of another example circuit which is so popular, yet so poorly understood and so full of traps lurking everywhere.

Vivek

Title: Re: miller compensatone
Post by vineeth on Jun 11^th, 2011, 5:59am

i dnt have a load cap specification..but my high gain opamp has a ugb spec of 10mhz and a Imax of 500u...my rhp zero is taken care of and so is the phase margin..bt to reduce ma ugb and hence to push ma dominant pole inside i need to increase ma miller cap..i mt able to keep ma ugb within 10Mhz only with a miller cap minimum of 80pf

Title: Re: miller compensatone
Post by loose-electron on Jun 11^th, 2011, 2:15pm

Ahuja? I would not know, I devised that method about 20 years back and have used it in a number of designs, because of the fact that you get less space to do this, and less process varioance induced pole shifting

vivkr wrote on Jun 10^th, 2011, 2:34am:

loose-electron wrote on Jun 8^th, 2011, 10:24am:

Title: Re: miller compensatone
Post by RobG on Jun 13^th, 2011, 8:03am

vineeth wrote on Jun 11^th, 2011, 5:59am:

Vineeth - you can also reduce ugbw by reducing the gm of the input diff pair by using less current or more Vgs-Vt. 100pF is huge -- I've only needed something like that to lower the ugb to much less than 100kHz to get total noise down. Why do you need such high tail current?

You are going to have very low slew rate. On the other hand it does not necessarily waste area since mim caps can be placed over active circuitry which would otherwise be unused. Still, it is large compared to most designs.

Vivek - I believe LE's method isn't Ahuja. With Ahuja the feedback path is output -> Ccomp-> source of common gate amp -> gate of output device. With LE method the path is output -> gate of source follower -> Ccomp -> gate of output device.

Title: Re: miller compensatone
Post by RobG on Jun 13^th, 2011, 9:08am

loose-electron wrote on Jun 8^th, 2011, 10:24am:

Then again, maybe I don't understand what you are doing. How do you get rail-to-rail if you have a source follower in the feedback path?

Title: Re: miller compensatone
Post by vivkr on Jun 14^th, 2011, 1:03am

RobG wrote on Jun 13^th, 2011, 8:03am:

[quote author=vineeth4tech link=1307475244/0#9 date=1307797180]i dnt

Vivek - I believe LE's method isn't Ahuja. With Ahuja the feedback path is output -> Ccomp-> source of common gate amp -> gate of output device. With LE method the path is output -> gate of source follower -> Ccomp -> gate of output device.

Rob,

Thanks for pointing out the error. Indeed, the use of a source follower to turn the bidirectional miller coupling cap into a unidirectional (to first order) feedback network would not be the same as the Ahuja method.

The source follower method would also not give rail-rail output swing, although the Ahuja method might work there still. Your comments Jerry?

Vivek

Title: Re: miller compensatone
Post by loose-electron on Jun 14^th, 2011, 1:29pm

RobG wrote on Jun 13^th, 2011, 9:08am:

loose-electron wrote on Jun 8^th, 2011, 10:24am:

Then again, maybe I don't understand what you are doing. How do you get rail-to-rail if you have a source follower in the feedback path?

AC coupled signal, Gate drive of source follower is on OTA output, with a current source off of it to keep it biased properly (goes triode when its down, but still works). Source feeds a cap (drive the bottom plate side) and the other side of the cap goes to your feedback point. Done.

Title: Re: miller compensatone
Post by vivkr on Jun 16^th, 2011, 1:01am

[quote author=loose-electron link=1307475244/0#14

AC coupled signal, Gate drive of source follower is on OTA output, with a current source off of it to keep it biased properly (goes triode when its down, but still works). Source feeds a cap (drive the bottom plate side) and the other side of the cap goes to your feedback point. Done.[/quote]

Jerry,

Would you mind posting a schematic? Still cannot see how you get rail-rail output with the arrangement you are describing.

By the way, the Ahuja compensation scheme was published in JSSC about 30 years ago (Dec. 1983), around the time when MOS circuits started to phase out BJTs in common telecomm ICs, i.e. when MOS started to go mainstream, and the problem of the RHP zero started to get more attention as BJT amps didn't suffer much from this.

There were countless variations published around this time and earlier which all attempted to get rid of the nasty RHP zero. The scheme using a MOS biased in triode region as a resistor to push out the zero is a little older and was published by Tsividis if I recall correctly. The approach using a source follower which you describe was also published around this time although I cannot recall by whom (The output swing limitation was a problem there as most people want to use two-stage opamps precisely to get high swing).

The difference between using an active stage (source foll./Ahuja) and a passive R is of course one of topology. You are burning a bit more current but you have an active stage. In principle, you can be about gm*ro times better in mitigating the problem loosely speaking.

Vivek

Title: Re: miller compensatone
Post by fonseca.ha on Jun 16^th, 2011, 9:26am

To add that I agree, 100pF is way too much. To be honest for me 10pF is when a compensation capacitors starts looking too big and when that happens its normally time to think about what else can be done in the design.

I agree with loose electron that the source follower driving the miller capacitor is a nice technique, but sometimes it can be hard to implement if you're doing a rail to rail design.

You also have to think about which device to use as compensation capacitor. For 90nm core moscaps are no longer a good option because they leak and you can actually see a reduction in DC gain because the leakage grows with gate voltage which you can model in the 1st order as being a resistor in parallel.

The need to use thick oxide moscaps or metal caps may make the area look even less attractive.

Humberto

Title: Re: miller compensatone
Post by RobG on Jun 16^th, 2011, 9:39am

fonseca.ha wrote on Jun 16^th, 2011, 9:26am:

The need to use thick oxide moscaps or metal caps may make the area look even less attractive.

Humberto

Is there a reason you can't put the metal caps over the active area?

Title: Re: miller compensatone
Post by fonseca.ha on Jun 16^th, 2011, 12:19pm

RobG wrote on Jun 16^th, 2011, 9:39am:

fonseca.ha wrote on Jun 16^th, 2011, 9:26am:

The need to use thick oxide moscaps or metal caps may make the area look even less attractive.

Humberto

Is there a reason you can't put the metal caps over the active area?

Hi RobG
When you refer to active area do you mean other circuitry? If so I guess it depends what the circuitry is. The issues may be related to capacitive coupling.. possible psrr degradation depending where that area is connected to. It may affect matching,
100pF of metal cap may also considerably constraint routing for other signals.

Humberto

Title: Re: miller compensatone
Post by RobG on Jun 16^th, 2011, 12:40pm

I was just responding to your comment that metal caps even look less attractive. They can be pretty capacitive dense (especially if stacked) and can be put over devices. This makes them pretty attractive to me, but figured that maybe something was going on in the process you were talking about. I've never been a fan of gate caps for compensation... the non-linearity is just one more thing to complicate the design.

Title: Re: miller compensatone
Post by loose-electron on Jun 16^th, 2011, 4:35pm

See the attached drawing. This is a simplified variant that I use when teaching a mixed signal design course. This version is **not** rail to rail.

However, if you add a level shifter (up), from the output to the feedback source follower, this can then be a rail to rail functional output.

Title: Re: miller compensatone
Post by fonseca.ha on Jun 17^th, 2011, 12:44am

loose-electron wrote on Jun 16^th, 2011, 4:35pm:

Hi Loose-electron,
Thank you for sharing the schematic. In this circuit the consumption of the output stage (Q8, Q9) seems to be supply dependent, I say this because the gate of Q9 is driven by the source of Q6 which shares its gate with Q8. Therefore VgsQ9+VgsQ6+VgsQ8=VDD. So if supply changes the Vgs change as well affecting the bias of the output stage. It may impact psrr.
Humberto

Title: Re: miller compensatone
Post by vivkr on Jun 17^th, 2011, 3:35am

loose-electron wrote on Jun 16^th, 2011, 4:35pm:

However, if you add a level shifter (up), from the output to the feedback source follower, this can then be a rail to rail functional output.

Won't your "up" level shifter (the one not shown here but needed for rail-rail output) clamp hard about 1 Vgs below supply, breaking the feedback loop? The Ahuja scheme puts that level shifter on the other end of the cap avoiding that problem rather neatly.

Vivek

Title: Re: miller compensatone
Post by loose-electron on Jun 17^th, 2011, 3:15pm

fonseca.ha wrote on Jun 17^th, 2011, 12:44am:

In this circuit the consumption of the output stage (Q8, Q9) seems to be supply dependent, I say this because the gate of Q9 is driven by the source of Q6 which shares its gate with Q8. Therefore VgsQ9+VgsQ6+VgsQ8=VDD. So if supply changes the Vgs change as well affecting the bias of the output stage. It may impact psrr.
Humberto

No kidding - as I said before, I use this material in a class to illustrate a concept. The concept being illustrated here is the use of a source follower to eliminate a feedforward path and thus eliminate the resistor in the compensation circuit.

Title: Re: miller compensatone
Post by loose-electron on Jun 17^th, 2011, 3:30pm

vivkr wrote on Jun 17^th, 2011, 3:35am:

loose-electron wrote on Jun 16^th, 2011, 4:35pm:

However, if you add a level shifter (up), from the output to the feedback source follower, this can then be a rail to rail functional output.

You may want to provide a schematic for the "Ahuja scheme" so we know what you are talking about.

Your level shifting mentioned implies something like diode connected MOSFEt, and that will have limitations with it shutting off. Agreed. A complimentary driver in the feedback is an alternative. (think AB amp, like a logic inverter in architecture) It does not need to be linear.

Title: Re: miller compensatone
Post by vivkr on Jun 20^th, 2011, 12:57am

Jerry,

The scheme is described in the JSSC paper I already mentioned in my previous posts (JSSC, Dec. 1983, An Improved Frequency Compensation Technique for CMOS Operational Amplifiers). There is no diode connected MOSFET there. Many variants (with some improvements) also exist, but the concept can be seen here.

Vivek

Title: Re: miller compensatone
Post by fonseca.ha on Jun 20^th, 2011, 1:16am

Ill have to try this one,
Humberto

Title: Re: miller compensatone
Post by loose-electron on Jun 20^th, 2011, 4:50pm

Ah, its a folding method that injects a current differential to drive gain down. Similar to a folded cascode. Interesting.

Thing is, if you are doing op-amps, I would hope that the system is fully differential with common mode feedback. So the ground referenced op-amps shown here are essentially good for illustration but need to be converted appropriatly.

Title: Re: miller compensatone
Post by RobG on Jun 20^th, 2011, 4:58pm

In cascode amplifiers you can connect the capacitor to the source of the cascode. It is in the signal path, which can cause peaking if the output gm is much larger than the cascode gm, but it otherwise works pretty well.

Title: Re: miller compensatone
Post by vivkr on Jun 21^st, 2011, 12:08am

Jerry,

Yes! it is kind of a folding. Of course, the figure here is only an illustration. I think the original paper only had a single-ended amp though (it was 1983).
But a fully-differential version is preferable also because you will avoid systematic offset (The level shifter branch is sharing current with the input branch).

Rob,

Indeed, the most popular variant of this scheme is to return the feedback to the cascode. However, as you point out, that causes peaking. My experience was that this peaking makes life quite painful, so much so that you do not really reap the full benefits of using this scheme. It turns out to be better to just put the extra stage although you end up burning more power and adding an extra source of noise to your amp.

Vivek

Title: Re: miller compensatone
Post by RobG on Jun 21^st, 2011, 7:00am

vivkr wrote on Jun 21^st, 2011, 12:08am:

Rob,

Indeed, the most popular variant of this scheme is to return the feedback to the cascode. However, as you point out, that causes peaking. My experience was that this peaking makes life quite painful, so much so that you do not really reap the full benefits of using this scheme. It turns out to be better to just put the extra stage although you end up burning more power and adding an extra source of noise to your amp.

Vivek

What kind of amps are you building? It usually isn't a problem for me unless I'm driving a large DC current.

Title: Re: miller compensatone
Post by vivkr on Jun 21^st, 2011, 7:55am

well! I am not building any of those fancy amps anymore, but I did spend quite a while playing with all these topologies in an earlier job. There were different kinds of applications there for sensor interfaces.

I didn't use it for driving R-loads though as there were hardly any such applications. I found it quite handy as you can get away with very low compensation cap and still be stable. A very handy advantage if making high-voltage amps by the way as those caps can be rather huge and you want to save every bit of area.

Also tried it once when I had a tricky amp to stabilize but gave up on it because the noise requirements were too tight and it does give you more noise. I think I ended up using the return to the cascode in that one.

Vivek

Title: Re: miller compensatone
Post by kaart on Aug 29^th, 2011, 9:51pm

@Loose-electron: Can you please attach the circuit???

Title: Re: miller compensatone
Post by harpoon on Aug 30^th, 2011, 6:03am

100pF is just too wasteful (unless you have lots of silicon space) ...

try the following link for ideas :-
http://cmosedu.com/cmos1/bad_design/bad_design3/bad_design_3.htm

(I think it is similar to what vivek pointed out).

Title: Re: miller compensatone
Post by loose-electron on Aug 30^th, 2011, 3:58pm

sushan wrote on Aug 29^th, 2011, 9:51pm:

@Loose-electron: Can you please attach the circuit???

See attached - Q10 & Q11 buffers the output and drives a feedback capacitor, Cc. Because of the Q10 source follower there is no high frequency feed forward.

Title: Re: miller compensatone
Post by harpoon on Aug 31^st, 2011, 1:28am

loose-electron wrote on Aug 30^th, 2011, 3:58pm:

See attached - Q10 & Q11 buffers the output and drives a feedback capacitor, Cc. Because of the Q10 source follower there is no high frequency feed forward.

This is a classic "indirect feedback" topology that is much better than the standard (and outdated) Miller compensation (but uses a tiny bit more current)

However, I just wish to point out that this is also a classic output stage that is a bit "dangerous" as the output stage current is not defined properly.