The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Analog Design >> How to reduce 1/f noise in BandGap
https://designers-guide.org/forum/YaBB.pl?num=1303130233

Message started by ic_engr on Apr 18^th, 2011, 5:37am

Title: How to reduce 1/f noise in BandGap
Post by ic_engr on Apr 18^th, 2011, 5:37am

Hello,

I have a design of BandGap circuit based on Banba's paper. I am interested in low noise from 0.1Hz-20Hz. Presently the noise is around 30uVrms across 0.1Hz-30Hz and most of it is 1/f noise.
I am wondering if I can make use of chopper stabilizing technique to reduce this noise since this is mostly 1/f ?

Any suggestions ?

Regards,

ic_engr

Title: Re: How to reduce 1/f noise in BandGap
Post by raja.cedt on Apr 18^th, 2011, 9:50am

hi,
theoretically you can do it but its gives some other problems like clock glitches on bias line through some Cgd or other means, but before doing you can verify the following things.

Opamp and current mirror are main noise sources. So estimate how much current mirror is generating by replacing opamp with an ideal VCVS, if it contributing more then try optimize like by increasing the device size or introducing degeneration. Run separate noise simulation for opamp and try to optimize.

Thanks.

Title: Re: How to reduce 1/f noise in BandGap
Post by Rakesh on Apr 18^th, 2011, 10:38am

Hi,
I think if u r going to degenerate with resistors the noise is going to be high. Its better to increase both the length and width of the device, so that the flicker noise comes down.
Correct me if i am wrong.
Averging will reduce all the high frequency stuff. Flicker noise is maximum at low frequency.
Can some one intutively explain how increasing the size will reduce flicker noise.
Rakesh

Title: Re: How to reduce 1/f noise in BandGap
Post by loose-electron on Apr 18^th, 2011, 11:36am

Large W and L will drop the flicker noise down,. but not elinminate it.

Intuitive explanation? - A larger portion of the current thru the transistor passes thru the silicon away from edges of the transistors. The catch-release of the electron/hole is less when not along the edge of a junction.

Chopper methods can be used to push flicker to a HF and get rid of (reduce!) mismatch effects, and then you filter the DC output and get rid of the HF content of the chopper and the up converted flicker noise.

Title: Re: How to reduce 1/f noise in BandGap
Post by raja.cedt on Apr 18^th, 2011, 11:51am

hi lose electron,
do have any reference/document for BGR with chopper stabilization?

Thanks.
Raj.

Title: Re: How to reduce 1/f noise in BandGap
Post by Rakesh on Apr 18^th, 2011, 2:02pm

Thanks jerry,
Please let me know some good material on the chopper stabilization

rakesh

Title: Re: How to reduce 1/f noise in BandGap
Post by Alexandar on Apr 19^th, 2011, 12:10am

Use google. There are so many papers with bandgap/choppers out there.

Title: Re: How to reduce 1/f noise in BandGap
Post by ic_engr on Apr 19^th, 2011, 8:33am

My BW is from 0.1Hz to 20Hz.
Before I posted, I had already checked that even with replacing the opamp with ideal opamp (AHDL), the total noise was dominated by the 1/f noise of the PMOS current sources feeding into the resistors of the core bandgap (e.g in Banba's circuit). Would chooping this help ?

ic_engr

Title: Re: How to reduce 1/f noise in BandGap
Post by rfcooltools.com on Apr 19^th, 2011, 5:42pm

ic_engr,

The 1/f likely is probably coming from the fact that there are two pfets that feed the bandgap core where the 1/f noise difference (not coherent) is multiplied by the DC gain of the opamp with that said I would try to make a common device (coherent) be a current mirror then split the current with either resistors or resistors and two cascodes. It should improve the 1/f, but will likely have other which you will need to resolve. just a suggestion.

http://rfcooltools.com

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Apr 20^th, 2011, 11:18am

A few things.
Increasing the width and length of the MOS will help, but not as much as having leaving the length short and using resistance degeneration for the same amount of Vgs + IR voltage drop that you use. Since you are using Bamba I assume you are already short of headroom so the IR degeneration will use it more efficiently.

... unless your resistors have 1/f! Many of them do.

You can chop the opamp easy enough, but I think your bigger problem will be chopping the mirror. The load resistor will see zero (or 2x) current for short periods of time. Maybe you will be able to filter this out....

You might look into some switched cap bandgaps. Their output can be as small as you want.

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Apr 20^th, 2011, 11:30am

Lex wrote on Apr 19^th, 2011, 12:10am:

Use google. There are so many papers with bandgap/choppers out there.

For those who don't know, probably because you are too old to know better, google scholar is an awesome tool:
http://scholar.google.com/advanced_scholar_search?hl=en&as_sdt=8000000000

Glancing briefly, I didn't see any current mode solutions. I do remember one presented at ISSCC although I'm not sure it was a 1/f solution... I'm guessing it was presented 8-10 years ago but I've been to so many I don't remember the years.

rg

Title: Re: How to reduce 1/f noise in BandGap
Post by ic_engr on Apr 20^th, 2011, 1:04pm

I had already tried the larger widths, an increase of 4X still does not reduce the noise due to 1/f being higher in the process.

The opamp transistors are only contributing 25%. The major noise contribution is from the PMOS transistors feeding into the Ressitors of Banba's core BG circuit.

Any help would be appreciated.

ic_enger

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Apr 20^th, 2011, 1:26pm

Larger widths by themselves do not help because the gm scales up the same amount that the noise goes down. This is why you degenerate with resistance or make the device longer as you go wider. You want to lower the gm so that the flicker voltage at the gate produces less current at the drain.

Also, in general, the more current you burn the less noise you will get for the same voltage at the gates because (among other reasons) the devices are larger. If you burn a lot of current in the Bamba core and then divide it down for the leg that biases the resistor I bet you will have less noise.

Title: Re: How to reduce 1/f noise in BandGap
Post by vivkr on Apr 21^st, 2011, 12:31am

loose-electron wrote on Apr 18^th, 2011, 11:36am:

Intuitive explanation? - A larger portion of the current thru the transistor passes thru the silicon away from edges of the transistors. The catch-release of the electron/hole is less when not along the edge of a junction.

Jerry,

That doesn't sound quite correct to me. If I recall, 1/f noise in MOSFETs is due to carrier trapping at the Si-SiO2 interface. That's where most of the traps are, not so much at the P-N junctions created at the S/D implants. The surface-dominated behavior of 1/f noise is one of the reasons why bulk devices such as JFETs or BJTs show very little 1/f, which was also true for PFETs in older technologies using buried channels.

A larger gate area will lead to an ensemble averaging of the trapping behavior. In other terms, if you are sending the same amount of current through a larger gate, then the carriers are likely to get stuck in many different, independent traps on their way home. For very tiny devices, you may be unlucky enough to get one single trap which looks rather nasty. I think there's an illustration in Tsividis' book.

Regards,
Vivek

Title: Re: How to reduce 1/f noise in BandGap
Post by loose-electron on May 2^nd, 2011, 1:11pm

vivkr wrote on Apr 21^st, 2011, 12:31am:

loose-electron wrote on Apr 18^th, 2011, 11:36am:

You just said the same thing that I said, with a few more details. :D

Title: Re: How to reduce 1/f noise in BandGap
Post by analog_wiz on Nov 30^th, 2012, 12:26am

You can also use on chip RC (Small area no external components) filters to

filter out the noise.

Title: Re: How to reduce 1/f noise in BandGap
Post by loose-electron on Nov 30^th, 2012, 1:39pm

analog_wiz wrote on Nov 30^th, 2012, 12:26am:

You can also use on chip RC (Small area no external components) filters to

filter out the noise.

Flicker noise is generally DC and near DC,
so filtering does not get you much other than limiting the BW a bit.

Google and IEEE Explore will be your research sources.

Title: Re: How to reduce 1/f noise in BandGap
Post by analog_wiz on Nov 30^th, 2012, 7:37pm

To Optimize a BG for noise will include reducing both (thermal as well as flicker). You can use (on chip rc using special techniques) to get cutoffs as low as 500hz(3db).Loose electron, have you used any other techniques other than those discussed here.

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Nov 30^th, 2012, 7:44pm

analog_wiz wrote on Nov 30^th, 2012, 7:37pm:

The original poster was concerned with bandwidths less than 20 Hz. To do that your options are pretty much larger devices, correlated double sampling or chopping.

Title: Re: How to reduce 1/f noise in BandGap
Post by analog_wiz on Nov 30^th, 2012, 8:26pm

OK Robg. Thanks for getting me back to the topic. Also do you have any no's for integrated noise for frequencies upto 10k with chopping.what kind of applications will need that kind of low noise from bandgap.high resolution adc's?

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Nov 30^th, 2012, 8:56pm

analog_wiz wrote on Nov 30^th, 2012, 8:26pm:

Sorry, I don't have any numbers off the top of my head.

Title: Re: How to reduce 1/f noise in BandGap
Post by raja.cedt on Dec 1^st, 2012, 1:08am

hello all,
please the following website papers, you get real numbers and current state of art.

http://wwwetis.et.tudelft.nl/people/biography/projectleaders/makinwa_kofi/publications_2012.php

Thanks,
Raj.

Title: Re: How to reduce 1/f noise in BandGap
Post by loose-electron on Dec 1^st, 2012, 3:58pm

analog_wiz wrote on Nov 30^th, 2012, 7:37pm:

After you got a BG adjusted where you wanted (curvature correction, offsets and all that)

To the best of my knowledge you got:

Bigger W & L for lower flicker

Lower impedance, higher currents for lower White-thermal-johnson noise.

Chopping to up-convert the near DC noise flicker noise

Bipolar design to get away from MOS flicker noise. (reminder - BJT
devices have flicker noise as well, just orders of magnitude
smaller than MOS devices)

Filtering output

Filtering select nodes in the BG

That's what I am aware of.

Title: Re: How to reduce 1/f noise in BandGap
Post by Kevin Aylward on Jul 6^th, 2013, 7:44am

Use a process with a half decent bipolar. This incessant drive for cmos only is pretty much daft. Many “cmos” processes actually do have a decent bipolar. Just avoid those silly 45nm ones. Who wants to do analog with a few atoms of silicon anyway…

Title: Re: How to reduce 1/f noise in BandGap
Post by Venky_analog on Feb 3^rd, 2016, 6:23pm

Hi,

I have a question here. How can source degeneration reduce flicker noise? If it's because of the reduction in the 'gm' of the transistor, then isn't it going to affect the signal by the same amount, thereby keeping the SNR the same?

Title: Re: How to reduce 1/f noise in BandGap
Post by RobG on Feb 3^rd, 2016, 8:02pm

Venky_analog wrote on Feb 3^rd, 2016, 6:23pm:

In this case it is a current mirror that doesn't have a signal - just noise.

In general, for a given DC current, the larger the gate voltage, the lower the noise. So start out with the assumption that we have made the gate voltage as large as headroom will allow. There are two ways you can get this large gate voltage:
1) Small W, large L (gives a large Vgs-Vt)
2) Large W, small L and resistive degeneration

If the total W*L is the same for both cases then (2) will have less 1/f noise current because its gm is less.

The optimum for noise would be to use a really wide device with minimum length, however, that maximizes drain capacitance and minimizes output impedance so there is a trade-off.

Title: Re: How to reduce 1/f noise in BandGap
Post by DanielLam on Feb 5^th, 2016, 12:40pm

Venky_analog wrote on Feb 3^rd, 2016, 6:23pm:

Source degeneration can reduce both flicker and thermal noise (and any other noise produced by that transistor). First, you must understand why the noise of a cascode does not matter much. Then apply this thinking to the source degeneration case. They are very similar, just one has a degeneration resistor instead of a degeneration ro.

Assume there is a noise source at the input of the cascode, what happens to its gain as you increase the source impedance/resistance?

The cascode case is described in Razavi's book in the noise chapter.