HelloYawei,

flicker noise is directly related to the quality of the oxyde-channel interface. In more modern technologies, this interface is usually of better quality, so that, for equal device size (i.e. not scaled), the flicker noise in the modern technology should drop, not increase. But I guess you scaled down your devices, so that flicker noise should remain approx constant (for a constant scaling factor).

Not, the problematic issue with flicker noise is that the quality of the model depends a lot on the efforts made by your foundry to characterize it. You know, the longer the measurement, the more noise you get (as it raises at lower frequencies).

It is not clear in your post whether you are talking about simulation results or measurements. In the first case, I would check the quality of the model (the process documentation should give some characterized values). In the second cas, I would carefully check the scaling rules operated during the design transfer.

Hope this helps

Paul

Hi, Yawei,
Looks to me that you are using BSIM3 models, as the
noise flags in BSIM4 are TNOIMOD and FNOIMODE.

There are 2 types of mosfet noise models in BSIM3:
1.Hspice mosfet noise model ~ NLEV, AF, KF and GDSNOI
2.Berkeley mosfet noise model ~ NOIMOD, NOIA, NOIB, NOIC, EM, AF, EF, KF and GDSNOI.

For NOIMOD 1, spice2 noise model is used for flicker noise,
which is a function of Ids, Leff and f.  For NLEV3, the
flicker noise is a function of gm, Leff, Weff, and f.  So, even
for the same process from the same fountry, and the same
design, the flicker noise should not be the same.

>Does the flicker noise of MOS transistor vary so much >(about 2 order of magnitude) from one generation to >next generation?
It should not be that different.  Please check the KF value
in your spice models.

>I will check them again and try to figure out the equation >of both noise model.  
You can refer to the following for details:
For  Hspice noise model:
Hspice MOSFET models manual -> Chapter 2 Technical
Summary of MOSFET Models -> Noise models

For Berkely noise mode:
BSIM3V3 Manual -> Chapter 8 Noise Models
The BSIM3V3 manual URL:
http://www-device.eecs.berkeley.edu/~bsim3/arch_ftp.html

Let me know if I can provide you with further info.

Thanks.

--Charlie

Hi, crhu,

I have checked SPICE model for 0.25um CMOS process of Foundry B.

For its n-channel transistor, KF=3.454E-24, AF=0.8824.
For its p-channel transistor, KF=1.184E-23, AF=1.2828.

Is that reasonable?

Thanks
Yawei

Yawei,

There is no chance the flicker noise jums by a factor of 100!
It only can happen in case of gross misprocessing.

Do you use binned model in old or new process? Do noise paremeters present in all bin sections?
Do old and new schematics differ in gain, dc state or some other significant parameter, other than flicker noise?

As for your second question, in 0.25um process pmos still should have an edge in 1/f noise performance.  How reputable is your foundry B?

Hi Yawei,

In fact 10 times difference in flicker noise is still too large to be explained by process differences. Can it be that your old 0.35um models is problematic and shows too low 1/f noise? Does it correlate with measurements?