AnsweredAssumed Answered

Issue on Noise Model of MOSFET

Question asked by TNAKA on Jun 12, 2003
Latest reply on Jun 18, 2003 by TNAKA
In BSIM3v3, ADS adopts a following equation as channel thermal noise
for Noimod=1(most basic model).
   8*k*T/3*(gm)
This is consistent with Nlev=-1,0,1,2 of HSPICE.

But other vendor tools such as Cadence-Spectre and Mentor-Eldo
adopt different equation for Noimod=1(SPICE2).
These tools adopt following as most basic model.
   8*k*T/3*(gm+gds+gmb)
Here gm=transconductance, gds=channel conductance, gmb=body transconductance

This causes different NF results when comparing with Spectre using Noimod=1(SPICE2).
According to chapter-8 in BSIM3v3 manual from UCB,
I think Cadence-Spectre's equation is major now.
So ADS should have same equation.
This issue is true for all other MOSFET models in ADS.

I compared results for other Noimod.

 For Noimod=1(Flicker=SPICE2,Thermal=SPICE2),
   ADS doesn't give same NF with Spectre.

 For Noimod=2(Flicker=BSIM3v3,Thermal=BSIM3v3),
   ADS gives same NF with Spectre.
 
 For Noimod=3(Flicker=BSIM3v3,Thermal=SPICE2),
   ADS doesn't give same NF with Spectre.
 
 For Noimod=4(Flicker=SPICE2,Thermal=BSIM3v3),
   ADS gives slightly different NF with Spectre.

Why Noimod=4 can't give same result ?
Noimod=4 should be same between ADS and Spectre
since it uses BSIM3v3 model not SPICE2 model
as channel noise.
Here I set flicker noise off(Kf=0), so this discrepancy is not due to flicker noise model difference.

Flicker noise model for noimod=1(SPICE2) is almost same except for Leff^2 or Leff*Weff.  

Outcomes