Flicker Noise Extraction for Scalable MOS Simulation Models · Page 18 Summary Together with the...
Transcript of Flicker Noise Extraction for Scalable MOS Simulation Models · Page 18 Summary Together with the...
Flicker Noise Extraction forScalable MOS Simulation Models
Dr. Thomas Gneiting
AdMOS GmbH
Page 2
Content
Flicker noise equations in modern MOS models
Noise simulation in different simulators
Steps in parameter extraction using a PSP example
Problems to be solved
Summary
Page 3
Common Flicker NoiseModels
100 101 102 103 104 105
10-17
10-18
10-19
10-20
10-21
10-22
10-23
10-24
f [Hz]
Sid [A²/Hz]
Frequencyexponent
Start ofthermal noise
100 101 102 103 104 105
10-17
10-18
10-19
10-20
10-21
10-22
10-23
10-24
f [Hz]
Sid [A²/Hz]
DC biasdependency
+ Dimension dependency
The most MOS simulation models provide the following 2 principaldegrees of freedom to adjust the flicker noise behavior.
Page 4
BSIM3 Flicker Noise
EFeffOX
AF
DSid
fLC
IKFfS
2)(
*
2
82
2
22
14
14
82
2
10*2
2102
102log
10*2)(
NN
NNOICNNOIBNOIA
fLINTNOILW
LITk
NNNOIC
NNNOIBN
NNOIA
fALINTNOILC
IqTkfS
l
ll
EFeffeff
CLMDSB
lOlO
l
O
EFbulkeffoxe
DSeffB
id
36
2
'
'
104)(,)(
EFeffeff
DStmwi
wimitli
wimitliid
fLW
IVNOIAfS
SS
SSfS
NOIMOD=1,4,5
NOIMOD=2,3,6
Vgs>Vth+0.1:
Vgs<Vth+0.1:
“classic” SPICE model
BSIM3 model
Page 5
BSIM4 Flicker Noise
EFeffOX
AF
DSid
fLC
IKFS
2
2*
2
102
2
22
*
*
102
2
,
10*2
2log
10*2)(
NN
NNOICNNOIBNOIA
fLINTNOILW
LITk
NNNOIC
NNNOIBNN
NNNOIA
fALINTNOILC
IqTkfS
l
ll
EFeffeff
CLMDSB
lOlO
l
O
EFbulkeffoxe
DSeffB
invid
102*
2
,
,,
,,
10)(,
)()(
)()()(
NfLW
ITkNOIAfS
fSfS
fSfSfS
EFeffeff
DSBsubVtid
subVtidinvid
subVtidinvid
id
FNOIMOD=0
FNOIMOD=1
Inversion:
SubVth:
“classic” SPICE model
Page 6
PSP Flicker Noise
NNNNFCNFB
NN
NNNNFCNNFBNFA
NGCf
IqfS
m
m
m
vsatox
DSTfl
**
*
*2**
*
2
2
2/
2/ln)(
EE
ENEN
EE
ENEN
EE
ENEN
LW
LWNFCLWNVC
LW
LWNFBLWNFB
LW
LWNFALWNFA
Local model:
Scalingequation:
Fixed frequency exponent of 1!
Page 7
Noise in ADS
define noise_circ (D G OUT)Flicker_Noise:MAIN D1 G 0 0 \L=10u W=10u Mult=1;; sense the noise currentR:Rdummy D D1 R=1m Noise=no;;convert noise current into anidentical voltageZ_Port:v D1 D OUT 0 Z[2,1]=1R:Raux OUT 0 R=1k Noise=no;end noise_circ
Page 8
Noise in Spice, HSPICE
.subckt noise_circ 1 2 5 7MAIN 9 6 0 0 Flicker_Noise+ L=10u W=10u NF=1L2 6 2 1000000C2 5 6 1000000H1 8 0 V1 1R1 7 8 1e-4V1 1 9 0.ends
Page 9
Content
Flicker noise equations in modern MOS models
Noise simulation in different simulators
Steps in parameter extraction using a PSP example Measurement (see paper from IHP, Falk Korndörfer)
Determination of the frequency exponent
Determination of bias dependency
Extraction of parameters for scalable noise models shown withan example using the PSP model
Problems to be solved
Summary
*) Based on the work of Knoblinger, Grabinski, Sischka
*)
Page 10
Frequency Exponent (1)
100 101 102 103 104 105
10-17
10-18
10-19
10-20
10-21
10-22
10-23
10-24
f [Hz]
Sid [A²/Hz]
EF
If the model provides afrequency exponent, it can bederived from the slope of themeasured curves in logarithmicrepresentation:(Example for SPICE model, the othermodels behave similar)
xmcy
fEFconfS
fconfS
fLC
IKFfS
id
EFid
EFeffOX
AF
DSid
loglog)(log
)(
)(2
Page 11
Frequency Exponent (2)
In the real world:
EF must fitseveral deviceswith diff. L, Wsimultaneously
Take intoaccount onlyparts of thecurves due tomeasurementrestrictions.
Page 12
DC Bias dependency (1)
100 101 102 103 104 105
10-17
10-18
10-19
10-20
10-21
10-22
10-23
10-24
f [Hz]
Sid [A²/Hz]
Sid [A²/Hz]
0 1 2 3 4
10-17
10-18
10-19
10-20
10-21
Vg[V]
Once the slope is given, the noise values at 1Hz can be determined byextrapolating the measured curve to 1Hz.
EFeffOX
AF
DSid
fLC
IKFfS
2)(
=1 at 1Hz
(Example for SPICE model, the othermodels behave similar)
Page 13
DC Bias dependency (2)
3 Steps:
Select range ofcurves todetermine slope
Make curvesslopeless
Apply linearfitting to theselected range ofthe curve anddetermine cross-point at 1Hz
Page 13
DC Bias dependency (2)
3 Steps:
Select range ofcurves todetermine slope
Make curvesslopeless
Apply linearfitting to theselected range ofthe curve anddetermine cross-point at 1Hz
Page 13
DC Bias dependency (2)
3 Steps:
Select range ofcurves todetermine slope
Make curvesslopeless
Apply linearfitting to theselected range ofthe curve anddetermine cross-point at 1Hz
Page 13
DC Bias dependency (2)
3 Steps:
Select range ofcurves todetermine slope
Make curvesslopeless
Apply linearfitting to theselected range ofthe curve anddetermine cross-point at 1Hz
Page 14
Parameter Extraction – PSP (1)
Sid(f=1Hz)@ diff Vg,Vd, L, W
SolverSfl(1Hz)=f(Parameter,Dimensions,DC Bias)
Iterative Solutionusing a modifiedLevenberg-Marquardt algorithm
Noise parameters
PSP:PSP1020 Par.L, W, MULTVG, VD
PSP:NFALW, NFBLW,NFCLW
Page 15
Parameter Extraction – PSP (2)
1. Simulation withdefault valuesof NFALW,NFBLW,NFCLW
2. Performextraction inca. 1s andrepeatsimulation
Example parameterextraction process:
Page 15
Parameter Extraction – PSP (2)
1. Simulation withdefault valuesof NFALW,NFBLW,NFCLW
2. Performextraction inca. 1s andrepeatsimulation
Example parameterextraction process:
Page 15
Parameter Extraction – PSP (2)
1. Simulation withdefault valuesof NFALW,NFBLW,NFCLW
2. Performextraction inca. 1s andrepeatsimulation
Example parameterextraction process:
Page 16
Parameter Extraction – PSP (3)
Final resultfor 4differenttransistordimensions.
Page 17
Problems to be solved
The scalability of the simulation models isstill not good enough.
The PSP scaling equation:
does not allow to modify parametersindependently for L and W. Actually, onlythe product L·W is taken into account.
EE
ENEN
EE
ENEN
EE
ENEN
LW
LWNFCLWNVC
LW
LWNFBLWNFB
LW
LWNFALWNFA
Fitting for smalldevice cannot beimproved in this areawithout the distortionof the other devices.
Page 18
Summary
Together with the measurement of flicker noise ( presentedby Falk Korndörfer, IHP), the shown extraction strategy isthe basis of a complete noise modeling method.
The very effective simultaneous extraction of flicker noiseparameters for different devices with different dimensions isthe key improvement of this tool.
The shown methodology for PSP was implemented forcommon MOS models (BSIM3, BSIM4) and can be easilyextended to other models like HiSIM2 etc.
The co-operation between IHP and AdMOS resulted in acommercial available Flicker Noise Modeling Tool. Fordetails, please see:www.admos.de Products Flicker Noise System