Statistical Characterization of Flicker

Noise Fluctuation of a Nano-Scale

NMOS Transistor

Sarmista Sengupta and Soumya Pandit

Abstract The statistical variability of flicker noise for a nano-scale NMOS tran-
sistor due to intra-die variations in channel length, oxide thickness and mobility has
been characterized. This has been done through theoretical estimations as well
Monte Carlo-HSPICE simulation technique. BSIM-SPICE process parameters have
been chosen for characterization purpose. The present work explicitly depicts the
dependence of flicker noise variability on various causes of variability and device
design parameters. 45 nm CMOS process technology has been considered in the
present work. The theoretical model can be effectively used for statistical charac-
terization which is essential for robust analog and RF circuit design.


Keywords Flicker noise Intra-die process variability  Channel length  Oxide
  
thickness Mobility Mean Standard deviation

1 Introduction

With the downscaling of CMOS transistors in the sub-90 nm regime, the impact of
process parameter variations on the performances of CMOS analog and digital
circuit is becoming increasingly important [1]. The increasing amount of intra-die
variability leads to critical challenges in the conventional design procedure of
nano-scale integrated circuits. A circuit optimized with the conventional design
methodology is more susceptible to random performance variability. For robust

S. Sengupta (&)  S. Pandit

Institute of Radio Physics and Electronics, University of Calcutta, Kolkata, India
e-mail: [email protected]
S. Pandit
e-mail: [email protected]

P. K. Giri et al. (eds.), Advanced Nanomaterials and Nanotechnology, 203

Springer Proceedings in Physics 143, DOI: 10.1007/978-3-642-34216-5_21,
Ó Springer-Verlag Berlin Heidelberg 2013
204 S. Sengupta and S. Pandit

design of integrated circuits, it becomes essential to accurately characterize and

model the amount of uncertainty of circuit performances due to the process
variations [1, 2].
The intra-die variation consists of the parametric fluctuations between identi-
cally designed MOS transistors within a short distance. It introduces systematic
variability in patterning and random variability in patterns. The major sources of
random intra-die variability are random discrete dopants, channel length variation,
oxide thickness variation, mobility variation etc., [3]. The variations in the channel
length of the transistor have significant impact on the various electrical perfor-
mances of a MOS transistor [4]. In advanced CMOS technology with oxide
thickness of about 1 nm, the oxide thickness variation due to the comparable
thickness of interface roughness affects many electrical performances of a MOS
transistor [5]. Mobility variations arise from several complex physical mechanisms
such as fluctuations in effective fields, fixed oxide charges, doping, surface
roughness etc. The variations of mobility have also significant impact on the
various electrical performances of a MOS transistor [6].
Flicker noise is an important parameter that needs to be critically considered for
the design of analog circuits as well as RF circuits. However, this has not received
enough attention by the researchers except recently in [7]. In [7], a flicker noise
variability model has been developed considering the intra-die variations of traps.
However, the issues of channel length variation, oxide thickness variation and
mobility variation have not been taken care of. This paper for the first time
presents an approach for statistical characterization of the variability of flicker
noise power spectral density of an NMOS transistor due to intra-die process
variations considering channel length variation, oxide thickness variation and
mobility variation. The BSIM parameters corresponding to these physical
parameters have been chosen for characterization. The statistical study has been
performed theoretically as well as through Monte Carlo-HSPICE simulation
technique. The theoretical estimates have been compared with simulation results
and good accuracy has been obtained. The theoretical study gives an insight about
the dependence of the coefficient of variation on various design parameters such as
channel length and oxide thickness.
The rest of the paper is organized as follows. Section 2 presents the theoretical
technique for characterization. Results and discussion are provided in Sect. 3 and
finally conclusion is drawn in Sect. 4.

2 Statistical Characterization

2.1 Flicker Noise Model

The various mechanisms responsible for flicker noise and their physical expla-
nations were reported much earlier [8]. There are two general concepts explaining
the source of flicker noise in MOS transistors. In the carrier number fluctuation
Statistical Characterization of Flicker Noise Fluctuation 205

theory, the source of flicker noise is explained by the random fluctuation in the
number of carriers in the channel and fluctuations in the surface potential. These
are caused by trapping and releasing of carriers by traps located near the silicon–
silicon dioxide interface. The mobility fluctuation theory attributes flicker noise to
mobility fluctuations due to carrier interactions with lattice fluctuations. The recent
approach is to unify the two theories [9, 10]. According to the unified theory the
total drain current noise power is [9]
 
2 kTq2 ID ls N0 1  2 2

in ð f Þ ¼ 0 Aln þ BðN0  NL Þ þ C N0  NL ð1Þ
acfL2eff Cox NL 2

Where A, B and C are technology dependent parameters, N0 and NL are the

carrier densities at the source and drain ends of the channel, respectively. ID is the
drain current, ls is the surface carrier mobility and Leff is the effective channel
0
length. Cox ¼ eox =tox is the oxide capacitance per unit area, tox is the oxide
thickness. The other constants bear their significances as described in [9]. The
surface carrier mobility is defined as
l0
ls ¼ ð2Þ
1 þ h1 ðVGS  VT Þ þ h2 ðVGS  VT Þ2
l0 is the low field carrier mobility. h1 and h2 are the mobility degradation coef-
ficients due to the vertical channel field. The values of these parameters depend
upon the oxide thickness. The effective channel length Leff is defined as
Leff ¼ Ldrawn þ XL  2dL ð3Þ

where Ldrawn is the drawn gate length, XL is the channel length offset due to mask/
etch effect and dL is the offset due to source/drain lateral diffusion.
Since it is difficult to handle (1) analytically for statistical characterization
purpose, we write (1) in the form
KFID
i2n ¼ 0 Df ð4Þ
Cox L2eff f

where KF is the Flicker noise coefficient. The value of the flicker noise coefficient
depends upon the operating voltages, surface carrier mobility and effective channel
length. These are approximated in the present model through the following three
parameter functions.
2
KF ðVGS Þ ¼ A1 þ A2:VGS þ A3:VGS ð5Þ
 
KF Leff ¼ B1 þ B2:Leff þ B3:L2eff ð6Þ

KF ðls Þ ¼ C1 þ C2:ls þ C3:l2s ð7Þ

206 S. Sengupta and S. Pandit

Here the values of the various parameters are to be determined based upon the
chosen process technology.
The equivalent input-mean-square noise voltage at the gate of the MOS tran-
sistor is given as
 
2 i2n KF VGS ; ls ; Leff
en ¼ 2 ¼ 02 Df ð8Þ
gm 2fCox Weff Leff leff

where Weff is the effective channel width.

In presence of the lateral field, the surface mobility of the
 carriers
 is modified
and the effective carrier mobility is given as, leff ¼ ls :f Leff : This functional
form is written as,
 
  Leff
f Leff ¼ D1  D2 exp  ð9Þ
D3
Here D1, D2 and D3 are three unknown parameters whose values are to be
determined based upon the chosen process technology.

2.1.1 Flicker Noise Coefficient Extraction

In order to match with HSPICE simulation results corresponding to 45 nm CMOS

process technology, the exact values of the flicker noise coefficient have to be
extracted. This depends upon the applied gate-source voltage. For extraction
purpose, it is assumed that the noise voltage is measured at 1 Hz bandwidth so that
the term Df is unity. With this (8) can be written as
" #
 2 KF
log en ¼ log 02  log½ f  ð10Þ
2leff Weff Leff Cox

The intercept of the log [f] vs. log e2n is given by
" #
KF
log 02 ð11Þ
2leff Weff Leff Cox

The input noise spectral density is simulated through HPSICE using 45 nm

process technology and the necessary plot is obtained. From (8), the value of the
coefficient KF can be extracted by substituting the values of the necessary process
and geometry parameters. However, the value of effective mobility leff needs to be
extracted from the HSPICE simulation of the ID - VGS curve, which is a standard
procedure.
Statistical Characterization of Flicker Noise Fluctuation 207

2.2 Theoretical Formulation


Let the process parameter set be p ¼ p1 ; p2 ; . . .; pNp and the mean values of the
process parameters be mp. Expanding the performance parameter e1 around the
mean and assuming that the elements of the set p are independent and normally
distributed gives
Np
  X oe1  
en ðpÞ  en mp þ pi  m p i ð12Þ
i¼1
opi p¼mp

The mean and variance of e1 are defined as follows

 
m en ¼ e n m p ð13Þ

Np
!2
X oen
r2en ¼ r2pi ð14Þ
i¼1
opi p¼mp

oen
In (13) and (15), opi p¼m represents the sensitivity of the performance param-
p

eter with respect to the chosen process parameters. An underlying assumption

behind this is that the variations in the electrical performances are approximated as
linear functions of the process parameters. However, this approximation is often
found to be valid and sufficient to model all of the observed electrical
performances.

2.2.1 BSIM Process Parameter Variability

The BSIM model parameters can be divided into two groups: process parameters
and fitting parameters. The process parameters are directly related to the chosen
process technology. In the present work, the BSIM process parameters that have
been chosen are (1), XL (2) TOXE which is the electrical gate equivalent oxide
thickness and (3) U0 which is the low field carrier mobility.
The variations of XL represent the intra-die variations of the channel length
variability. The variations of TOXE represent the intra-die oxide thickness varia-
tions and the variations of U0 represent the intra-die variations of low field carrier
mobility. The variations of these process parameters are assumed to be Gaussian.
The 3r variations are assumed to be ± 10 % of the technology specified
nominal values. These are based upon ITRS specifications and that available in
literature [2].
The sensitivity of the flicker noise with respect to the parameters XL, TOXE and
U0 are defined from (5)–(14) as follows.
208 S. Sengupta and S. Pandit

 
oen oLeff 2 2
r2enðXLÞ
¼ : :rXL ð15Þ
oLeff oXL
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
oen C u B1 þ B2Leff þ B3L2eff
¼ t n o
oLeff 2 Leff D1  D2 exp  Leff
D3
2 Leff
3
D2
B2 þ 2B3L 1 D3 exp  D3
4 eff 5
  ð16Þ
B1 þ B2Leff þ B3L2eff Leff D1  D2 exp  Leff
D3

oLeff tox 1
From (3) we have, ¼ 1: In (16), C ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi.
oXL eox 2fWeff ls
 
oen ols 2 2
r2enðU0Þ ¼ : :rU0 ð17Þ
ols oU0
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
oen A C1 þ C2ls þ C3l2s C2 þ 2C3ls 1
¼  ð18Þ
ols 2 ls C1 þ C2ls þ C3l2s ls

ols 1
¼ ð19Þ
oU0 1 þ h1 ðVGS  VTH Þ þ h2 ðVGS  VTH Þ2
tox
where, A ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
eox 2fLeff Weff f Leff
 2
oen
r2enðTOXEÞ
¼ :r2TOXE ð20Þ
oTOXE
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  E1 E2

oen E1 E2 B  TOXE 2  2 TOXE3
¼ B: 1 þ þ þ TOXE qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð21Þ
oTOXE TOXE TOXE2 2 1 þ E1 þ E2 TOXE TOXE2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 KF
where, B ¼  .
eox 2fWeff Leff U0f Leff

3 Results and Discussion

The theoretical as well as the numerical HSPICE simulation results that have been
provided in this work are based upon 45 nm CMOS Predictive Technology Model
[11].
In order to extract the values of the flicker noise coefficient KF as function of
the gate voltage VGS we assume Leff = 22.5 nm, Weff = 90 nm and tox = 1.75 nm.
Statistical Characterization of Flicker Noise Fluctuation 209

The necessary spectral density plot for this as obtained from HSPICE simulation
results is shown in Fig. 1. Using the extracted values of KF, we compare the
results obtained from HSPICE simulation and analytical model (8). The coinci-
dence of the simulated and model predicted noise spectral density proves accuracy
of the model. KF is extracted for five different VGS values. It is observed that with
increase of VGS, the value of KF increases which means that the noise also
increases. The flicker noise spectral density depends upon channel length and
carrier mobility. The variations of KF with effective length Leff and surface carrier
mobility ls are obtained from HSPICE simulation. We observe that as the effective
length is reduced, the noise increases so as the value of KF. These two variations
are justified from (1). The functional form of f(Leff) as described in (9) is shown in
Fig. 2. The values of the various unknown parameters of (5)–(7) and (9) are
extracted by curve fitting technique and are summarized in Table 1.
The statistical variability of flicker noise spectral density for an NMOS tran-
sistor due to intra-die variations of channel length, oxide thickness and mobility
variations are studied through Monte Carlo simulations using HSPICE.
The corresponding BSIM parameters, i.e., XL, TOXE and U0 are varied by
± 10 % of the technology specified mean values. We observe that the coefficient
of variations r=l for channel length variation, oxide thickness variation and
mobility variation are 4.21, 2.84 and 0.81 % respectively for fifty Monte Carlo
runs. Considering single MOS transistor, these figures are alarming about the
impact of intra-die process variations on flicker noise variability. This demon-
strates the need of accurate characterization of the various causes of flicker noise
variability. It is also observed that channel length variation is the primary source of
variation regarding flicker noise variability among the three sources of variations
considered in the present work.
In order to validate the theoretical estimations of flicker noise variability with
actual HSPICE-Monte Carlo simulation results, we compare the two. These are
shown in Figs. 3, 4 and 5. It is performed for VGS = 0.7 V. Here the process input

Fig. 1 Flicker noise spectral

density plot used to extract
KF
210 S. Sengupta and S. Pandit

Fig. 2 Functional form of

f(Leff)

Table 1 Values of the various unknown parameters as used in the present work
Parameters Values Parameters Values
A1 5.69e–27 F-A C1 -5.98e–28 F-A
A2 -2.80e–26 F-A/V C2 8.40e–26 C2/m2
A3 3.66e-26 F-A/V2 C3 8.91e–25 C2V sec/m4
B1 1.05e-26 F-A D1 0.2
B2 -3.76e-19 F-A/m D2 7.94
B3 3.73e-12 F-A/m2 D3 3.73e–9

parameters are assumed to follow a Gaussian distribution, and resulting

performance variations are also found to follow the same. One thousand Monte
Carlo runs are performed for the simulation purpose. The values of the mean and
standard deviation as estimated from theoretical model and obtained through

Fig. 3 Comparison of
distribution curves
(theoretical model and Monte
Carlo-HSPICE simulation) of
flicker noise variability with
channel length variations.
VGS = 0.7 V
Statistical Characterization of Flicker Noise Fluctuation 211

Fig. 4 Comparison of
distribution curves
(theoretical model and Monte
Carlo-HSPICE simulation) of
flicker noise variability with
oxide thickness variations.
VGS = 0.7 V

Fig. 5 Comparison of
distribution curves
(theoretical model and Monte
Carlo-HSPICE simulation) of
flicker noise variability with
mobility variations.
VGS = 0.7 V

Table 2 Comparison between mean and standard deviation of flicker noise as obtained from
theoretical model and Monte Carlo-HSPICE simulation results
VGS (V) Parameters Mean m Mean m % Standard Standard deviation %
(model)
 pffiffiffiffiffiffi (Monte error deviation r r
 (Monte carlo) error
pffiffiffiffiffiffi
V= Hz carlo)
 pffiffiffiffiffiffi (model)
 pffiffiffiffiffiffi V= Hz
V= Hz V= Hz
0.7 XL 18.90e–6 19.00e–6 0.53 794.97e–9 835e–9 4.79
TOXE 19.21e–6 19.00e–6 1.11 586.72e–9 586e–9 0.12
U0 18.95e–6 19.00e–6 0.26 153.43e–9 155e–9 1.01

Monte Carlo simulation results are summarized in Table 2. It is observed from all
the comparisons that the theoretical estimations are reasonably accurate, as it is
evident from the % errors reported which has a maximum value of 4.79 % in one
of the cases and in most of the cases it is around 1 % or even less.
212 S. Sengupta and S. Pandit

Fig. 6 Variation of flicker

noise variability due to XL
and TOXE as functions of
effective channel length

Fig. 7 Variation of flicker

noise variability due to XL
TOXE as functions of oxide
thickness

Fig. 8 Variation of flicker

noise variability due to XL,
TOXE and U0 as a function of
gate area
Statistical Characterization of Flicker Noise Fluctuation 213

Figure 6 shows the variations of flicker noise variability due to channel length
variation and oxide thickness variation as function of the channel length. It is
observed from Fig. 6 that higher effective channel length is beneficial so that the
variability is less for both the variations.
Figure 7 shows the flicker noise variability due to XL and TOXE as function of
oxide thickness. It is observed that the flicker noise variability is reduced for lower
oxide thickness in both the cases. This is significant as far as device design is
considered. Thus we can conclude from these graphs that flicker noise variability
due to channel length variation can be minimized by choosing some optimum
combination of TOXE and Leff.
Figure 8 explicitly shows the area dependence of the flicker noise variability. It
is observed that the variability increases with decrease in device sizes. The flicker
noise variability model we derive correctly describes this fact.

4 Conclusion

This paper presents statistical study and characterization of flicker noise variability
for nano-scale NMOS transistor with intra-die channel length variations, oxide
thickness variations and mobility variations. The study has been made theoreti-
cally as well as through HSPICE-based Monte Carlo simulation results. BSIM-
SPICE process parameters have been chosen for characterization purpose. The
theoretical estimates have been compared with MC simulation and good accuracy
has been achieved. The theoretical model thus can be used for accurate charac-
terization of flicker noise variability, which is essential for robust design of analog
and RF circuits in the nano-scale regime.

Acknowledgments The authors thank the Centre for Research in Nanoscience and Nanotech-
nology, University of Calcutta and Department of Science and Technology, Govt. of India (under
Fast Track Young Scientist Scheme) for supporting the research work financially.

References

1. S.K. Saha, Managing process variability in scaled CMOS technology. IEEE Des. Test
Comput. 27(2), 8–16 (2010)
2. S.S. Sapatnekar, Overcoming variations in nanometer-scale technologies. IEEE J. Emerg.
Sel. Top. Circuits Syst. 1(1), 5–18 (2011)
3. K. Kuhn et al., Managing process variations in intel’s 45 nm CMOS technology. Intel
Technol. J. 12(2), 92–110 (2008)
4. D. Sylvestar, K. Agarwal, S. Shah, Variability in nanometer CMOS: impact, analysis, and
minimization. Integr. VLSI J. 41, 319–339 (2008)
5. A. Asenov, S. Kaya, J.H. Davis, Intrinsic threshold voltage fluctuations in decanano
MOSFETs due to local oxide thickness variations. IEEE Trans. Electron Devices 49(1),
112–119 (2002)
214 S. Sengupta and S. Pandit

6. S.K. Saha, Process variability modeling for VLSI circuit simulation. Workshop Compact
Model. NSTI-Nanotechnol. 2, 751–755 (2011)
7. T.H. Morshed, M.V. Dunga, J. Zhang, D.D. Lu, A.M. Niknejad, C. Hu, Compact modeling of
Flicker noise variability in small size MOSFETs’, IEDM, pp. 30.6.1–30.6.4 (2009)
8. M.B. Weissman, 1/f noise and other slow, nonexponential kinetics in condensed matter, Rev.
Mod. Phys., 60(2) (1988)
9. K.K. Hung, P.K. Ko, C. Hu, Y.C. Cheng, A unified model for the Flicker noise in metal-
oxide-semiconductor field-effect transistors. IEEE Trans. Electron Devices 37(3), 654–665
(1990)
10. Y. Nemirovsky, I. Brouk, C.G. Jakobson, 1/f noise in CMOS transistors for analog
applications. IEEE Trans. Electron Devices 48(5), 921–927 (2001)
11. W. Zhao, Y. Cao, New generation of predictive technology model for sub-45 nm early design
exploration. IEEE Trans. Electron Devices 53(11), 2816–2823 (2006)