AIN SHAMS UNIVERSITY FACULTY OF ENGINEERING Electronics and Communications Engineering Department Low Phase Noise VCO Design A Thesis Submitted in partial fulfillment for the requirements of Master of Science degree in Electrical Engineering Submitted by: Ahmed Kamal Abd Elhamid Mahmoud B.Sc. of Electrical Engineering (Electronics and Communications Engineering) Ain Shams University, 1996. Supervised by: Prof. Dr. Hani Fikry Ragaie Dr. Khaled Sharaf Cairo-2003
This is Ahmed Kamal Kassem M.Sc thesis. It is obtained in Electrical Engineering from Ain Shams University, Cairo, Egypt at 2004.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
AIN SHAMS UNIVERSITY
FACULTY OF ENGINEERING Electronics and Communications Engineering Department
Low Phase Noise VCO Design
A Thesis
Submitted in partial fulfillment for the requirements of Master of Science
degree in Electrical Engineering
Submitted by:
Ahmed Kamal Abd Elhamid Mahmoud
B.Sc. of Electrical Engineering
(Electronics and Communications Engineering)
Ain Shams University, 1996.
Supervised by:
Prof. Dr. Hani Fikry Ragaie
Dr. Khaled Sharaf
Cairo-2003
STATEMENT
This dissertation is submitted to Ain Shams University for the degree of
Master of Science in Electrical Engineering, Electronics and
Communications Engineering.
The work in this thesis was carried out by the author at the Electronics and
Communications Engineering Department, Faculty of Engineering, Ain
Shams University.
No part of this thesis was submitted for a degree or a qualification at any
other university or institution.
Date :
Signature :
Name : Ahmed Kamal Abd Elhamid.
Acknowledgment I would like first to thank my supervisors, prof. Hany Fikry Ragai and Dr.
Khaled Sharaf for their very helpful technical support.
I would like also to thank Mentor Graphics Corporation for support by tools
and time and I thank specially Dr. Mohamed Tawfik, Egypt Mentor
Consulting Manager for his technical and time support. I would like also to
thank Mohamed Hassan from Device Modeling team in Mentor Graphics
who helped me in the circuit functionality verification.
2
M.Sc. Thesis
Name : Ahmed Kamal Abd Elhamid.
Thesis Title : Low Phase Noise VCO Design.
Degree Title : Masters of Science.
Department : Electronics & Communications Department- Ain Shams
University.
Referee Committee Name and Position • Prof Dr. AbdelHalim Mahmoud Shousha
Electronics & Communications Dept. Ain Shams University.
• Prof Dr. Hani Fikry Ragaie
Electronics & Communications Dept. Ain Shams University.
• Dr. Khaled Sharaf Electronics & Communications Dept. Ain Shams University.
3
ABSTRACT
The oscillator phase noise is one of the most difficult issues in the design of
modern RF telecommunication systems. One of the most problematic issues
in the low phase noise oscillator design is the upconversion of the low
frequency noise (flicker noise) from the baseband to the oscillation
frequency. The severity of the flicker noise is coming from its large
amplitude compared of other types of noise.
A new LC voltage-controlled oscillator circuit topology is proposed, in
which the phase noise generated by the tail transistor is noticeably reduced
by utilizing the phenomenon of flicker noise intrinsic reduction due to
switched biasing. A macro model of MOSFET under switched biasing is
used to prove the phase noise improvement in the new technique. Circuit
simulations were done on two oscillators with the same tail current value;
one with fixed biasing and the other with the proposed switching.
About 4.5 dB phase noise improvement is achieved at 1kHz frequency offset
in the switched biasing scheme with the same oscillation frequency and
tuning range.
4
SUMMARY
RF VOLTAGE-CONTROLLED oscillators (VCOs) are a vital part of the
RF transceivers, phase-locked loops, clock recovery circuits and frequency
synthesizers and are an important integral part of phase-locked loops, clock
recovery circuits, and frequency synthesizers. The timing accuracy of the
clock recovery circuits and the signal-to-noise ratio where frequency
translation is performed are affected dramatically by the oscillator phase
noise. In particular, it is a must for the RF oscillators employed in wireless
transceivers to be very low phase noise, mandating special attention and
effort in reducing the phase noise.
In the last decade a great effort was paid to the phase noise nature and
modeling. This because the general convince that when designer understand
the nature and the sources of the phase noise, he will be able to avoid it. The
Linear time invariant theory is the theory that most designers rely on in the
past. But it was in need of a lot of modifications and empirical data to be
valid (Leeson’s model). After that the linear time variant model has been
proposed in the research area. It succeeded to explain the behavior of the
phase noise including the cyclostationary behavior of the noise sources in
qualitative and quantitative approaches.
Applying these different theories, lot of researchers was able to minimize the
oscillator phase noise through either inventing new architectures or
optimization of the current architectures. One can say that great efforts were
5
done in the last decade to reduce the oscillator phase noise. These efforts
were in both the architecture-based work and the optimization-based work.
A remaining problem that has not been solved with satisfactory solutions is
the flicker noise upconversion to the oscillation frequency.
Although a great effort was done in the field of Flicker noise modeling, there
is still some lack in the large signal and RF flicker noise issues. One of these
effects is the flicker noise modeling under switched biasing.
This work has proposed a new oscillator circuit that overcomes the problem
of flicker noise upconversion from the tail current source baseband noise to
the oscillation frequency.
6
List of Symbols
µ Carrier mobility (cm2/V.s)
W MOS transistor width (µm)
L MOS transistor length (µm)
COX Capacitance of oxide per unit area (F/µm2)
εo Permitivity of oxide (F/µm)
NA Doping concentration of acceptors (1/µm3)
dOX Oxide thickness in a MOS (µm)
VTH Threshold voltage of a MOS (V)
φF Work function between fermi level and the intrinsic Fermi level
ni Intrinsic doping concentration (1/µm3)
k Boltzmann constant
gDS Conductance between drain to source (mho)
AF Flicker noise exponent
EF Flicker noise frequency exponent
KF Flicker noise coefficient
Q Quality factor
fo Oscillator carrier frequency (Hz)
gm Trans-conductance of the amplifier (mho)
in noise current
L(∆ω) Phase noise value at frequency offset ∆ω from the carrier
Γ(x) Impulse sensitivity function (ISF)
α(x) The noise modulation function (NMF)
7
List of Figures Figure page number Fig. 1.1 The ideal and non-ideal oscillator harmonic 13 Fig. 1.2 The random fluctuations in real oscillator periodic time 14 Fig. 1.3 A typical Heterodyne transceiver 15 Fig. 1.4 The bad effect of phase noise in presence of adjacent strong channel 15 Fig. 1.5 Typical plot of phase noise versus frequency offset 17 Fig. 1.6 An ideal impulse parallel to ideal tank 18 Fig. 1.7 Injected impulse impact at different instances 19 Fig. 1.8 The impulse sensitivity function of a sine wave 20 Fig. 1.9 Phase impulse response of the oscillator system 21 Fig. 1.10 Output voltage relationship with injected current 22 Fig. 1.11 Equivalent system of current voltage transformation 22 Fig. 1.12 Thermal and flicker noise contribution 23 Fig. 1.13 Colpitts Oscillator 26 Fig. 1.14 (a) Output voltage (b) Collector current (c) bipolar shot noise in Colpitts oscillator 27 Fig. 2.1 The tradeoff between power saving, tuning range and phase stability of the oscillator under the same architecture 31 Fig. 2.2 Innovation in architecture can increase all tradeoffs together 32 Fig. 2.3 The voltage-biased negative resistance oscillator 33 Fig. 2.4.The current-biased negative resistance oscillator 34 Fig. 2.5 The oscillator with filtering capacitor 35 Fig. 2.6 The same oscillator with added inductor 36 Fig. 2.7 Adding off-chip inductor and capacitor 37 Fig. 2.8 Complementary versus only-NMOS structures comparison 38 Fig. 2.9 (a) The equivalent circuit of the VCO in Fig. 2.8-b (b) The bias current I(t) versus time 39 Fig. 2.10 The dominant noise sources in oscillator 41 Fig. 2.11 (a) Simplified model for transistor noise sources (b) Differential equivalent circuit 42 Fig. 2.12 The tail current noise transportation 43 Fig. 2.13 Three-dimensional diagram for the on-chip square inductor type 44 Fig. 2.14 The relationship between
22LgL and L 51
8
Fig. 3.1 Drain current noise spectral density of an n-channel MOSFET 53 Fig. 3.2 Gate oxide thickness dependence of flicker noise in n-channel MOSFETs with 0. 15µm and 0.2µm gate channel lengths 59 Fig. 3.3 Gate length dependence of flicker noise in n-channel MOSFETs with various gate oxide thickness 60 Fig. 3.4 Noise reduction as a function of the "off' voltage for an nMOS, VGS_ON = 2.5V, Vth, = 1.9 V, fSWITCH=10 kHz, duty cycle = 50% 61 Fig. 3.5 Two transistors (a) with fixed bias and the (b) with switched bias 62 Fig. 3.6 Noise reduction while switching at different frequencies for an nMOS, VGS = 2.5 V, VGS.off = 0 V, duty cycle = 50%. Also shown is the noise floor under the same conditions 63 Fig. 4.1. N-type negative differential resistance VCO 65 Fig. 4.2. Complementary negative differential resistance VCO 66 Fig. 4.3. An oscillator with reduced tail current flicker noise effect due to filtering 67 Fig. 4.2 The proposed macro-model schematic 69 Fig. 4.3 Comparison of fixed bias case (curve a), switched case with classical model (curve b) and switched case with new macro-model (curves c, d and e) 70 Fig. 4.4 Proposed complementary biased architecture 71 Fig. 4.5 The practical implementation of the proposed complementary biasing (VCO-I) 72 Fig. 4.6 Tail current transistors gate voltages 73 Fig. 4.7 Proposed circuit layout 74 Fig. 4.8 Reference architecture (VCO-II) 75 Fig. 4.9 Waveform comparison of switched and fixed bias schemes 76 Fig. 4.10 Tuning behavior of both oscillators 77 Fig. 4.11 Phase noise comparison at 1kHz offset of three cases 78 Fig. B.1 An ideal LC oscillator, with the maximum voltage amplitude of V0 124
9
Table of Contents Chapter 1 ......................................................................................12 Phase Noise Models...................................................................12
1.1 Introduction. 12 1.2 Phase Noise Description 12 1.3 Bad Effect of the phase noise: timing jitter 15 1.4 Linear Time Invariant (LTI) Model 16 1.5 Linear Time Variant (LTV) theory [3] 18
Chapter 3 ......................................................................................52 Flicker Noise Modeling of MOSFET .....................................52
3.1 Introduction.......................................................................................................... 52 3.2 The Physical Mechanisms of Flicker Noise. 53 3.3 Flicker Noise existing Models. ............................................................................ 56 3.4 Future Work in Flicker Noise Modeling. 58 3.5 Flicker Noise Modeling of MOSFET under Switching Biasing 61
Chapter 4 ......................................................................................64 A New Circuit to Reduce Flicker Noise Effect in VCOs....64
4.1 Introduction. 64 4.2 A new macro-model for transistor noise under switched biasing......................... 67 4.3 Complementary Biasing Technique 70 4.4 Design Considerations 72 4.5 Switched biasing versus constant biasing; a comparison 75
Conclusion and Future Work...................................................79 APPENDIX A: Test Benches and Macro-models.................80
A.1 Main test bench: 80 A.2 Simulation models: 86
APPENDIX B: Calculation of ISF for Ideal LC Oscillator.......................................................................................................124
10
Introduction During the last five decades, there has been tremendous growth in wireless
mobile systems [2]. The implementation of these systems has been made
feasible by scientific and technological advances in the field of integrated
circuits (ICs) [2], allowing a high level of integration at low cost because of
low power dissipation and small integration area. This resulted in a new
oscillator design challenge due to the known tradeoff between the oscillator
power saving and its phase stability [8]. On the other hand, there is a great value for the efficiency of using the
frequency spectrum in wireless communication. One of the most important
factors that affect the efficiency of utilizing the frequency spectrum is the
local oscillators characteristic. This is because local oscillators are used
heavily in both transmitters and receivers. And because of dependence of
signal upconversion and downconversion quality on the local oscillator
timing accuracy, the most important characteristic of the oscillator is the
phase noise. Phase noise dramatically worsens the RF transceivers
performance. For that reason, a deeper understanding of the fundamental
issues limiting the oscillator timing accuracy, and development of design
guidelines to improve it, are necessary.
High speed digital applications performance is also badly impacted by the
oscillator phase noise. The maximum clock rate and hence the maximum
number of operations per unit time is determined by the timing accuracy of
the clock signal [2]. In the huge synchronous digital system such as
microprocessors, the clock is generated through a phase locked loop (PLL).
Introduction
11
This PLL synchronizes the clock to a very accurate external oscillator. The
timing accuracy of the oscillator as a vital part of the PLL impacts the total
timing accuracy.
The known design techniques used for LC oscillator phase noise reduction
are studied in this work and a new oscillator circuit that has low tail current
flicker noise is suggested reducing the total oscillator phase noise.
In Chapter 1, the different basic theories that try to explain the nature and
sources of the phase noise are presented. A special care is given to the linear
time variant model presented in [5].
In Chapter 2, different efforts and design techniques to enhance the phase
noise are shown. The design techniques are divided in this chapter into two
parts. The first is the architecture based design techniques and the other is
optimization based design techniques. In the first part, various architectures
built to improve the phase noise are presented and the short comings of some
of them are shown. In the second part, the methodology of how to enhance
the phase noise given a certain pre-determined architecture is presented.
In Chapter 3, a special care is given to the flicker noise modeling, as it is one
the hardest obstacles in the way of phase noise improvement. The shortage
in the present MOSFET devices models regarding the physical phenomenon
used in the new oscillator circuit is shown.
In Chapter 4, a new LC voltage controlled oscillator is proposed utilizing a
newly discovered physical phenomenon in which the MOSFET flicker noise
is reduced substantially by switched biasing. A new macro model is built to
Introduction
12
overcome the current shortage in MOSFET flicker noise. This macro model
has helped in illustrating how this reduction in flicker noise due to switching
can reduce the total phase noise of the oscillator.
The spurious signals emerged due to switching are easily overcome by
making the switching frequency much larger than the oscillator tank
bandwidth. Finally, the thesis ends by drawing conclusions and suggesting
possible directions for future work.
Chapter1: Phase Noise Models
12
Chapter 1
Phase Noise Models.
1.1 Introduction.
As all types of electronic circuits, the oscillator is greatly affected by both
noise and interferer sources. Thermal, shot and flicker noise of the different
devices are examples for the former sources, while supply and substrate
noise are examples for the latter sources. Those sources affect both
amplitude and phase of the real oscillator, but due to the amplitude limitation
mechanisms in practical oscillators the amplitude noise is suppressed such
that the dominant noise is the phase noise [4].
1.2 Phase Noise Description
1.2.1 Frequency Domain.
The ideal oscillator output oscillating by frequency ωo is perfect periodic
time function
( )ooout tfV ψω += ; ( ) ( )oooo Ttftf ψωψω ++=+ )( and T = periodic time (1-1)
Chapter1: Phase Noise Models
13
accordingly, the Fourier expansion of such function is a series of Dirac
Deltas at nωo since the general expression of ideal periodic waveform is
1( . [ ] . [ ])
2o
out n no o
aV a Cos n t b Sin n tω ω∞
= + +∑
(1-2)
Both the amplitude and frequency of the ideal oscillator are constant, while
the real oscillator amplitude and frequency have statistical nature around
certain average. The non-ideal waveform can be expressed as:
)]([)].(1.[ ttftAVVout oo ψω ++= (1-3)
So the difference between the ideal oscillator harmonics and real oscillator
harmonics is shown in Fig. 1.1.
Fig. 1.1 The ideal and non-ideal oscillator harmonic [3]
The fluctuations in both amplitude and frequency generate the sidebands
shown in the figure. These sidebands are expressing the phase noise
sidebands to a far extent[1]. To be accurate, the phase noise is expressed
only by the sidebands resulted from the phase perturbation frequency
modulation as shown in [4]. To quantify the phase noise, we consider a unit
Chapter1: Phase Noise Models
14
bandwidth at an offset ω∆ from the carrier, calculate the noise power in this
bandwidth, and divide the result by the carrier power, so phase noise can be
written as shown in [1] as:
=∆
powerCarrierbandwidthHzinPowerL
__1__log.10}{ ω (1-4)
1.2.2 Time Domain
The equivalent picture of the phase noise in the time domain is known as
Timing Jitter . In the ideal oscillator, the periodic time is constant with time,
while the real oscillator periodic time has random fluctuations shown in Fig.
(1.2).The above sine wave is an ideal one and the bottom sine wave is the
actual waveform.
Fig. 1.2 The random fluctuations in real oscillator periodic time [3]
Defining the waveform phase as φ(t), then the cycle to cycle phase noise is
proportional to its statistical variance E[φ(t)] [3].
Chapter1: Phase Noise Models
15
1.3 Bad Effect of the phase noise: timing jitter There is a lot of harm that phase noise can cause to different applications.
One of the most important applications that are badly affected by the phase
noise is the heterodyne transceiver Fig. 1.3.
Fig. 1.3 A typical Heterodyne transceiver [1]
Fig. 1.4 demonstrates the bad effect of the noisy local oscillator in the
mentioned transceiver.
Fig. 1.4 The bad effect of phase noise in presence of adjacent strong channel
[1]
Chapter1: Phase Noise Models
16
If strong channel is located near the desired channel, the local oscillator will
transfer both of them in frequency domain. The transferred channels will be
affected by the oscillator noise, hence the noise level of the transferred
adjacent channel may be comparable with the level of the desired channel
resulting in very boor signal to noise ratio [1].
On the other side the timing jitter can badly affect the data handling in the
digital circuits and sampling levels accuracy in the sample and hold circuits.
1.4 Linear Time Invariant (LTI) Model The linear time invariant model of phase noise in oscillators is widely
known as Leeson model and is the most well-known model. The phase noise
according to this model can be written as:
32 1
2{ } 10.log . 1 . 12
o
s L
FkT fLQP
ωωω
ωω ∆ = + + ∆∆
(1-5)
Where F is the oscillator excess noise factor [4]. This number is obtained
empirically, k is Boltzmann’s constant, T is the absolute temperature, Ps is
the tank average power dissipation, and QL is the loaded quality factor of
the tank. Fig. 1.5 is a plot of eqn.(1-4).
Chapter1: Phase Noise Models
17
Fig. 1.5 Typical plot of phase noise versus frequency offset [5]
This graph can be divided to three regions, Region I which is low frequency
offsets region, in which the phase noise slope is -30dB/decade,
-20dB/decade of them is due to noise filtering by the tank and the other
-10dB/decade is due to the upconversion of low frequency noise to the
oscillation frequency. Region II in which the phase noise slope is just
-20dB/decade due to the tank circuit noise filtering, and region III that has
flat noise level. This flattening can be a result of buffer noise, test
instrumentation noise or series resistances of both inductor and capacitor
that put a limit of tank circuit noise filtering [4].
1.4.1 LTI Model Limitations Although the Leeson’s model is directly deduced from experimentation, the
existence of parameter F has led to a serious problem. There is no precise
picture on how F depends on the circuit parameters. One of the most
deceiving results from this model is that the blind application of it leads to
using additional circuits to boost Q of the tank to reduce the phase noise, but
Chapter1: Phase Noise Models
18
sadly, adding these circuits raises the value of F, wasting the phase noise
reduction by Q boosting [4].
1.5 Linear Time Variant (LTV) theory [3] One of the most important steps in the phase noise-modeling trip is the
theory introduced by Thomas Lee and Ali Hajimiri [4], [5]. This new theory
has introduced a robust quantitative model for the phase noise based on new
linear time-variant model to build the relationship between noise and phase
perturbation. There is an important point here that although the oscillator is
mainly non-linear circuit, the relationship between the noise and the
corresponding zero-crossing deviations can be considered linear to a far
extent[5].
1.5.1 Tank voltage Phase change with parallel current Impulse Assume we have an ideal tank circuit in parallel with an impulse current
source Fig. 1.6. The current impulse is located at time τ and has charge ∆q.
Fig. 1.6 An ideal impulse parallel to ideal tank [5]
As the current impulse spectrum is white, the energy of the pulse will be
absorbed almost completely by the capacitor resulting in abrupt change in
tank voltage expressed by:
totalCqV ∆=∆ (1-6)
This abrupt change is illustrated in Fig. 1.7. Top part shows the impulse
effect on the phase when injected at the sine wave peak, it causes no phase
Chapter1: Phase Noise Models
19
difference from the reference sine wave. Down part shows the impulse effect
on the phase when injected at the zero crossings, it causes maximum phase
difference from the reference sine wave.
Fig. 1.7 Injected impulse impact at different instances [5]
It is clear that the waveform is sensitive to current impulse at zero crossings
much more than at peaks, so the sensitivity is dependent on the instance of
injection with respect to the zero crossings. Based on that notice an impulse
sensitivity function (ISF) can be defined as Γ(ωoτ), this function is periodic
function with the same period of the original waveform and dependent on
the time of impulse injection (τ) Fig. 1.8.
Chapter1: Phase Noise Models
20
Fig. 1.8 The impulse sensitivity function of a sine wave [5]
An important approximation can be applied here for the small values of ∆q
is that:
maxVV∆∝∆φ (1-7)
For a certain time instance τ. To apply (1-6) on all time instances we have to
utilize the ISF resulting in:
)(..).().(maxmax
ττωτωφ −∆Γ=∆Γ=∆ tuq
qV
Voo (1-8)
Where
totalCVq .maxmax = (1-9)
The unity step function u (t-τ) is multiplied here because the effect of the
impulse does not occur before the instant τ.
Chapter1: Phase Noise Models
21
1.5.2 Phase Impulse Response
According to the previous analysis, since it was proved that the relationship
between ∆q and ∆φ is linear time-variant, and substituting in (1-8) by ∆q=1,
the oscillator system can be expressed as its impulse response hφ(t,τ).
)(.)(
),(max
ττωτφ −Γ
= tuq
th o (1-10)
Fig. 1.9 illustrates this meaning clearly.
Fig. 1.9 Phase impulse response of the oscillator system [6]
1.5.3 Phase Response for an arbitrary current source
According to the linearity, the superposition principle can be applied to
obtain the phase response to an arbitrary current source i(t) using eqn (1-10)
we obtain:
τττωτττφ φ diq
dithtt
o ).(.)(
).().,()(max
∫∫∞−
∞
∞−
Γ== (1-11)
1.5.4 Oscillator Output Response for an arbitrary current source
The relationship between current and phase is mainly linear time-variant as
seen in (1-8), while the output voltage relationship with the excess phase φ(t)
is non-linear relationship expressed as:
)](cos[)( tttV o φω += (1-12)
The complete process is shown by Fig. 1.10
Chapter1: Phase Noise Models
22
Fig. 1.10 Output voltage relationship with injected current [6]
To have more insight in such process Γ(ωoτ) is analyzed to its Fourier
components:
∑∞
++=Γ1
).cos(.2
)( oono
o ncc θτωτω (1-13)
Substituting in (1-11):
+= ∫ ∑ ∫
∞−
∞
∞−
t t
ono dnicdi
cq
t1max
)cos()()(2
1)( ττωτττφ (1-14)
Combining (1-13) with information in Fig. 1.10 we can obtain Fig. 1.11,
which is another version of Fig. 1.10 in which the effect of each Γ(ωoτ)
harmonic in the total transformation process appears.
Fig. 1.11 Equivalent system of current voltage transformation
Chapter1: Phase Noise Models
23
1.5.5 Single tone current injection
It can be shown according to (1-14) that injecting a single tone sine wave current ( )[ ]tnIti on ωω ∆+= cos)( into the ideal tank output node at offset ∆ω from oscillation frequency multiples nωo creates time varying φ(t) given by:
ωωφ
∆∆
=max2
)sin()(
qtcI
t nn (1-15)
Substituting eqn. (1-15) into (1-12) assuming that )(tφ <<1, the sidebands power relative to carrier power is:
ωωω 22
max
222
max 162
4)(
∆=
∆=∆
qcI
qcI
P nnnndBc
RMS (1-16)
1.5.6 Phase noise due to thermal and flicker noise
According to the last section, the noise sources beside nωo are the main
contributor in the oscillator phase noise. Two types of noise build the phase
noise sidebands, thermal noise and 1/f flicker noise. Thermal noise around
all nωo frequencies affect phase noise, while only low frequency flicker
noise affect phase noise. Fig. 1.12 illustrates this contribution of both
thermal and flicker noise in oscillator phase noise.
Fig. 1.12 Thermal and flicker noise contribution [5]
Chapter1: Phase Noise Models
24
To calculate the noise contribution we have to divide the problem into two
parts thermal noise contribution and flicker noise contribution.
The expressions of device thermal and flicker noise are given respectively
by equations (1-15) and (1-16).
Kf
i thermaln =∆
− 2
_ (1-17)
ωω fflicn K
fi /1
2
ker_ =∆
−
(1-18)
where K is independent on frequency and ω1/f is the flicker noise corner (the
frequency at which the device flicker noise equals device thermal noise).
In the following two sections a detailed description of how both thermal and
flicker noise contribute in the oscillator phase noise is presented.
I-Thermal noise contribution
To calculate the effect of 1Hz bandwidth of thermal noise at offset ∆ω from
the nth harmonic nωo, Equation (1-17) is substituted in equation (1-16), we
have:
ωω 22
max
2
___ 8)(
∆=∆
qKcP n
harmonicnthermaldBc (1-19)
Summing all contributions from all harmonics:
ω
ω 22max
0
2
_ 8)(
∆=∆
∑∞
q
cKP
n
thermaldBc (1-20)
Chapter1: Phase Noise Models
25
Using Parseval’s relation: 22 2 RMSo nc Γ=∑
∞ (1-21)
}4
.log{.10)( 22
max
2
_ ωω
∆Γ
=∆qKP RMS
thermaldBc (1-22)
II-Flicker noise contribution
To calculate the effect of 1Hz bandwidth of flicker noise at frequency ∆ω o,
Equation (1-18) is substituted in equation (1-16), we have:
}8
log{.10)( 22max
2/1
ker;_ ωω
ω
ω∆
∆=∆q
cKP
of
icfdBc (1-23)
Now 3/1 fω can be calculated as the frequency offset at which phase noise due
to flicker noise equals phase noise due to thermal noise. Equating (1-22) and
(1-23) we get:
2/1
2
/1 23
RMS
fof
cΓ
=∆ω
ω (1-24)
This equation is one of the most important keys in the VCO design, because
it tells that it is possible to eliminate the flicker noise effect through making
co=0. This can be made simply by keeping the oscillator waveform
symmetric as possible as we can. This is because when the oscillator
waveforms are symmetric the ISF waveform is symmetric w.r.t. zero
accordingly. This leads to having co=0.
III-Total phase noise From (1-22) and (1-23) we have:
+
∆
∆+
∆Γ
=∆ floornoiseqK
L fRMS ...1.4
.log10}{
3/122
max
2
ωω
ωω (1-25)
Chapter1: Phase Noise Models
26
1.5.7 Cyclostationary Noise Sources In practical oscillators, the statistical properties of some of the random noise
sources may change with time in a periodic manner. These sources are
referred to as cyclostationary. For instance, the channel noise of a MOS
transistor in an oscillator is cyclostationary because the gate-source
overdrive, which varies with time periodically, modulates the noise power.
There may be other noise sources in the circuit whose statistical properties
do not depend on time and the operating point of the circuit. And are
therefore called stationary. Thermal noise of a resistor is an example of a
stationary noise source.
Fig. 1.13 Colpitts Oscillator [3]
These concepts can be understood best in the context of an example.
Consider the Colpitts oscillator of Fig. 1.13. The simulated collector voltage
and current of the transistor are shown in Fig. 1.14 [2]. Note that the
collector current consists of a short period of large current followed a quiet
interval. The power of collector shot noise is proportional to the
instantaneous collector current of the transistor; therefore it has the
maximum power during the peak of collector current.
Chapter1: Phase Noise Models
27
Fig. 1.14 (a) Output voltage (b) Collector current (c) bipolar shot noise in
Colpitts oscillator [3]
Fig. 1.14c shows one sample of collector shot noise of the bipolar transistor.
A white Cyclostationary noise current in(t) can always be decomposed as
)().()( 00 ttiti nn ωα= (1-26)
Where )(0 tin is a white stationary process and )( 0tωα is a deterministic
periodic function describing the noise amplitude modulation and therefore is
referred to as the noise modulating function (NMF). The NMF )( 0tωα is
normalized to a maximum value of 1. This way fin ∆/20 is equal to maximum
of the periodically varying noise power density ftin ∆/)(2 . Applying (1-26) to
(1-11), )(tφ may be rewritten as:
Chapter1: Phase Noise Models
28
∫∫∞−∞−
Γ=
Γ=
t
n
t
n dq
idq
it ττωτωατττωτφmax
000
max
0 )().()(
)()()( (1-27)
As can be seen, Cyclostationary noise can be treated as a stationary noise
applied to a system with a new Impulse sensitivity function (ISF) given by
)().()( xxxNMF αΓ=Γ (1-28)
, where )( 0τωα can be derived easily from device noise characteristics and
the noiseless steady–state waveform. There is a strong correlation between
the Cyclostationary noise source and the waveform of the oscillator. The
maximum of the noise power always recurs at a certain point of the
oscillatory waveform, thus the average of the noise may not be a good
representation of the noise power.
Also note that as the waveform deviates in time from the noiseless
waveform due to phase noise )( 0τωα shifts by exactly the same amount
because the oscillator voltages and currents modulate the noise sources.
Therefore, they will always have a constant phase relationship and (1-25)
will be valid at all times. The relative timing of the cyclostationary noise
sources with respect to the impulse sensitivity function can dramatically
change the effect of those noise sources.
Chapter1: Phase Noise Models
29
The Colpitts oscillator example is considered to provide some design insight
into the effect of cyclostationary noise sources .As a first example consider
the Colpitts oscillator of Fig. 1.13. As can be seen in Fig. 1.14, the surge of
collector current occurs at the minimum of the voltage across the tank,
where the ISF is small. The collector shot noise has its maximum power
when the collector current is maximal, as shown in Fig. 1.14. This fortunate
coincidence lowers the phase noise deterioration due to the bipolar transistor
shot noise, because the maximum noise power always coincides with the
minimum phase noise sensitivity.
Another example is the single ended ring oscillator example, which has the
opposite phenomenon. In ring oscillator the maximum of the ISF function is
at the time of the maximum noise power so the phase noise is deteriorated a
lot.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
31
Chapter 2
Low Phase Noise Design Techniques in LC VCOs
2.1 Introduction It is well known that there is a tradeoff between power saving, tuning range
and phase noise of the voltage controlled oscillators for the same technology
and architecture [7]. This is shown in Fig. 2.1. The gray plane represents
certain architecture under certain technology. The phase noise enhancement
task can be achieved through one of two ways. First is the optimization of
the circuit parameters under fixed architecture. Second, is playing with
Fig. 2.1 The tradeoff between power saving, tuning range and phase stability of the oscillator under the same architecture
Chapter2 Low Phase Noise Design Techniques in LC VCOs
32
architecture innovation to reduce the phase noise. The first approach is in
fact to have the tuning range and power saving above certain values
(determined by the standards/applications) with minimizing the phase noise.
Optimization techniques through computer-aided methods such as
geometrical programming were successfully used [13] for that purpose. The
second way is through moving the tradeoff plane far from the origin as
shown in Fig. 2.2. This enables the designer to obtain improvement in tuning
range, phase stability and power consumption at the same time.
Fig. 2.2 Innovation in architecture can increase all tradeoffs together
Such kind of movement cannot be achieved without changing in the
oscillator architecture. It is shown in Fig. 2.1 that architecture A may have
better phase noise without badly impacting either tuning rage or power
dissipation.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
33
2.2 Architecture Based Improvements It has been shown in a lot of researches that phase noise can be improved by
changing the architecture itself. The N-type negative resistance LC
oscillators different architectures are taken here as an example of how
architecture can play a major role in phase noise enhancement.
2.2.1 Voltage-biased negative resistance oscillators One of the simplest architectures of the negative resistance oscillators is
shown in Fig. 2.3. This architecture has some advantages and some
disadvantages.
Fig. 2.3 The voltage-biased negative resistance oscillator
The advantage of this oscillator is that having low transistor count realizing
potential decrease in phase noise. On other hand, the oscillation amplitude of
this oscillator depends on both voltage supply level and temperature because
the current passing the tank depends on supply voltage and switching
transistors impedance (which depend on temperature variation). Another
drawback is that the current passing to ground has all frequency components
including ωo component. This loads the tank with the switching transistors
Chapter2 Low Phase Noise Design Techniques in LC VCOs
34
while they are in triode regime, subsequently reducing the loaded quality
factor. This reduction increases the phase noise substantially.
While this current bias scheme solves some problems it does create other
problems. The practical implementation of this current source is done by
noisy transistors. This leads to contribution of current source noise in
oscillator phase noise. The thermal noise of the tail current will be mixed
through M1 and M2 to reach the tank. According to LTV phase noise theory
introduced in [5], only the thermal noise beside the multiples of oscillation
frequency ωo will affect the tank phase noise. This effect will be weighted
Chapter2 Low Phase Noise Design Techniques in LC VCOs
35
by the ISF coefficients (c0, c1… cn …). On other hand low frequency noise
(mainly flicker noise) of the tail current also undergoes transfer to the tank.
There are two mechanisms of upconverting the flicker noise of the tail
current to tank nodes. The first is through conveying the noise power
directly to the oscillation frequency fo by the mixing effect made by
switching transistors M1 and M2. The second mechanism is through
upconverting some of the flicker noise power to 2fo by channel length
modulation of the tail current transistor. Then, translating such upconverted
noise back to fo through mixing effect by the negative Gm transistors M1 and
M2.
2.2.3 Negative resistance oscillators with noise filtering techniques
It was shown that adding a large capacitor between the tail current drain and
ground (see Fig. 2.5) short-circuits the noise upconverted at 2ωo. This
protects the oscillator core from that part of tail current flicker noise.
Fig. 2.5 The oscillator with filtering capacitor [9]
Chapter2 Low Phase Noise Design Techniques in LC VCOs
36
It was shown in [11] that the major drawback of this capacitor is passing the
high frequency current (2ω0). This loads the tank (at ω0) by the –ve Gm
while they are in triode. This reduces the effective quality factor of the tank,
losing some of the current source benefits[11].
A modification of the circuit in Fig. 2.5 was made in [11] to overcome its
drawback as shown in Fig. 2.6.
The inductor Lf is added here to resonate with the parasitic capacitance Cp at
frequency 2ωo to solve the quality factor degradation by –ve Gm transistors.
Actually there are two notes on this circuit. The first is that the noise
Fig. 2.6 The same oscillator with added inductor [11]
component that was suppressed is the upconverted part to 2ωo while there is
a part of noise left at low frequency. This part is still harmful to the circuit
since it is upconverted directly to ωo deteriorating the phase noise of the
oscillator. The second note is that the filter does not cover the whole tuning
Chapter2 Low Phase Noise Design Techniques in LC VCOs
37
range since this filter is fixed while the oscillator frequency is tuned. Hence,
the tank circuit immunity of quality factor degradation is not covering all the
tuning range.
A solution for the first note was introduced in [12] by adding off-chip
inductor and capacitor to eliminate most of low frequency noise component
as shown in Fig. 2.7.
Fig. 2.7 Adding off-chip inductor and capacitor [12]
The problem here is that the on-chip solution trend has been violated.
Another solution is suggested in [40] and is fully described in chapter 3.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
38
2.2.4 Complementary NMOS and PMOS Structure
It was shown in [9] that the phase noise of complementary MOS structure
(see Fig. 2.8) shows a better phase noise about 6dB lower than the only-
NMOS architecture. This can be explained by two reasons. First is that the
oscillation amplitude can be approximated by (2/π)*I for architecture (a)
while (4/π)*I for architecture (b).
Fig. 2.8 Complementary versus only-NMOS structures comparison
This means that the oscillation amplitude can be doubled by using
complementary structure, under the same bias current I. Second reason
behind this improvement is that; with the complementary architecture, the
designer can obtain a better waveform symmetry. This symmetry leads to a
symmetry in the ISF as well, hence making c0 = 0. The minimization of c0
minimizes the upconverted low flicker noise. This minimization improves
the total oscillator phase noise [4].
Chapter2 Low Phase Noise Design Techniques in LC VCOs
39
2.3 Optimization-based phase noise improvement Given certain architecture, optimization techniques should be followed to
get the minimum phase noise under certain power dissipation and tuning
range. To do this optimization process, two pieces of information are
required. The first is the noise sources in the circuit and the second is their
share in the total phase noise of the oscillator. The share of this noise sources
in the total oscillator phase noise is significantly affected by oscillation
amplitude.
2.3.1 Tank circuit amplitude calculation and its effect on VCO phase noise
Tank voltage amplitude has an important effect on the phase noise [2], as
emphasized by the presence of qmax in the (1-14). A simple expression for
the tank amplitude can be obtained by assuming that the differential stage
switches quickly from one side to another. For that, the differential pair
oscillator shown in Fig. (2.8-b) can be modeled as current source switching
between Itail and –Itail in parallel with an RLC tank, as shown in Fig. 2.9. Req
is the equivalent parallel resistance of the tank.
Fig. 2.9 (a) The equivalent circuit of the VCO in Fig. 2.8-b (b) The bias
current I(t) versus time [2]
Chapter2 Low Phase Noise Design Techniques in LC VCOs
40
At the frequency of resonance, the admittances of L and C cancel, leaving
Req. Harmonics of I(t) is strongly attenuated by the LC tank. The only
component that remains un-attenuated is the fundamental component. The
waveform in Fig. 2.9 can be approximated as rectangular waveform. It can
be shown that the fundamental frequency current amplitude is given by the
equation:
tailIfundI *4_
=π
(2-1)
This fundamental component will be simply multiplied by Req to give the
oscillation voltage amplitude:
qIoscV tail Re**4_
=π
(2-2)
At high frequencies, the current waveform may be approximated as
sinusoidal wave due to finite switching time and limited gain. In this case,
the tank amplitude may be approximated as:
qIoscV tail Re**4_
=π
(2-3)
This mode of operation is referred as current-limited regime of operation
since in this regime, the tank voltage amplitude is determined mainly by bias
current I and equivalent tank parallel resistance Req.
Equations (2-2) and (2-3) are no longer valid when tank amplitude reaches
supply voltage since the –ve Gm transistors enter triode regime at the peaks
of the voltage. Also the tail NMOS transistor may spend most or even all its
time in the linear region. The tank voltage in this case will be clipped at
VDD by PMOS transistor and at Ground by NMOS transistors. The tail
Chapter2 Low Phase Noise Design Techniques in LC VCOs
41
NMOS transistor helps tank voltage to reach 0V since it will be at triode
region in that case [2].
2.3.2 Noise sources in negative resistance oscillator
As shown in Fig. 2.10 the major noise sources affecting the oscillator phase
noise are transistors noise.
Fig. 2.10 The dominant noise sources in oscillator [2]
The noise power densities of these noise sources are required to calculate the
phase noise. In general, these noise sources are cyclostationary because of
the periodic change in current and voltages of the active devices. A
simplified stationary model for the noise sources is discussed then the true
cyclostationary behavior is considered.
In a simplified stationary approach, the power densities of the noise sources
can be evaluated at most sensitive time (zero crossings of the differential
voltage waveform). To estimate the stationary effect of all sources except
the tail current noise source, Fig. 2.11 is a simple model for the tank circuit
exposed to the differential couple stationary noise sources.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
42
Fig. 2.11 (a) Simplified model for transistor noise sources (b) Differential
equivalent circuit [2]
Fig. 2.11a shows a simplified model of the sources in this balance case.
Fig. 2.11b is the effective noise sources seen by the tank circuit. The total
differential noise power due to the four cross-coupled transistors is:
+=
+++=
___2
___2
___2
2
___21
___22
___21
___2
21
41
pnppnneq iiiiiii
Where (2-4)
==
==
___2
2
___21
___2
___22
___21
___2
pppnnn iiiandiii
And the following expression is valid for both ___
2ni /∆f and
___2pi /∆f thermal
noise components of both short-channel and long-channel approximation
[2]:
( )___
2 / 4n GS Tox
Wi f kT V VLcγµ∆ = − (2-5)
Where µ is the carrier mobility at the channel, Cox is the oxide capacitance
per unit area, W and L are the width and length of the transistor gate,
respectively, VGS is the dc gate source voltage difference and VT is the
transistor threshold voltage. Note that:
3/2≈γ for long channel transistors
Chapter2 Low Phase Noise Design Techniques in LC VCOs
43
And 3~2≈γ for short channel region
2.3.3 Tail current noise source
Tail current noise source impulse sensitivity function (ISF) can be shown [2]
to have 2ω0 oscillation frequency. This is expected, since the tail current
drain is oscillation with this frequency. This is expected because the tail
node is pulled up every time each one of the differential NMOS transistors
turns on and thus the tail node moves at twice the frequency of the
differential voltage. Because of this frequency doubling, only the even ISF
Fourier coefficients (c0 c2 c4 ………) have a value. All the odd coefficients are
0. This makes the noise spectrum beside the odd harmonics has no effect on
the oscillator. The noise spectrum beside 2ω0, 4ω0 ….. has been removed by
filtering techniques introduced by [18] and [18]. But the solution introduced
to eliminate the low frequency noise in [10] has contradicted the on-chip
solution trend nowadays.
Fig. 2.12 The tail current noise transportation [1]
Chapter2 Low Phase Noise Design Techniques in LC VCOs
44
A new technique will be introduced in chapter 4 reducing the tail current
flicker noise without off-chip components.
2.3.4 High Q on-chip inductor design as a key factor of low phase noise
oscillator
It can be shown that the noise source associated with the inductor:
LQkTfiL ω
4/___
2 =∆ (2-6)
So, as one leverages the quality factor of the inductor, the associated noise
decreases for the same oscillation frequency and the same inductance value.
And it is well known that as Q increases the effective filter bandwidth
associated with the tank decreases. This filters the noise around the oscillator
carrier decreasing the phase noise.
Fig. 2.13 Three-dimensional diagram for the on-chip square inductor type
Chapter2 Low Phase Noise Design Techniques in LC VCOs
45
2.3.4.1 High Q inductor design tips
I-Circular geometry
When the inductor lay-out is done as circular as possible it gives good
results. That is because circular geometry has larger perimeter than other
geometries with the same radius. However, increasing the number of sides of
a spiral (Number of sides = ∞ represents circular geometry) increases the
inductor resistance. But, the inductance increases also, because of the
increase in wire length. However, the Q rises because the inductance
increases faster than the resistance. Several simulations were carried out [19]
to verify this rule (Fig. 2.13) at 1.5 GHz with ASITIC [26], considering the
characteristics of a CMOS standard process. The Q is found to rise with the
number of sides.
II- Multi-metal spiral inductors Reducing the resistance per length can increase the inductor quality factor Q.
The task can be accomplished using a thicker track. In conventional CMOS
processes, inductor metal lines can be thickened by shunting metal layers.
As the number of layers shunted becomes higher, the inductor resistance
becomes smaller and the quality factor rises. The number of metal layers
used to make the spiral depends on the CMOS process. Some designers
recommend not using the closest metal layer to the substrate, due to the
increase of the parasitic capacitance to the substrate [27], which reduces the
inductor self-resonant frequency. For example, a spiral of value 3 nH
fabricated using the closest metal layer has a self-resonant frequency of 4.5
GHz [24].For designers working less than 2.4 GHz, it is recommended to
use the metal layer closest to the substrate, because resistance reduction is
more important than resonant frequency reduction [19].
Chapter2 Low Phase Noise Design Techniques in LC VCOs
46
III- Hollow coil Due to the generation of eddy currents at high frequencies, the innermost
turns of the coil suffer from an enormously high resistance and their
contribution to the inductance value is minimal [19]. By including these
turns into the inductor layout, the overall Q is reduced. Therefore, it is
essential to use a hollow coil. A rough percentage of a good inner radius is
approximately equal to the 25 to 40 percent of the outer radius [19].
IV- Metal width and spacing
The spiral traces should be as wide as possible until the skin effect becomes
significant. The increase of the metal width results in a rise of Q because the
resistance of the inductor decreases and the inductance remains the same.
However, due to the skin effect, increasing the metal width much greater
than twice of the skin depth leads to no more decrease in the series
resistance. Thus the increase in metal width is just waste of area and
decreases the self-resonant frequency. The good inductor design fixes the
width between 9 to 14 µm, depending on the design criteria [21]. These
values have been adopted as one of the rules to avoid FEM simulation [21].
The spacing between the metal lines should be as small as possible (just if
this does not increase the inter-winding capacitance significantly).
Increasing the spacing decreases the total inductance because of the
decreasing mutual inductance. It also increases the series resistance and the
total area [21]. If the metal spacing is increased, having the same inductor
area, the Q will increase slightly and the inductance will decrease. Several
simulations were done in [21] to conclude the following rule of thumb; the
Chapter2 Low Phase Noise Design Techniques in LC VCOs
47
quality improves with a minimum spacing and a high width (8 to 12 µm)
[21].
V- Spiral radius
The spiral radius is a very complex parameter to select. As the radius
increases [21], so does the Q. This assumption is valid for small radius. For
values greater than 90 to 100 µm the losses induced by the eddy currents are
heavy, which degrade the quality factor. Another point should be taken into
account when selecting the spiral radius. As the radius increases, the metal
area shared between the spiral and the substrate grows and increases the
parasitic capacitance between the substrate and the spiral. This reduces the
self-resonant frequency of the spiral inductor. Therefore, the radius should
be selected so that the frequency of operation of the spiral is not close to the
resonance. A good design should have a radius under 100 µm [21]. Over 100
µm the parasitic capacitance between the spiral and the substrate and the
induced currents become very high, degrading the Q.
2.3.4.2 Simple Quick Design Rules
1- Do not over increase the line width.
a-This limits the inductance value, assuming constant inductor area.
b-Due to the skin effect, the center of the conductor will not be used.
2- Use the minimum spacing between the adjacent conductors.
a-To enlarge the value of L (increase the turns number without increasing the inductor area).
b-To allow for largest center hole. c-The inter-winding fringing capacitance will be negligible in most cases.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
48
3- Do not fill the inductor up to the center.
a-To avoid the reduction in quality factor due to the eddy currents effect in inner turns. b-To avoid the reduction in inductance value due to the reverse magnetic field produced by eddy currents effect in inner turns.
4- Limit the inductor area.
a-To limit the substrate losses.
b-To Limit the decrease in the inductance value.
2.3.5 Low Phase-Noise Oscillator Design Strategy It was shown in [14] that a robust design strategy can be followed for
negative resistance LC oscillator (Fig. 2.8b).
2.3.5.1 Design Constraints
Assume that the following constrains are imposed on the design:
1- Power dissipation: The maximum power consumption constraint is expressed through the maximum bias current, given certain supply voltage Vsupply
max_biasbias II ≤ (2-7) 2- Tank voltage amplitude: The tank amplitude should be larger than a certain value. This value is determined by the system specifications
min_tantan kk VV ≥ (2-8) And assuming that the oscillator in the current-limited regime, we have:
max_tantan
k
biask g
IV = (2-9)
From (2-8) and (2-9) we have:
min_tanmax_tan
kk
bias Vg
I≥ (2-10)
Chapter2 Low Phase Noise Design Techniques in LC VCOs
49
3- Frequency tuning range: The oscillator is required to have certain tuning range determined by the relationship:
max_00min_0 ωωω ≤≤ (2-11) Or
2max_0
20
2_0
111ωωω
≥≥mn
(2-12)
And we have:
kk CL tantan20
.1 =ω
(2-13)
Then we have the constraint:
2max_0
tantan2_0
1.1ωω
≥≥ kkmn
CL (2-14)
4- startup condition: To guarantee safe startup, the following constraint should be taken on the minimum gactive
max_tanmin . kactive gg α≥ (2-15) Where αmin is the safety factor (a proper value is 3) 5- Diameter of spiral inductors.
According to the total chip area, the inductors should not exceed certain
area. This leads to a maximum diameter dmax:
maxdd < (2-16)
2.3.5.2 Design Equations
It can be shown that [13]:
2tan
2
}{k
bias
VILL ∝∆ω (2-17)
Then, if the oscillator is current limited region we have:
k
biask g
IV
tantan = (2-18)
And for voltage-limited region we have
Chapter2 Low Phase Noise Design Techniques in LC VCOs
50
plyk VV suptan = (2-19)
So the design equations will be
bias
L
IgLL
22
}{ ∝∆ω For current-limited region (2-20)
2sup
2
}{ply
bias
VILL ∝∆ω For voltage-limited region (2-21)
2.3.5.3 Design Strategy
Examining equations (2-20) and (2-21) we find the following observations:
a- Phase noise decrease with Ibias in the current-limited regime while –
surprisingly – increases with Ibias in the voltage-limited region.
b- Phase noise increases with 22LgL in the current-limited regime while
increases with 2L in the voltage-limited region.
From the first observation we can say that we increase Ibias until one of the
following constraints is met; maximum power dissipation or entering the
voltage-limited regime.
The second observation mandates a piece of information to be complete.
How does 22LgL vary with the inductance L?
According to [13], 22LgL has the relationship shown in Fig. 2.13 with L
provided that the gL of the inductor is optimized and the diameter is
constrained by a maximum value.
Chapter2 Low Phase Noise Design Techniques in LC VCOs
51
Fig. 2.14 The relationship between 22LgL and L [13]
Hence, the phase noise is proportional to inductance value L. So it is
recommended to decrease L. But L cannot be decreased without limit.
Decreasing L will decrease the oscillation amplitude. This may violate either
the start-up condition or the amplitude value constraint.
According to the last two observations and their interpretation, the
conclusion is the following design strategy:
Find the minimum inductance that satisfies both the tank amplitude and
startup constraints for the maximum bias current allowed by the design
specifications [13].
Chapter3 Flicker Noise Modeling in MOSFET
52
Chapter 3
Flicker Noise Modeling of MOSFET
3.1 Introduction. Among all noise sources, the flicker noise is the dominant source for phase
noise in silicon MOSFET circuits, especially in the low-frequency-range
[28]. It sets a - lower limit on the level of signal detection and spectral
purity and is one of the factors limiting, the achievable dynamic range of
MOS ICs, so it is important for device and circuit designers to minimize this
effect in order to improve the circuit performance. As designers begin to
explore circuits with low-power and low-voltage MOSFETS, the impact of
low-frequency flicker noise becomes more and more crucial for providing
enough dynamic rang and better circuit performance.
In principle, flicker noise is a low-frequency noise and it mainly affects the
low-frequency performance of the device, so it can be ignored at very high
frequency. However, the contribution of flicker noise should be considered
in designing some radio frequency (RF) circuits such as mixers, oscillators,
or frequency dividers that up-convert the low-frequency noise to higher
frequency and deteriorate the phase noise or the signal to-noise ratio.
Channel resistance and all terminal resistances contribute to the thermal
noise at high frequency (HF), but typically channel resistance dominates in
Chapter3 Flicker Noise Modeling in MOSFET
53
the contributions of the thermal noise from the resistances in the device.
Induced gate noise is generated by the capacitive coupling of local noise
sources within the channel to the gate, and usually it plays a more important
role as the operation frequency goes much higher than the frequency at
which channel thermal noise dominates [28].
3.2 The Physical Mechanisms of Flicker Noise. Noise at low frequencies in a MOSFET is dominated by flicker noise. The
current noise spectral density is roughly inversely proportional to frequency,
as shown in Fig. 3.1. Therefore, flicker noise is also called 1/f noise. Much
effort has been made in understanding the physical origin of flicker noise.
However, the physical mechanism is still not very clear so far. A lot of
discussions and investigations are continuing to find a universal model to
explain the experimental results reported by different research groups that
use devices from different manufacturers [28].
Fig. 3.1 Drain current noise spectral density of an n-channel MOSFET [29].
Chapter3 Flicker Noise Modeling in MOSFET
54
Although there are probably several different physical mechanisms resulting
in noise in MOSFETS, there is a strong indication that traps at the Si-SiO2
interface play the most important role, as discussed in [31]. Electron
trapping and de-trapping can lead to conductance variations. The exact
mechanism is still under discussions however basically; there are three
different theories on the mechanism of flicker noise as follows:
The oscillator with switched biasing shown in Fig. 4.4 can be implemented
by several schemes. The selected scheme is illustrated in Fig. 4.5.
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
72
Fig. 4.5 The practical implementation of the proposed complementary
biasing (VCO-I)
4.4 Design Considerations 1-Biasing oscillator frequency: The frequency of the biasing oscillator (2.24GHz) is selected such that is far from the oscillating frequency fundamental and harmonics of the main oscillator. This is due to two reasons. First is to guarantee no mutual interaction of both oscillators phase noise. Of course, the tuning behavior of the main oscillator must be taken into consideration in this case. And second is to guarantee elimination of any modulation effect due to the biasing oscillator.
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
73
2-Biasing oscillator waveform: The biasing oscillator waveform must be adjusted such that the tail current transistors of the main oscillator switch from the strong inversion regime to depletion regime. Both active transistor dimension of the biasing oscillator and Vbias can be adjusted to achieve this goal. As shown in Fig 4.6, the tail current transistors gate voltages are driven to negative values to guarantees depletion regime entrance.
Fig. 4.6 Tail current transistors gate voltages
3-Layout precautions:
It is essential to layout the switched tail current near the biasing
oscillator as shown in Fig.4.7. This helps in two ways. One is to avoid
have parasitic inductance series with the switched tail transistors gates
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
74
due to routing large distance. The second ways is to avoid collecting
noise on the routes going to the gates.
Fig. 4.7 Proposed circuit layout
Filter resonance capacitor
MOS varactor
Filter noise shorting capacitor
Filter resonance inductor
Switched tail current
Core current path
Main inductor
Biasing oscillator inductor
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
75
4.5 Switched biasing versus constant biasing; a comparison The proposed oscillator shown in Fig. 4.5 (VCO-I) is compared with a
reference oscillator (VCO-II) as shown in Fig 4.8. This reference oscillator
has the same core of the switched bias oscillator but takes its bias from
traditional current mirror. Following are some comparison aspects:
1-Power dissipation:
The current supply of the traditional oscillator, VCO-II, is chosen to
be equal to the core current of VCO-I (I [in both Figs 4.5 and 4.8]
=3mA). This is to obtain the same oscillation frequency and tuning
range for both VCO-I and VCO-II. This power dissipation can by
calculated from the equation: Power dissipation switched scheme = 3.3V*(3mA + 2mA)=16.5mW (4-1) Power dissipation constant bias scheme=3.3V*(3mA )=9.9mW
Fig. 4.8 Reference architecture (VCO-II)
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
76
2-Output waveform:
Fig. 4.9 illustrates a time domain picture of VCO-I and VCO-II output
waveforms respectively.
Fig. 4.9 Waveform comparison of switched and fixed bias schemes
3-Tuning behavior:
Since we have the core of VCO-I and VCO-II the same including Icore
value, the tuning range is expected to be the same. The tuning
characteristics of both oscillators are shown in Fig 4.10.
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
77
Fig. 4.10 Tuning behavior of both oscillators
4-Phase noise:
Finally the phase noise at a low frequency offset (1kHz) is simulated
for three cases. The first is for VCO-II and the second is for VCO-I
and the third is for ideally switched bias oscillator shown in Fig. 4.4.
The result of mentioned simulations is shown in Fig 4.11.
An improvement 4.9dB is noticed between the constant bias phase
noise and ideally switched scheme. This improvement is deteriorated
a little (decreased to 4.6dB) in the practical implementation due to
Chapter4 A New Circuit to Overcome Flicker Noise Effect in VCOs
78
biasing oscillator noise.
Fig. 4.11 Phase noise comparison at 1kHz offset of three cases.
4-Design area:
As shown in Fig 4.7 the switched bias layout is considerably larger
than the classical oscillator with the same core. This is because the
switched scheme has an extra area due to the existence of biasing
oscillator inductors. Moreover these inductors should be separated by
a minimum spacing to reduce coupling between the two oscillators.
Switched design layout area=0.77mm2
Classical design layout area=0.36mm2
Conclusion and Future Work
79
Conclusion and Future Work A complementary switched biasing technique of an LC CMOS VCO tail
current has been proposed. The main idea was to convert the current source
from static current source to two complementary switched current sources.
The total phase noise of such scheme has been improved by a factor up to
4.5dB compared to the classical reported schemes with the same oscillation
frequency and tuning range. The design area of the switched scheme is
nearly double of classical scheme area due to use of two extra inductors. The
switched scheme dissipates extra 6.6mW compared to classical scheme.
Because of the current lack of switched MOSFET flicker noise modeling, a
macro model for switched MOSFET flicker noise has been used in the
oscillator circuit simulation. The macro model has been found to be in
qualitative agreement with the published results on this phenomenon. The
simulation was done using Eldo-RF simulator. The technology used in
simulation is a standard CMOS 0.35µm technology, the total power
dissipation of the new oscillator is 16.5mW, the center frequency is
3.35GHz, the tuning range is 300MHz and its design area = 0.77mm2.
Based on the experience gained in this thesis, several useful future work
trends can be introduced.
• The first is to concentrate on the innovation in oscillator architecture.
Special care can be paid to cyclostationary behavior of noise sources.
• The second is to pay more attention and research to the RF device
modeling.
• The third potential future work is to elaborate more deeply the phase noise
simulation based on zero crossings statistics.
APPENDIX A: Test Benches and Macro-models
80
APPENDIX A: Test Benches and Macro-models.
A.1 Main test bench:
***######### INCLUDES ######################### .include ../custom.mod
+AT =3.300e+04 UTE =-1.300e+00+KT1 =-5.403e-01 KT2 =2.200e-02 KT1L =0.000e+00 +UA1 =0.000e+00 UB1 =0.000e+00 UC1 =0.000e+00+PRT =0.000e+00 * *** Overlap capacitance related and dynamic model parameters ***+CGDO =8.600e-11 CGSO =8.600e-11 CGBO =1.100e-10 +CGDL =1.080e-10 CGSL =1.080e-10 CKAPPA =6.000e-01+CF =0.000e+00 ELM =5.000e+00 +XPART =1.000e+00 CLC =1.000e-15 CLE =6.000e-01* *** Parasitic resistance and capacitance related model parameters ***+RDSW =1.033e+03+CDSC =2.589e-03 CDSCB =2.943e-04 CDSCD =4.370e-04 +PRWB =-9.731e-02 PRWG =1.477e-01 CIT =0.000e+00* *** Process and parameters extraction related model parameters ***+TOX =7.754e-09 NGATE =0.000e+00+NLX =1.770e-07 +XL =0.000e+00 XW =0.000e+00* *** Substrate current related model parameters ***+ALPHA0 =0.000e+00 BETA0 =3.000e+01* *** Noise effect related model parameters ***+AF =AFo KF =KFo EF =EFo+NOIA =NOIAo NOIB =NOIBo NOIC =NOICo* *** Common extrinsic model parameters ***+ALEV =2 RLEV =2+RD =0.000e+00 RS =0.000e+00 RSH =1.290e+02+RDC =0.000e+00 RSC =0.000e+00 LD =-7.130e-08+WD =3.449e-08+LDIF =0.000e+00 HDIF =8.000e-07 WMLT =1.000e+00+LMLT =1.000e+00 DEL =0.000e+00 XJ =3.000e-07+DIOLEV =4 JS =9.000e-05 JSW =0.000e+00+IS =0.000e+00 N =1.000e+00+DCAPLEV=2 CBD =0.000e+00 CBS =0.000e+00+CJ =1.360e-03 CJSW =3.200e-10 FC =0.000e+00+MJ =5.600e-01 MJSW =4.300e-01 TT =0.000e+00+PB =1.020e+00 PBSW =1.020e+00* ----------------------------------------------------------------------.SUBCKT PD A C PARAM: AREA=1e-12 PERI=4e-6* ----------------------------------------------------------------------************************* SIMULATION PARAMETERS ************************* ----------------------------------------------------------------------* format : ELDO, AccusimII, Continuum* model : DIODE* process : C35* revision : 2;* extracted : C64685 ; 2002-12; ese(487)* doc# : ENG-182 REV_2* ----------------------------------------------------------------------* TYPICAL MEAN CONDITION* ----------------------------------------------------------------------* TERMINALS: A=anode=P-region C=cathode=N-region* VARIABLES: M (mulitiplier), AREA [m^2], PERI [m].* NOTE: The role of a protection DIODE is to conduct ESD current to VDD* (or from VSS). This forward bias is NOT modelled, only leakage current* and capacitance during normal operation. Any inductive load etc that* will give forward bias, must be limited by other components to within* Operating Conditions, otherwise parasitic bipolar action can occur.*D1 A C PDINSUB AREA=AREA PERI=PERI.ENDS PD*.MODEL PDINSUB D LEVEL=1 MODTYPE=ELDO+IS =9.000e-05 ISW =0.000e+00 N =1.000e+00 +CJ =1.360e-03 M =5.600e-01 VJ =1.020e+00 TT =0.000e+00
APPENDIX A: Test Benches and Macro-models
101
+CJSW =3.200e-10 MJSW =4.300e-01 FC =0.500e+00 +EG =1.110e+00 XTI =3.000e+00 AF =1.000e+00 KF =0.000e+00* ----------------------------------------------------------------------
.SUBCKT MODNH D G S B PARAM: W=1e-6 L=1e-6 AD=0 AS=0 PD=0 PS=0 NRD=0 NRS=0* VARIABLES: W,L,AD,AS,PD,PS,NRD,NRS = standard MOSFET parameters*M1 D1 G S B MODNHINSUB W=W L=L AD=AD AS=AS PD=PD PS=PS NRD=NRD NRS=NRSRD D1 D {1.328e+03*4.000e-06/(W)} TC=6.200e-03 .ENDS MODNH.MODEL MODNHINSUB NMOS LEVEL=53 MODTYPE=ELDO* ----------------------------------------------------------------------************************* SIMULATION PARAMETERS ************************* ----------------------------------------------------------------------* format : ELDO, AccusimII, Continuum* model : MOS BSIM3v3* process : C35* revision : ;* extracted : C35 B11004.L2; 2002-11; hhl(5481)* doc# : REV_2.0* ----------------------------------------------------------------------* TYPICAL MEAN CONDITION* ----------------------------------------------------------------------*+THMLEV =0.000e+00 * *** Flags ***+MOBMOD =1.000e+00 CAPMOD =2.000e+00 NQSMOD =0.000e+00 +NOIMOD =1.000e+00 DERIV =1.000e+00 * *** Threshold voltage related model parameters ***+K1 =6.2697e-01 +K2 =-4.966e-03 K3 =-2.240e+00 K3B =6.954e-01 +NPEAK =2.236e+17 VTH0 =4.460e-01 +VOFF =-5.090e-02 DVT0 =4.985e+01 DVT1 =1.296e+00 +DVT2 =1.311e-02 KETA =-4.553e-02 +PSCBE1 =1.000e+10 PSCBE2 =1.024e-05 +DVT0W =0.000e+00 DVT1W =0.000e+00 DVT2W =0.000e+00 * *** Mobility related model parameters ***+UA =1.000e-30 UB =1.949e-18 UC =1.217e-10 +U0 =3.427e+02 * *** Subthreshold related parameters ***+DSUB =5.000e-01 ETA0 =3.075e-02 ETAB =-5.261e-02 +NFACTOR=2.034e-01 * *** Saturation related parameters ***+EM =4.100e+07 PCLM =2.940e-01 +PDIBLC1=3.090e-02 PDIBLC2=6.375e-04 DROUT =5.000e-01 +A0 =1.893e-01 A1 =0.000e+00 A2 =1.000e+00 +PVAG =0.000e+00 VSAT =2.402e+05 AGS =1.245e-01 +B0 =6.790e-08 B1 =0.000e+00 DELTA =1.729e-02 +PDIBLCB=2.067e-01 * *** Geometry modulation related parameters ***+W0 =1.145e-07 DLC =6.000e-07 +DWC =2.605e-08 DWB =0.000e+00 DWG =0.000e+00 +LL =0.000e+00 LW =0.000e+00 LWL =0.000e+00 +LLN =1.000e+00 LWN =1.000e+00 WL =0.000e+00 +WW =0.000e+00 WWL =0.000e+00 WLN =1.000e+00 +WWN =1.000e+00 * *** Temperature effect parameters ***+TNOM =2.700e+01 AT =3.300e+04 UTE =-1.800e+00 +KT1 =-3.302e-01 KT2 =2.200e-02 KT1L =0.000e+00 +UA1 =0.000e+00 UB1 =0.000e+00 UC1 =0.000e+00 +PRT =0.000e+00 * *** Overlap capacitance related and dynamic model parameters ***+CGDO =1.200e-10 CGSO =1.200e-10 CGBO =1.100e-10
APPENDIX A: Test Benches and Macro-models
102
+CGDL =0.000e+00 CGSL =0.000e+00 CKAPPA =6.000e-01 +CF =0.000e+00 ELM =5.000e+00 +XPART =1.000e+00 CLC =1.000e-15 CLE =6.000e-01 * *** Parasitic resistance and capacitance related model parameters ***+RDSW =1.092e+03 +CDSC =7.944e-03 CDSCB =0.000e+00 CDSCD =8.448e-05 +PRWB =0.000e+00 PRWG =0.000e+00 CIT =1.000e-03 * *** Process and parameters extraction related model parameters ***+TOX =7.700e-09 NGATE =0.000e+00 +NLX =1.132e-07 +XL =0.000e+00 XW =0.000e+00 * *** Substrate current related model parameters ***+ALPHA0 =0.000e+00 BETA0 =3.000e+01 * *** Noise effect related model parameters ***+AF =1.400e+00 KF =2.810e-27 EF =1.000e+00 +NOIA =1.000e+20 NOIB =5.000e+04 NOIC =-1.400e-12 * *** Common extrinsic model parameters ***+ALEV =2.000e+00 RLEV =2.000e+00 +RD =0.000e+00 RS =0.000e+00 RSH =8.200e+01 +RDC =0.000e+00 RSC =0.000e+00 LD =6.000e-07 +WD =2.605e-08 +LDIF =0.000e+00 HDIF =6.000e-07 WMLT =1.000e+00 +LMLT =1.000e+00 DEL =0.000e+00 XJ =3.000e-07 +DIOLEV =4.000e+00 JS =6.000e-05 JSW =0.000e+00 +IS =0.000e+00 N =1.000e+00 +DCAPLEV=2.000e+00 CBD =0.000e+00 CBS =0.000e+00 +CJ =8.000e-05 CJSW =5.100e-10 FC =0.000e+00 +MJ =3.900e-01 MJSW =2.700e-01 TT =0.000e+00 +PB =5.300e-01 PBSW =6.900e-01 * ----------------------------------------------------------------------
.SUBCKT MODNMH D G S B PARAM: W=1e-6 L=1e-6 AD=0 AS=0 PD=0 PS=0 NRD=0 NRS=0* VARIABLES: W,L,AD,AS,PD,PS,NRD,NRS = standard MOSFET parameters*M1 D1 G S B MODNMHINSUB W=W L=L AD=AD AS=AS PD=PD PS=PS NRD=NRD NRS=NRSRD D1 D {1.547e+03*4.000e-06/(W)} TC=6.200e-03 .ENDS MODNMH.MODEL MODNMHINSUB NMOS LEVEL=53 MODTYPE=ELDO* ----------------------------------------------------------------------************************* SIMULATION PARAMETERS ************************* ----------------------------------------------------------------------* format : ELDO, AccusimII, Continuum* model : MOS BSIM3v3* process : C35* revision : ;* extracted : C35 B11004.L2; 2002-11; hhl(5481)* doc# : REV_2.0* ----------------------------------------------------------------------* TYPICAL MEAN CONDITION* ----------------------------------------------------------------------*+THMLEV =0.000e+00 * *** Flags ***+MOBMOD =1.000e+00 CAPMOD =2.000e+00 NQSMOD =0.000e+00 +NOIMOD =1.000e+00 DERIV =1.000e+00 * *** Threshold voltage related model parameters ***+K1 =9.5409e-01 +K2 =4.9101e-02 K3 =-2.439e+00 K3B =4.077e-01 +NPEAK =2.092e+17 VTH0 =6.449e-01 +VOFF =-4.948e-02 DVT0 =4.985e+01 DVT1 =1.683e+00 +DVT2 =4.126e-02 KETA =-7.397e-02 +PSCBE1 =4.000e+10 PSCBE2 =1.000e-10 +DVT0W =0.000e+00 DVT1W =0.000e+00 DVT2W =0.000e+00
APPENDIX A: Test Benches and Macro-models
103
* *** Mobility related model parameters ***+UA =1.000e-12 UB =3.768e-19 UC =6.391e-12 +U0 =4.394e+02 * *** Subthreshold related parameters ***+DSUB =5.000e-01 ETA0 =1.616e-03 ETAB =-1.373e-02 +NFACTOR=3.455e-01 * *** Saturation related parameters ***+EM =4.100e+07 PCLM =1.055e-01 +PDIBLC1=1.000e-10 PDIBLC2=1.000e-10 DROUT =5.000e-01 +A0 =2.190e-01 A1 =0.000e+00 A2 =1.000e+00 +PVAG =0.000e+00 VSAT =5.129e+04 AGS =9.448e-02 +B0 =-3.629e-08 B1 =0.000e+00 DELTA =3.370e-03 +PDIBLCB=3.872e-01 * *** Geometry modulation related parameters ***+W0 =6.289e-08 DLC =8.917e-08 +DWC =4.938e-08 DWB =0.000e+00 DWG =0.000e+00 +LL =0.000e+00 LW =0.000e+00 LWL =0.000e+00 +LLN =1.000e+00 LWN =1.000e+00 WL =0.000e+00 +WW =0.000e+00 WWL =0.000e+00 WLN =1.000e+00 +WWN =1.000e+00 * *** Temperature effect parameters ***+TNOM =2.700e+01 AT =3.300e+04 UTE =-1.760e+00 +KT1 =-4.502e-01 KT2 =2.200e-02 KT1L =0.000e+00 +UA1 =0.000e+00 UB1 =0.000e+00 UC1 =0.000e+00 +PRT =0.000e+00 * *** Overlap capacitance related and dynamic model parameters ***+CGDO =1.080e-10 CGSO =1.080e-10 CGBO =1.100e-10 +CGDL =0.000e+00 CGSL =0.000e+00 CKAPPA =6.000e-01 +CF =0.000e+00 ELM =5.000e+00 +XPART =1.000e+00 CLC =1.000e-15 CLE =6.000e-01 * *** Parasitic resistance and capacitance related model parameters ***+RDSW =5.304e+02 +CDSC =1.000e-02 CDSCB =0.000e+00 CDSCD =8.448e-05 +PRWB =0.000e+00 PRWG =0.000e+00 CIT =8.122e-04 * *** Process and parameters extraction related model parameters ***+TOX =1.514e-08 NGATE =0.000e+00 +NLX =1.593e-07 +XL =-1.050e-06 XW =0.000e+00 * *** Substrate current related model parameters ***+ALPHA0 =0.000e+00 BETA0 =3.000e+01 * *** Noise effect related model parameters ***+AF =1.400e+00 KF =2.810e-27 EF =1.000e+00 +NOIA =1.000e+20 NOIB =5.000e+04 NOIC =-1.400e-12 * *** Common extrinsic model parameters ***+ALEV =2.000e+00 RLEV =2.000e+00 +RD =0.000e+00 RS =0.000e+00 RSH =7.946e+01 +RDC =0.000e+00 RSC =0.000e+00 LD =8.917e-08 +WD =4.938e-08 +LDIF =0.000e+00 HDIF =6.000e-07 WMLT =1.000e+00 +LMLT =1.000e+00 DEL =0.000e+00 XJ =3.000e-07 +DIOLEV =4.000e+00 JS =6.000e-05 JSW =0.000e+00 +IS =0.000e+00 N =1.000e+00 +DCAPLEV=2.000e+00 CBD =0.000e+00 CBS =0.000e+00 +CJ =8.000e-05 CJSW =5.100e-10 FC =0.000e+00 +MJ =3.900e-01 MJSW =2.700e-01 TT =0.000e+00 +PB =5.300e-01 PBSW =6.900e-01 * ----------------------------------------------------------------------* Owner: austriamicrosystems* HIT-Kit: Digital
ed2 8 0 d 0 1eg2 11 0 g 0 1es2 12 0 s 0 1eb2 10 0 b 0 1
fflicker 0 vn vsen1 1fnonoise vn 0 vsen2 1
rcon vn 0 1 nonoise
.ends
Appendix B: Calculation of ISF for Sinusoidal Wave
124
APPENDIX B: Calculation of ISF for
Ideal LC Oscillator.
Consider the LC oscillator of Fig. B.1.Assuming that the tank has maximum voltage amplitude of V0, the voltage across the capacitor and the current through the inductor can be written as:
Fig. B.1 An ideal LC oscillator, with the maximum voltage amplitude of V0
[2]
)cos()( 0 ttV V ω= (B-1)
)sin()(0
tLCti V ω=
Where L and C are the values of inductor and capacitor, respectively and LC/1=ω is the angular frequency of oscillation.
Fig. B.2. Induced phase and amplitude changes due to a change in the voltage [2]
Appendix B: Calculation of ISF for Sinusoidal Wave
125
If a current impulse with an area of ∆q is injected into the tank at t=t0, it will induce a voltage change of ∆q/C in the capacitor voltage, as shown in Fig. B.2.Therefore, the capacitor voltage at t0
+ is Cqt /)cos( 00 ∆+ων and the inductor current does not change and is )sin(/ 00 tLC ων the capacitor voltage and inductor current after t0 will be sinusoid with a phase shift, φ∆ , and amplitude change, ν∆ , with respect to the initial sinusoid, i.e.…
)cos()()( 0 φωνν ∆+∆+= tt v (B-2)
)sin()()( 0 φωνν ∆+∆+= tLCti
The voltage and current given by (B-2) should be equal to the initial condition at t0
+
Cqttv ∆+=∆+∆+ )cos()cos()( 0000 ωφων ν
(B-3) )sin()sin()( 0000 tt ωφων νν =∆+∆+
Expanding the cosine and sine functions (B-3) can be written as
Appendix B: Calculation of ISF for Sinusoidal Wave
126
Since φ∆ and ν∆ are small, cos 1)( ≈∆φ , sin φφ ∆≈∆ )( and νν 00 ≈∆+v , are valid approximation. Using these approximations, (B-4) can be written as
Cqtt ∆=∆−∆ )sin()cos( 000 ωφων ν
(B-6)
0)cos()sin( 000 =∆+∆ tt ωφων ν Multiply the first and second equations in (B-5) by sin )( 0tω and cos )( 0tω respectively, and subtracting the first from the second, the following is obtained
)sin(0
0tc
q ωφν∆−=∆ (B-7)
And, therefore, the phase impulse response is
)()sin(),( τχωττφ −−= tu
qmath (B-8)
Comparing (B-7) with (4.3), the phase ISF for an ideal oscillator is
)sin()( χχ −=Γ (B-10) 1
Bibliography
127
Bibliography
[1] Behzad Razavi, “A Study of Phase Noise in CMOS Oscillators,” IEEE
Journal Of Solid-State Circuits, vol. 31, no. 3, pp. 331-343, Mar. 1996.
[2] Ali Hajimiri and Thomas H. Lee, The Design of Low Noise
Oscillators. 1999 Kluwer Academic Publishers.
[3] Asad A. Abidi, Analog Circuit Design-RF Analog-to-Digital Con-
verters; Sensor and Actuator Interfaces; Low-Noise Oscillators, PLLs
and Synthesizers. 1997 Kluwer Academic Publishers.
[4] Thomas H. Lee, and Ali Hajimiri, “Oscillator Phase Noise: A
Tutorial,” IEEE Journal Of Solid-State Circuits, vol. 35, no. 3, Mar
2000.
[5] Thomas H. Lee, and Ali Hajimiri, “A General Theory of Phase Noise
in Electrical Oscillators” IEEE Journal Of Solid-State Circuits, vol.
33, no. 2, Feb 1998.
[6] Ali Hajimiri and Thomas H. Lee, “Phase Noise in CMOS Differential
LC Oscillators, ” Symposium on VLSI Circuits, 1998.
[7] Asaad Abidi and A. Kral, “RF-CMOS Oscillators with Switched
Tuning,” in Custom IC Conf. 1998.
[8] Jae Joon Kim and Beomsup Kim, “ A Low Phase Noise CMOS LC
Oscillator with ring structure,” in Solid-State Circuits Conference,
2000, pp. 430 –431.
[9] HongMo Wang, “A Solution for Minimizing Phase Noise in Low-
Power Resonator-Based Oscillators,” ISCAS 2000, IEEE International
Symposium on Circuits and Systems, pp. 53 –56 Geneva Switzerland,
May 2000.
Bibliography
128
[10] Bram De Muer, M. Borremans, M. Steyaert, and G. Li Puma, “A 2-
GHz Low-Phase-Noise Integrated LC-VCO Set with Flicker-Noise
Upconversion Minimization,” IEEE Journal of Solid-State Circuits,
vol. 35 no. 7, pp. 1034 – 1038, July 2000.
[11] Hegazi E., Sjoland. H. and Abidi A., “A filtering technique to lower
oscillator phase noise,” IEEE Journal of Solid-State Circuits, vol. 36
no. 12, pp. 1921-1930, Dec 2001.
[12] Pietro Andreani, and Henrik Sjlِand, “Tail Current Noise Suppression
in RF CMOS VCOs,” IEEE Journal of Solid-State Circuits, vol. 37 no.
3, pp. 342 –348, Mar 2002.
[13] Eric A. M. Klumperink, Sander L. J. Gierkink, Arnoud P. van der Wel
and Bram Nauta, “Reducing MOSFET 1/f Noise and Power
Consumption by Switched Biasing,” IEEE Journal of Solid-State
Circuits, vol. 35 no. 7, pp. 994 –1001, July 2000.
[14] Donhee Ham, and Ali Hajimiri, “Concepts and Methods in
Optimization of Integrated LC VCOs,” IEEE Journal of Solid-State
Circuits, vol. 36 no. 6, pp. 896 –909, June 2001.
[15] C. Samori, A. L. Lacaita, A. Zanchi, S. Levantino and G. Cali, “Phase
Noise Degradation at High Oscillation Amplitudes in LC-Tuned
VCOs,” IEEE Journal of Solid-State Circuits, vol. 35 no. 1, pp. 96 –
99, Jan. 2000.
[16] Maria del Mar Hershenson, Ali Hajimiri, Sunderarajan S. Mohan,
Stephen P. Boyd and Thomas H. Lee, “Design and optimization of LC
oscillators,” in 1999 IEEE/ACM International Conference, 1999 pp. 65
–69.
[17] Jaime Aguilera and Joaquín de Nó, Andrés García-Alonso,
CEIT;Frank Oehler, Heiko Hein and Josef Sauerer, “A Guide for On-
Bibliography
129
Chip Inductor Design in a Conventional CMOS Process for RF
Applications,” Applied Microwave & Wireless, Oct 2001.
[18] J. Craninckx, S.J. Steyaert, “A 1.8 GHz Low-Phase-Noise CMOS
VCO using Optimized Hollow Spiral Inductors,” IEEE Journal of
Solid State Circuits, vol. 32, May 1997.
[19] Y. Hong and M. Frei, “An Approach for Fabricating High-
Performance Inductors on Low-Resistivity Substrate,” IEEE Journal
of Solid State Circuits, vol. 33, Sep 1998.
[20] J. Y. Chang and A. Abidi, “Large Suspended Inductors on Silicon and
their Use in a 2-mm CMOS RF Amplifiers,” IEEE Electron Device
Letters, vol. 14, May 1993.
[21] P. Yue, “On-chip Spiral Inductors with Patterned Ground Shields for
Si-Based RF ICs,” IEEE Journal of Solid-State Circuits, vol. 33, May
1998.
[22] K. Khong, G. Kenneth, “Characteristics of an Integrated Spiral
Inductor with an Underlying N-well,” IEEE Transactions on Electron
Devices, vol. 44, no. 9, Sep 1997.
[23] A. Niknejad and R. Meyer, “Analysis, Design and Optimization of
Spiral Inductors and Transformers for Si RF ICs,” IEEE Journal of
Solid State Circuits, vol. 33, Oct 1998.
[24] A. Kral, “A 2.4 GHz CMOS Frequency Synthesizer,” UCLA, Mar
1998.
[25] N. Nguyen, R. Meyer, “Si IC-Compatible Inductor and LC Passive
Filters,” IEEE Journal of Solid State Circuits, vol. 25, Aug 1990.
[26] J. Burghartz, D.C. Edelstein and M. Soyuer, “RF Circuit Design
Aspects of Spirals Inductors on Silicon,” IEEE Journal of Solid State
Circuits, vol. 33, Dec 1998.
Bibliography
130
[27] T. E. Kolding, “On wafer Calibration Techniques for Giga-Hertz
CMOS Measurements,” Proceedings of IEEE International
Conference on Microelectronic Test Structures (ICMTS), Mar 1999.
[28] Tond Yttedral, Yuhua Cheng and Tor A. Fjeldly, Device Modeling for
Analog and RF CMOS Circuits, John Wiley & Sons Ltd, Chichester,
England.
[29] Hung et al., “a physical-based MOSFET noise model for circuit
simulation,” IEEE Trans Electron Devices, vol.37, pp. 1323-1333,
1990.
[30] Momose H. S. et al., “A study, of flicker noise in n- and p-MOSFETs
with ultra-thin oxide in the direct-tunneling,” Tech. Dig, Int. electron
device meeting, pp. 923-926, 1998.
[31] Jindal R. P. and Ziel A. Van der “phonon fluctuation model for flicker
noise in elemental semiconductor,” J. APPL. Phys. Vol. 52, p. 2884,
1978.
[32] Vandamme L. K. J “1/f noise in MOS devices, mobility or number
fluctuations,” IEEE Trans. Electron Device, vol. 41, pp. 1936-1945,
1994.
[33] Vandamme E. P. et al. “Critical discussion on unified 1/f noise models
for MOSFETs,” IEEE Trans. Electron Device, vol. 47 no. 11, pp.
2146-2153, 2000.
[34] Cheng Y. et al. “BSIM3v3. 1 User's Manual,” Memorandum No.
UCB/ER M97/2, 1997.
[35] Vladimirescu A., The Spice Book, John Wiley & Sons, 1994, New
York.
Bibliography
131
[36] Hung K. K. et al. “A physical-based MOSFET noise model for circuit
simulation,” IEEE Trans. Electron Devices, vol. 37, pp. 1323-1333,
1990.
[37] Cheng Y. and Hu A., MOSFET modeling & BSIM3 User's Guide,
Kluwer Academic publishers, 1999, Norwell, MA.
[38] Wel A. Pvan der et al. “MOSFET 1/f noise measurement under
switched bias condition,” IEEE Electron Device Lett. vol. 21 no. 1, pp.
43-46, 2000.
[39] Klumperink E. A. M. et al “Reducing MOSFET 1/f noise and power
consumption by switched biasing,” IEEE Journal of Solid State
Circuits, vol. 35 no. 7, pp. 994-999, 2000.
[40] Ahmed K. Kassim, Khaled Sharaf, and Hani Ragaie, “Tail Current
Flicker Noise Reduction in LC VCOs by Complementary Switched