The State of The Book
The State of The Book
The State of The Book
Vadim Issakov
Vadim Issakov (Vadim.Issakov@ovgu.de) is with the Otto von Guericke University Magdeburg, Germany.
This work was written when he was working at Infineon Technologies, Neubiberg, Germany.
Digital Object Identifier 10.1109/MMM.2019.2941633
Date of current version: 12 November 2019
December 2019
1527-3342/19©2019IEEE 59
advanced nanoscale CMOS nodes even surpass the fT of a number of delay cells (CMOS inverters or simple
and fmax values achieved by silicon germanium (SiGe) differential pairs loaded by resistive loads), designed
heterojunction bipolar transistor technologies [3]. such that they fulfill the Barkhausen criteria to sustain
The voltage-controlled oscillator (VCO) is the heart a stable oscillation [5]. The main advantage of these
of every mm-wave radar or communication system. It circuits is the lack of integrated inductors and, thus,
needs to fulfill very challenging requirements, such as a very compact chip area and wide tunable frequency
low phase noise (PN), and, simultaneously, a wide fre- range. However, ring oscillators have a significantly
quency tuning range (FTR). This is particularly challeng- higher PN than their LC oscillator counterpart.
ing at mm-wave frequencies. The lowest possible PN Therefore, despite many advantages, this VCO type is
is required to achieve the best target discrimination in not usable at mm-wave frequencies. To my knowledge,
radar systems or to enable efficient modulation schemes the fastest operation frequency of an inductorless ring
with large symbol constellations in communication sys- oscillator reported so far is 32 GHz in [6].
tems. Additionally, a wide FTR is required in frequency- The LC oscillator is based on a resonant tank circuit
modulated, continuous-wave radar sensors to achieve consisting of an integrated inductor L and capacitor
the high range resolution needed for many applications, C. The losses of the physical components can be rep-
such as car radar or gesture recognition. Multigigahertz resented by a parallel resistor Rp (Figure 1). To have a
FTR requires the use of large varactors, which degrade sustainable oscillation, the losses of the tank must be
the VCO PN due to the low quality factor they exhibit compensated for by a negative resistance R act, gener-
at mm-waves. This poses a tradeoff between low PN ated by the active circuitry. For example, as shown in
and achievable FTR [4]. In this article, these tradeoffs Figure 1(b), a cross-coupled pair is used as an active
are addressed systematically, with a discussion of the circuit, with the negative resistance seen looking into
technology limitations, and an overview is given of the the drains given by [7]
recently published mm-wave state-of-the-art VCOs.
R act = - 2 , (1)
gm
RF CMOS VCO Basics
There are two main types of commonly used where g m is the transconductance. This implies that, to
integrated CMOS VCOs: RC ring oscillators [Fig- generate more negative resistance, one needs to burn
ure 1(a)] and inductance–capacitance (LC) oscilla- more current. The center oscillation frequency is sim-
tors [Figure 1(b)]. The ring oscillators are composed ply given by
f0 = 1 . (2)
2r LC
By varying the capacitance, one can tune the center
frequency. However, the varactor has a limited qual-
τ τ τ τ
ity factor at mm-wave frequencies. This deteriorates
the quality factor of the resonator and degrades the PN.
(a) Hence, there is an inherent tradeoff between the wide
Vdd FTR and low PN. Nevertheless, the LC VCO is the
most popular type of oscillator in CMOS at mm-wave
L
frequencies due to the low power dissipation and low
Rp Passive achievable PN. The main disadvantage of the LC VCO is
LC Tank that it requires at least one tank inductor, which results
C
in a larger area consumption compared to the ring oscil-
lator. Finally, there are several options for distributed
–Ract oscillators that can be implemented in CMOS at mm-
Vo+ Vo– wave frequencies, such as the rotary traveling wave
[8]. However, these circuits are complex to design and
consume a larger chip area than a single LC VCO. Even
Active though distributed VCOs can achieve excellent per-
Circuit
formance, the following sections focus only on the LC
VCO, since it is the most common approach used in the
lower mm-wave frequency range, such as for commer-
(b) cial applications operating at 60 or 77 GHz. For VCOs
operating at fundamental oscillation frequencies in the
Figure 1. The common types of CMOS oscillators: (a) the upper mm-wave region, toward 300 GHz (approach-
RC ring oscillator and (b) the LC oscillator. ing fmax of the available technologies), both distributed
60 December 2019
standing-wave VCOs [9] and even inductively loaded mm-wave frequencies. An LC VCO usually has more
ring VCOs [10] again become a very interesting and than a single inductive device; therefore, coupling
viable option. between these devices must be carefully modeled in an
electromagnetic field solver. Several inductor-shielding
mm-Wave VCO Design Challenges techniques and “8”-shaped coils can be used to reduce
As opposed to a VCO operating at the classical RFs of a the crosstalk. Their efficiency is analyzed empirically up
few gigahertz, the design of VCOs at mm-wave frequen- to mm-wave frequencies in [21].
cies (per definition, 30–300 GHz) poses several additional Finally, as the operation frequency increases to-
challenges. First, running an LC VCO at the fundamental ward higher mm-wave frequencies, the inductor size
mm-wave frequency has disadvantages, since the qual- decreases almost quadratically [11]. This means that
ity factor of varactors or switched-capacitor banks drops the parasitic inductance of the traces becomes compa-
rapidly with increasing frequency. This leads to PN deg- rable to the values of wanted inductors in the range of a
radation in the thermal region and smaller output power few picohenrys; eventually, it makes more sense to use
[11]. A common solution to this is to operate a VCO at a short transmission lines instead.
lower RF and then increase it to the wanted mm-wave
frequency without introducing additional noise. This is PN of LC VCOs
achieved typically by a frequency multiplier (e.g., qua- The PN of resonator-based VCOs can be described
drupler [12]), subharmonic injection-locking techniques using an empirical expression by Leeson [22]. The PN
[13], extraction of harmonics (e.g., second [4] or third at a frequency offset Δf from a carrier at f0 is given by
harmonic [14], [15]), or any combination of these tech-
2
December 2019 61
Choice of VCO Circuit Topology seen as a series combination of C 1 and C 2. Hence, the
At mm-wave frequencies, a differential circuit topology oscillation frequency is given by
is highly recommended. When deciding to use any of
the three-point VCOs (Clapp, Colpitts, or Hartley) [7], f0 = 1 .(4)
.
one should choose a differential variant of such. The 2r L tank C 1 C 2
C1 + C2
reason for this is a higher immunity to CM interference
of differential signaling compared to the single-ended The negative resistance for these two types, seen in the
counterpart, when applied to distribution of the local terminals of the transistor, is given by
oscillator signal generated by the VCO. Additionally, gm
Rs = - . (5)
the cross-coupled topology is the most popular inte- ^ 2rf0 h2 C 1 C 2
grated CMOS VCO realization. The simplest circuit
realizations are shown in Figure 3. The passive reso- Using the well-known series-to-parallel transformation
nant tank is marked in blue for all three topologies. The and assuming a high value of tank quality factor, one
active parts, which generate the negative equivalent can approximate the equivalent parallel tank resis-
resistance that should compensate for the losses of the tance as R p . R s Q 2tank = R s ^L tank ~ 0 /R s h2 = ^L tank ~ 0 h2 /R s .
resonant tank, are marked in red. Ne x t, by c om bi n i ng t h i s e x pr e s sio n w it h (4)
Unfortunately, Clapp and Colpitts topologies are a nd (5) (note that ~ 0 = 2rf0 ), the start-up condi-
often confused in the literature. The operating fre- tion g m $ R p $ ^C 1 + C 2 h2 /C 1 C 2 = ^C 1 /C 2 h + ^C 2 /C 1 h + 2
quency and start-up condition are equal for these two is obtained [7]. The g m is minimized for C 1 = C 2, and
topologies. The equivalent input capacitance for both is the oscillation condition simplifies to g m $ R p $ 4. For
Vdd Vdd
LD LD Ltank Ltank
C2 C2
Vtune C3
LS LS C2 C2
(a) (b)
Vdd Vdd
Ltank Ltank
Cvar Cvar
Vb
Vo+ Vo– Vo+ Vo–
IB IB CT
(c) (d)
Figure 3. The common types of resonant-tank-based LC VCOs: (a) Clapp, (b) Colpitts, (c) class-B, and (d) ac-coupled cross-
coupled oscillators.
62 December 2019
the cross-coupled oscillator, the start-up condition is negligible by design and might mask the aforemen-
g m $ R p $ 1. This means that one needs one quarter of tioned idealized considerations.
the transconductance g m and, therefore, much less cur- In large signal operation, the Colpitts oscillator exhib-
rent to achieve the oscillation. Alternatively, one can its waveforms similar to class-C operation, whereas the
allow for a lower quality factor of the tank and lower classical cross-coupled VCO [see Figure 3(c)] exhibits a
R p while spending the same current. The more robust class-B-type operation. Idealized current and voltage
oscillation start-up is one of the reasons that cross-cou- waveforms are compared in Figure 4.
pled oscillators are preferred over three-point ones. The drain voltages in both topologies swing about the
Next, for a comparison of VCO topologies, con- CM level of Vdd . In class-B operation [Figure 4(a)], the cur-
sider t he f ig u re of mer it (FoM) a nd FOM T (FoM rent is fully steered into the left or right branch, resulting
including FTR) in a duty cycle (conduction angle) of about 50%. When the
almost-square-wave current with a period of T = 1/f0 is
2
December 2019 63
steady-state oscillation amplitude versus bias current I B the cross-coupled pair, the bias voltage can be set to a
is shown in Figure 5. At lower current levels, the ampli- desired value. In case of class C, the bias voltage VB in
tude is proportional to the bias current, and the oscilla- Figure 3(d) is set quite low, typically to . VDD /3 [25].
tor operates in the current-limited region. When further Additionally, a large capacitor CT is attached in parallel
increasing the current to the point at which the oscilla- to the current source. This causes the drain current to be
tion amplitude increases to almost 2Vdd, the transistors shaped as tall and narrow pulses, as shown in Figure 4(b).
enter the triode region, and the VCO enters the voltage- When Fourier decomposition of this waveform is con-
limited region. As a rule of thumb, using the qualitative sidered, the spectral component of the current at the
description in Figure 5, it is recommended that an ini- fundamental oscillation frequency f0 has the amplitude
tial bias point be placed at the transition between these of I B [26]. Therefore, the current efficiency h I reaches
two regions. Increasing the bias current beyond this 100%. However, class C has a disadvantage in that, at
point would not increase the oscillation amplitude, but large voltage-oscillation amplitudes, the transistors may
it would cause an additional power dissipation without enter the triode region and thus deteriorate the PN.
any advantage [24]. According to [23], the best ENF is achieved by the
Class-C operation can also be achieved using the class-B VCO by using an ac coupling and LC filter at
cross-coupled topology of the configuration shown in the tail node. Since the current efficiency in class B is
Figure 3(d). Thanks to the ac coupling at the gates of lower than 100%, one would need a peak-to-peak oscil-
lation amplitude above 2Vdd to achieve 100% power
efficiency, but this would pose device-reliability con-
cerns. Therefore, 100% power efficiency for class B is
Vosc
only a theoretical value.
2Vdd Another option is to use N-type MOS (NMOS)/
p-type MOS (PMOS) complementary VCO topology
to achieve 100% power efficiency [23]. However, this
approach is limited so far at mm-wave frequencies since
Current Voltage PMOS is slower than NMOS. However, PMOS is catch-
Limited Limited ing up in the newest nanoscale CMOS technologies.
Alternatively, excellent FoM values can be achieved
IB in class-F topology by means of waveform shaping
(Figure 6). In this way, the voltage waveform becomes
Figure 5. The differential amplitude in a class-B cross- more square-like, and the voltage efficiency increases
coupled VCO versus bias current. [24]. However, this is not easily applicable at mm-
wave frequencies because, in addition to the LC tank
resonator at the fundamental frequency f0, we need
Vdd to realize a resonator at 3f0 . Even more challeng-
ing is the condition that, to reduce the PN, the reso-
nance tank at 3f0 should have a higher quality factor
than the tank at fundamental frequency f0 [24]. So
Ltank
f0 far, there are several class-F topologies reported in
Ctank CMOS, mainly in the low-gigahertz RF frequencies
[27]. Fewer VCOs have been reported in the mm-wave
frequency range [28].
3f0 3f0 A systematic comparison between differential
Colpitts and differential cross-coupled LC VCOs in
class-B operation is presented in [29]. First, it shows
Vo+ Vo– that, under conditions of fair comparison (same oscil-
lation frequency, power consumption, tank inductors,
and losses), both topologies achieve the same voltage
ID 1 ID 2 amplitude. Also, the thermal noise contribution from
the lossy tank to the overall PN N L,Rt is equal for both
topologies. Finally, the ratio of PN contributions from
IB
transistors N L,Ids is given by [29]
64 December 2019
where n / C 1 /C 1 + C 2 for Colpitts topology and n opt frequency, at which we can minimize the expression
is equal to 1/3 for c Colpitts = 1. Thus, under the assump- R p /Q 2tank. Consider realizing a VCO at a lower fre-
tion that MOS transistors have the same excess noise quency fosc = f0 /N, where N is the ratio of the output
c Colpitts . c Cross - coupled = 1, the Colpitts VCO has twice frequency (in this example, f0 = 60 GHz) and the fun-
the higher PN contribution in the 1/Δf 2 region, resulting damental oscillation frequency of the VCO (e.g., fosc
in approximately 2-dB-degraded PN. = 15 GHz, if we use a frequency multiplier of N = 4).
The Colpitts VCO needs a higher current (or higher According to (3), reducing the oscillation frequency by
quality factor of the tank) to achieve a robust oscil- N results in a 20 log ^N h PN reduction (improvement).
lation start-up, and it generates a higher PN than its Assuming a noiseless frequency multiplication by
cross-coupled equivalent. The following sections focus N to get to an output frequency of f0 = 60 GHz, the
on the cross-coupled topology, which is the most popu- PN is increased by 20 log ^N h, ideally resulting in the
lar in CMOS. There are numerous options for optimi- same value. This means that there is no theoretical PN
zation of the FoM and absolute PN for this topology, penalty for doing so. On the contrary, considering the
e.g., by a higher class operation, biasing circuitry, and physical implementation, we can physically achieve
coupling of several cores. higher quality factor values of components at lower
frequency and translate this advantage to the desired
Design Considerations for mm-wave output frequency by an almost noiseless
Cross-Coupled LC VCOs frequency multiplication.
This section systematically addresses the design For low PN, one needs to minimize R p and maxi-
aspects of mm-wave cross-coupled LC VCOs in CMOS. mize Q tank. Maximizing Q tank is also necessary for
increasing the maximum possible FoM , as seen in (8).
PN Optimization in the Thermal Both R p and Q tank are technology related and depend
Region 1/Δf 2 on the choice of metal stack and quality factor of varac-
By revisiting Leeson’s equation (3), considering only tors and capacitors.
the thermal noise frequency region for a class-B oscil- One can make another observation about the choice
lator with an ideal current source, and substituting of the tank inductance L value. The parallel resistance
2
Psig = V osc / ^2R p h, (3) may be rewritten in simplified of the tank can be rewritten as
form as
R p = L~ 0 Q tank = Q tank L . (11)
2
December 2019 65
of two scenarios for frequency generation and analysis of Another effect that contributes to PN degradation in
the optimal oscillation frequency was carried out in [16] the 1/∆f 2 region is related to the thermal noise of the
(Figure 8). The authors show that the optimal frequency tail-current source at 2f0, which is directly downcon-
for minimizing the expression R p /Q 2tank for an ultra- verted to PN around f0 . A solution to this is to filter
scaled CMOS technology lies at approximately 5 GHz. the noise at the second harmonic by a large tail capac-
They use the ratio of the inductors’ self-resonance fre- itance C tail . Additionally, a series inductance L tail is
quency (SRF) to the operation frequency for the analysis used to resonate with tail capacitance and parasitic
d = fSRF /fosc . A high value of d indicates a small induc- capacitance to form a high impedance at the second
tor value (and vice versa). As one can see in Figure 9(a), harmonic [32].
the sweet spot for lowest PN is achieved at 5 GHz and
for small inductor values (large d ). On the other hand, PN Optimization in the Flicker
the maximum achievable FoM in (8) depends on Q tank Noise Region 1/Δf 3
(power efficiency h P is fixed by the choice of topology). Flicker noise is modulated onto the oscillation as side-
According to the study in [16], the Q tank tends to increase bands and contributes to the PN close to the carrier. In
for large inductor values, as shown in Figure 9(b). This radar systems operating with very low intermediate fre-
means that optimizing for the lowest PN and highest quencies, the absolute value of PN in the flicker noise
FoM results in different inductance values, and these region is critical. Unfortunately, the MOS transistor has a
conditions cannot be satisfied simultaneously. Addi- very high flicker noise due to interface states and, thus,
tionally, there is a physical limit to increasing quality exhibits very high corner frequencies of several mega-
factor by increasing the inductor size due to the sub- hertz, as opposed to the few kilohertz exhibited by a
strate losses, resulting in optimum inductance value. bipolar transistor [33], making it very difficult to achieve
PN values in CMOS comparable to SiGe realizations.
Rael and Abidi [34] provide an initial description
of two mechanisms describing how the flicker noise
of the tail current source (M T in Figure 10) at low fre-
quency fm upconverts to the oscillation frequency
40 µ m 100 µ m 180 µ m 12
Our Design 10 δ=3
35 Choice 500 δ=5
8
Rp /Qtank (dBΩ)
Parallel Resistance (Ω)
30 6
Q 400 δ = 10
Quality Factor
25 4
2
20 Rp 300 2
15 0
200
–2 δ = 15
10
100 –4
5 0 10 20 30 40
100 200 300 400 500 fOSC (GHz)
Inductance (pH) (a)
20
δ=5
Figure 7. Tank inductor optimization [31].
16
δ = 10
Qtank
12
δ = 15
fLO fOSC fLO 8
Multi-N δ=3
4
0 10 20 30 40
(a) (b) fOSC (GHz)
(b)
Figure 8. The scenarios for generating a mm-wave
frequency analyzed in [16]: (a) a fundamental oscillator at Figure 9. An analysis of the optimal frequency for PN
f LO and (b) an oscillator at lower frequency with frequency minimization [16]: (a) Rp/Q 2tank for different values of d
multiplier. and (b) the quality factor for different values of d.
66 December 2019
f0 and results in PN sidebands f0 ! fm . In the first noise upconversion. This can be achieved either by
mechanism, the switching action of the differential increasing the bias current or using larger devices.
pair commutates the noise of the tail current source Next, we consider the flicker noise contribution of
similarly to a single-balanced mixer. Noise enters the the differential pair transistors M 1, M 2 . Another Gro-
resonator as amplitude modulation (AM). Then, it szkoswski effect takes place due to parasitic device
is converted to frequency modulation (FM) through capacitances, summed as C par in Figure 11(a). This
nonlinear parasitic capacitances of the transistors, capacitance is seen as negative capacitance at the dif-
varactors, and capacitors. One solution to eliminate ferential outputs. Thus, (1) needs to be corrected as
this effect would be to use a small varactor and a large R act = - ^2/g m h - j~aC par [34]. Flicker noise of the core
bank of switched capacitors as the tank capacitance. transistors is modulated onto the current through
Another is to omit a tail-current source completely C par at the second harmonic at zero crossings. After
and use a resistor instead, making the VCO operate in the commutation through M 1, M 2, the flicker noise
the voltage-biased regime [24]. This is why, in most of is downconverted to the oscillation frequency and
our VCO realizations, the tail-current source is omit- presents an effective fluctuating capacitance. Modu-
ted [4], [35]–[38]. lation of the capacitance due to the flicker noise
The second mechanism can be the dominant effect
if the countermeasures are already taken on the first
one, and this is related to the incremental Groszkowski
Vdd
effect [39]. The steady-state oscillation frequency of
a VCO is shifted by Δf from the expected resonance
frequency set by the tank f0 = 1/ ^2r LC h. The reason Ltank
is that the differential pair generates a current with a
rich harmonic content. Higher harmonics flow into
Cvar
the lower-impedance capacitor and upset the reactive
Ract
power balance, as shown in Figure 11(b). However, the
Vo+ Vo–
harmonic content of the current in the tank is a func-
tion of the bias current I B . The shift in frequency as a
Vn2
function of current fluctuations 2f0 /2I B results in an M1 M2
indirect VCO FM by the flicker noise of the tail-current 2f0
source [34]. A very thorough and systematic analysis Cpar
of the 1/f noise upconversion effect by the incremen- Ltail
tal Groszkowski effect is provided by Bevilacqua and
Andreani in [40]. As a solution, the authors suggest that
MT Ctail
more saturated active devices are less prone to flicker
(a)
Vdd Lp Lp
Cd Cd
Ltank Cc Cc
Rp Rp
0
π/2
π/2
Cvar
IH 1
IH 3
IH 2
IH 2
Vo+ Vo– ID 2 ID 1
M1 M2
π
π
IH 3
IH 1
MT (b)
December 2019 67
results i n frequency sh ifts a nd PN i n t he 1/Δf 3 resonance in the tank at 2f0 in addition to the funda-
r e g io n [41]. A solution to this is to suppress the mental resonance at f0 . This causes the current at 2f0 to
current at 2f0 by a tail filter at the second harmonic, flow into the equivalent resistance and the waveform
as shown in Figure 11(a). Additionally, the flicker to become more symmetric.
noise of the transistors M 1, M 2 can be reduced by This technique is taken further by pseudosquaring the
increasing the transistor area, as seen when looking drain-voltage waveform via second-, third-, and fourth-
at the expression for the 1/f noise voltage of the MOS harmonic peaking, that is, realizing a class-F234 VCO
transistor given by V 2n ^ f h = ^K/W $ L h $ ^1/f ch6 V 2 /Hz@ [28]. Additionally, a multiresonant tank proposed in [44]
[7]. However, this creates additional parasitic capac- enables waveform shaping at the second and third har-
itance, reduces the tuning range, and degrades monics and CM-only capacitor tuning.
the maximum oscillation frequency fmax . Another Furthermore, Hu et al. [15] provide the latest find-
solution is to add a differential source capacitor and ings of flicker noise upconversion mechanisms. If one
thereby keep the amplitude of the second harmonic revises the ISF theory, the NMF is modeled as a time-
constant [42]. varying conductance, describing only the effect of
More recent insights into the upconversion mecha- carrier number fluctuation when the carriers are ran-
nism of the flicker noise of transistors M 1, M 2 can domly trapped and released at the interface. Addition-
be gained by using the impulse sensitivity function ally, [15] describes 1/f noise mechanism-correlated
(ISF) theory proposed by Hajimiri and Lee [43]. With mobility fluctuation, which becomes more important
this approach, the flicker noise upconversion can be in the latest CMOS technologies with short chan-
described as follows [15]. First, low-frequency voltage nels. The trapped electrons change the mobility of the
noise at the gate [as shown in Figure 11(a)] is modulated free electrons due to scattering. Finally, Hu et al. [45]
by means of a noise-modulation function (NMF), which prove a very interesting observation: that flicker noise
is typically modeled by a time-varying transconduc- upconversion can be reduced by narrowing the con-
tance. This results in a cyclostationary current noise duction angle, that is, by operating the voltage-based
around the harmonics k~ 0 ! Δ~, which is converted VCO in class C.
into PN through the effective ISF. It can be shown that
upconversion of flicker noise is related to the dc value Advanced mm-Wave VCO Circuit Techniques
of the effective ISF; this can be reduced by making the In this section, we briefly discuss some recent circuit
oscillation waveform more symmetric [41]. The oscilla- techniques to overcome technology limitations at mm-
tion waveform of a voltage-biased VCO tends to have wave VCOs.
asymmetrical rise and fall times due to even-order cur-
rent harmonics flowing into the capacitive part of the PN Reduction by Multiple-Core Coupling
resonant tank, as shown in Figure 11(b). This results in As discussed in previous sections, there is an ultimate
a nonzero dc value of ISF and causes the 1/f noise cur- limit on the PN level that can be achieved by a single
rent of the switching MOS devices to be upconverted VCO core due to low supply and limited quality factor
to the 1/Δf 3 PN region. Therefore, Shahmohammadi of the tank. Thus, the only viable way to further lower
et al. [41] proposed a solution by introducing an auxiliary the PN is by bilateral coupling of several identical cores
(Figure 12). Ideally, this results in improvement of PN
by 10 $ log ^N h [12], [31], [46], where N is the number
of coupled cores. The coupling between the VCO cores
can be done resistively [12], [31], [46] or inductively [38].
However, to realize the PN advantage, the coupling
network must be designed carefully. This technique is
also bounded practically by power consumption and
chip area.
68 December 2019
At the circuit level, one can apply several tech- harmonic while suppressing the fundamental and sec-
niques. The varactor can be “removed” from the ond harmonic.
resonant tank by adding a transformer and tuning
the transformer on the secondary side, as demon- Techniques for Switched-Capacitor Banks
strated in [46] and analyzed in [47]. Additionally, The frequency resolution Δf of a DCO is determined as
Nakamura [48] has proposed using a loop-ground Δf = ^1/2 h^ f0 /C tank h ΔC [19], where f0 = 1/2r LC tank is
transmission line to achieve a wide tuning range. the center frequency, C tank is the total tank capacitance,
Another promising technique is inductive tuning, L is the tank inductance, and ΔC is the switched-capac-
as shown in [49]. itor minimum capacitance. To achieve a resolution of
several kilohertz at mm-wave frequencies, one would
Techniques to Extract the Second need ΔC below attofarads, which is below the physi-
and Third Harmonics cally achievable value of the technology. The work in
As discussed previously, it makes more sense to real- [18] proposes a distributed LC fine-tuning by loading
ize a VCO at a lower frequency instead of the fun- the inductor capacitively along the length. This enables
damental at mm-waves. However, instead of using a a resolution of 160 kHz at 60 GHz. The work in [19]
frequency doubler, one can get the second harmonic proposes a C-2C with exponential steps, achieving a
“for free” by realizing a push–push VCO, since the sec- resolution of 4 Hz for a DCO operating around 60 GHz.
ond harmonic is present in the CM in several nodes in
the circuit [see Figure 11(a)]. Conclusions
In reality, it is not really free because, if a buffer is In this article, we discussed the main challenges
attached at this node, a parasitic capacitance is added and related to the design of mm-wave VCOs in advanced,
might degrade the PN of the VCO. Therefore, a trans- ultrascaled CMOS technologies. Cross-coupled VCOs
former can be used to couple out the second harmonic are most suitable for CMOS realization. It is worth-
at the source node [4], as shown in Figure 13. Alterna- while to realize a VCO at lower frequency to achieve
tively, one can distinguish the fundamental from the a sweet spot in terms of quality factor of tank com-
second harmonic by the mode of signaling. The funda- ponents and then use a push–push configuration and
mental harmonic at f0 is present in the differential mode, frequency multipliers to achieve the needed mm-wave
whereas the second harmonic 2f0 is present in the CM frequency. A summary of the state-of-the-art mm-
and can be coupled out capacitively [50]. Another tech- wave VCOs around 60-GHz published in the last years
nique for extracting a higher frequency (the third har- is given in Table 1. Due to the tradeoff between FTR
monic in this case) is presented in [14] and [15]. By means and PN, a comment regarding how frequency tuning
of a two-stage power amplifier, one can boost the third is achieved has been added.
Coupling Device
C1
f1 f2
f1
Frequency (Hz) eld ZDM
Fi L2 Frequency (Hz)
M1 M2 Inp H
Pout at f1 Is Nonzero
L1 Pout at f2 Is Negligible
EF Pout, ZCM (dBm)
T1 iel L3
Inn d
Vbias Rs Rs
ZCM
at 2f0
Vin1∠0° at f1 Vin1∠180° at f1
Vin2∠0° at f2 Vin2∠0° at f2
f2
Frequency (Hz)
Pout at f1 Is Negligible
Pout at f2 Is Nonzero
(a) (b)
Figure 13. The circuit techniques used to couple out the second harmonic for push–push VCO realization: (a) coupling
inductively at the tail node [6] and (b) coupling from the tank by mode separation [43].
December 2019 69
TABLE 1. The state of the art for CMOS VCOs around 60 GHz.
PN at
CMOS 1 MHz at fosc FOMT Frequency Tuning
Reference Year Node (nm) fosc (GHz) (dBc/Hz) FTR (%) (dBc/Hz) Technique
[4] 2017 40 59.5 −91.8 25 182.28 Continuous
[35] 2018 28 60 –87 16 176 Continuous
[36] 2018 45 SOI 60 –101.7 19 187 Continuous
[37] 2019 45 SOI 63.1 –94.4 16 180 Continuous
[38] 2017 45 SOI 66.4 –87.1 36 186 Continuous
[51] 2013 90 55.6 −91.6 12.5 188 Split in switched modes
[52] 2016 130 57.7 −100.6* 16.8 190.85 Continuous
[14] 2016 40 58 −100 25.4 187.9 Switched capacitors
[53] 2014 65 66.1 −88.4 27.9 177.2 Split in two switched modes
[54] 2016 65 54 −95.5 9.1 179 Split in 16 switched modes
[55] 2013 65 73.8 −86* 41.1 192.2 Split in six switched modes
[56] 2013 65 67.9 −88.9* 22.3 187.4 Split in two switched modes
[57] 2015 65 59.3 −90.3* 39 190.6 Split in two switched modes
[58] 2015 40 58.1 −99.5 14.8 188.9 Split with switched modes
*These values are estimated from figures or extrapolated from data.
70 December 2019
[22] D. B. Leeson, “A simple model of feedback oscillator noise spec- cuits Syst. I, Reg. Papers, vol. 59, no. 5, pp. 938–945, May 2012. doi:
trum,” Proc. IEEE, vol. 54, no. 2, pages pp. 329–330, Feb. 1966. doi: 10.1109/TCSI.2012.2190564.
10.1109/PROC.1966.4682. [41] M. Shahmohammadi, M. Babaie, and R. B. Staszewski, “A 1/f noise
[23] P. Garampazzi et al., “An intuitive analysis of phase noise fun- upconversion reduction technique for voltage-biased RF CMOS
damental limits suitable for benchmarking LC oscillators,” IEEE J. oscillators,” IEEE J. Solid-State Circuits, vol. 51, no. 11, pp. 2610–2624,
Solid-State Circuits, vol. 49, no. 3, pp. 635–645, Mar. 2014. doi: 10.1109/ Nov. 2016. doi: 10.1109/JSSC.2016.2602214.
JSSC.2014.2301760. [42] A. Ismail and A. Abidi, “CMOS differential LC oscillator with
[24] C. Samori, “Understanding phase noise in LC VCOs: A key prob- suppressed upconverted flicker noise,” in Proc. IEEE Int. Solid-State
lem in RF integrated circuits,” IEEE Solid State Circuits Mag., vol. 8, Circuits Conf., 2003, pp. 98–99.
no. 4, pp. 81–91, 2016. doi: 10.1109/MSSC.2016.2573979. [43] A. Hajimiri and T. H. Lee, “A general theory of phase noise in
[25] L. Fanori and P. Andreani, “Highly efficient class-C CMOS electrical oscillators,” IEEE J. Solid-State Circuits, vol. 33, no. 2, pp.
VCOs, including a comparison with class-B VCOs,” IEEE J. Solid- 179–194, Feb. 1998. doi: 10.1109/4.658619.
State Circuits, vol. 48, no. 7, pp. 1730–1740, July 2013. doi: 10.1109/ [44] H. Guo, Y. Chen, P. Mak, and R. P. Martins, “A 0.08-mm2 25.5-
JSSC.2013.2253402. to-29.9 GHz multi-resonant-RLCM-tank VCO using a single-turn
[26] A. Mazzanti and P. Andreani, “Class-C harmonic CMOS multi-tap inductor and CM-only capacitors achieving 191.6 dBc/
VCOs, with a general result on phase noise,” IEEE J. Solid-State Hz FoM and 130 kHz 1/f3 PN corner,” in Proc. IEEE Int. Solid-State
Circuits, vol. 43, no. 12, pp. 2716–2729, Dec. 2008. doi: 10.1109/ Circuits Conf., Feb. 2019, pp. 410–412.
JSSC.2008.2004867. [45] Y. Hu, T. Siriburanon, and R. B. Staszewski, “Intuitive under-
[27] M. Babaie and R. B. Staszewski, “An ultra-low phase noise class-F standing of flicker noise reduction via narrowing of conduction
2 CMOS oscillator with 191 dBc/Hz FoM and long-term reliability,” angle in voltage-biased oscillators,” IEEE Trans. Circuits Syst. II, Ex-
IEEE J. Solid-State Circuits, vol. 50, no. 3, pp. 679–692, Mar. 2015. doi: press Briefs, Jan. 2019. doi: 10.1109/TCSII.2019.2896483.
10.1109/JSSC.2014.2379265. [46] F. Padovan, F. Quadrelli, M. Bassi, M. Tiebout, and A. Bevilac-
[28] H. Guo, Y. Chen, P.-I. Mak, and R. P. Martins, “A 0.083-mm 2 25.2- qua, “A quad-core 15 GHz BiCMOS VCO with −124 dBc/Hz phase
to-29.5 GHz multi-LC-tank class-F234 VCO with a 189.6-dBc/Hz noise at 1MHz offset, −189 dBc/Hz FOM, and robust to multimode
FOM,” IEEE Solid-State Circuits Lett., vol. 1, no. 4, pp. 86–89, Apr. concurrent oscillations,” in Proc. IEEE Int. Solid-State Circuits Conf.,
2018. doi: 10.1109/LSSC.2018.2851499. 2018, pp. 376–378.
[29] P. Andreani, X. Wang, L. Vandi, and A. Fard, “A study of phase [47] A. Bevilacqua, F. P. Pavan, C. Sandner, A. Gerosa, and A. Neviani,
noise in Colpitts and LC-tank CMOS oscillators,” IEEE J. Solid-State “Transformer-based dual-mode voltage-controlled oscillators,”
Circuits, vol. 40, no. 5, pp. 1107–1117, May 2005. IEEE Trans. Circuits Syst. II, Express Briefs, vol. 54, no. 4, pp. 293–297,
[30] D. Murphy, J. Rael, and A. A. Abidi, “Phase noise in LC oscillators: Apr. 2007. doi: 10.1109/TCSII.2006.889734.
A phasor-based analysis of a general result and of loaded Q,” IEEE [48] T. Nakamura, T. Masuda, K. Washio, and H. Kondoh, “A push–push
Trans. Circuits Syst. I, Reg. Papers, vol. 57, no. 6, pp. 1187–1203, June VCO with 13.9-GHz wide tuning range using loop-ground transmis-
2010. doi: 10.1109/TCSI.2009.2030110. sion line for full-band 60-GHz transceiver,” IEEE J. Solid-State Circuits,
[31] S. A.-R. Ahmadi-Mehr, M. Tohidian, and R. B. Staszewski, “Analy- vol. 47, no. 6, pp. 1267–1277, June 2012. doi: 10.1109/JSSC.2012.2187470.
sis and design of a multi-core oscillator for ultra-low phase noise,” [49] P. Chiang, O. Momeni, and P. Heydari, “A 200-GHz inductively
IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 63, no. 4, pp. 529–539, tuned VCO with –7-dBm output power in 130-nm SiGe BiCMOS,”
Apr. 2016. doi: 10.1109/TCSI.2016.2529218. IEEE Trans. Microw. Theory Techn., vol. 61, no. 10, pp. 3666–3673, Oct.
[32] E. Hegazi, H. Sjoland, and A. Abidi, “A filtering technique to low- 2013. doi: 10.1109/TMTT.2013.2279779.
er oscillator phase noise,” in Proc. IEEE Int. Solid-State Circuits Conf., [50] V. Issakov and F. Padovan, “A dual-core 60 GHz push–push VCO
2001, pp. 364–365. with second harmonic extraction by mode separation,” in Proc.
[33] V. Issakov, D. Siprak, M. Tiebout, A. Thiede, W. Simburger, and L. IEEE Radio Frequency Integrated Circuits Symp., 2018, pp. 208–211.
Maurer, “Comparison of 24 GHz receiver front-ends using active [51] P.-L. You and T.-H. Huang, “A switched inductor topology using
and passive mixers in CMOS,” IET Circuits Devices Syst., vol. 3, no. a switchable artificial grounded metal guard ring for wide-FTR
6, pp. 340–349, Dec. 2009. doi: 10.1049/iet-cds.2009.0134. MMW VCO applications,” IEEE Trans. Electron. Devices, vol. 60, no.
[34] J. Rael and A. Abidi, “Physical processes of phase noise in differ- 2, pp. 759–766, 2013. doi: 10.1109/TED.2012.2234750.
ential LC-oscillators,” in Proc. IEEE Custom Integrated Circuits Conf. [52] A. H. M. Shirazi et al., “On the design of mm-wave self-mixing-
(CICC), Orlando, FL, May 2000, pp. 569–572. VCO architecture for high tuning-range and low phase noise,”
[35] V. Issakov, F. Padovan, J. Rimmelspacher, R. Weige, and A. Geisel- IEEE J. Solid-State Circuits, vol. 51, no. 5, pp. 1210–1222, May 2016.
brechtinger, “A 52-to-61 GHz push–push VCO in 28 nm CMOS,” in [53] C.-H. Hung and R. Gharpurey, “A 57-to-75 GHz dual-mode wide-
Proc. IEEE 48th European Microwave Conf., 2018, pp. 1009–1012. band reconfigurable oscillator in 65nm CMOS,” in Proc. IEEE Radio
[36] J. Rimmelspacher, R. Weigel, A. Hagelauer, and V. Issakov, “A Frequency Integrated Circuits Symp., 2014, pp. 261–264.
quad-core 60 GHz push–push 45 nm SOI CMOS VCO with −101.7 [54] T. Xi, S. Guo, P. Gui, D. Huang, Y. Fan, and M. Morgan, “Low-
dBc/Hz phase noise at 1 MHz offset, 19% continuous FTR and −187 phase-noise 54-GHz transformer-coupled quadrature VCO and
dBc/Hz FoMT,” in Proc. IEEE European Solid-State Circuits Conf., 76-/90-GHz VCOs in 65-nm CMOS,” IEEE Trans. Microw. Theory Techn.,
2018, pp. 138–141. vol. 64, no. 7, pp. 2091–2103, 2016. doi: 10.1109/TMTT.2016.2574716.
[37] J. Rimmelspacher, R. Weigel, A. Hagelauer, and V. Issakov, “LC [55] J. Yin and H. C. Huong, “A 57.5-90.1 GHz magnetically-tuned mul-
tank differential inductor-coupled dual-core 60 GHz push–push timode CMOS VCO,” IEEE J. Solid-State Circuits, vol. 48, no. 8, pp.
VCO in 45 nm RF-SOI CMOS technology,” in Proc. IEEE Silicon 1851–1861, Aug. 2013.
Monolithic Integrated Circuits, 2019, pp. 1–3. [56] Y. Chao and H. C. Luong, “Transformer-based dual-band VCO
[38] J. Rimmelspacher, R. Weigel, A. Hagelauer, and V. Issakov, “36% and ILFD for wide-band mm-wave LO generation,” in Proc. IEEE
Frequency-tuning-range dual-core 60 GHz push–push VCO in 45 Custom Integrated Circuits Conf., San Jose, CA, 2013, pp. 47–50.
nm RF-SOI CMOS technology,” in Proc. IEEE Int. Microwave Symp., [57] H. Jia, B. Chi, L. Kuang, and Z. Wang, “A 47.6-71.0-GHz 65-nm
2017, pp. 1356–1358. CMOS VCO based on magnetically-coupled pi-type LC network,”
[39] J. Groszkowski, “The interdependence of frequency variation and IEEE Trans. Microw. Theory Techn., vol. 63, no. 5, pp. 1645–1657, May
harmonic content, and the problem of constant-frequency oscil- 2015. doi: 10.1109/TMTT.2015.2415487.
lators,” Proc. Inst. Radio Eng., vol. 21, no. 7, pp. 958–981, 1934. doi: [58] X. Luo, H. J. Qian, and R. B. Staszewski, “A waveform-shaping
10.1109/JRPROC.1933.227821. millimeter-wave oscillator with 184.7 dBc/Hz FOM in 40-nm digi-
tal CMOS process,” in Proc. IEEE Int. Microwave Symp., 2015, pp. 1–3.
[40] A. Bevilacqua and P. Andreani, “An analysis of 1/f noise to phase
noise conversion in CMOS harmonic oscillators,” IEEE Trans. Cir-
December 2019 71