Chapter 2

CMOS ACTIVE INDUCTORS
This chapter provides an in-depth treatment of the principles, topologies,
characteristics, and implementation of CMOS active inductors. Section 2.1
investigates the principles of gyrator-C based synthesis of inductors. Both
single-ended and floating (differential) configurations of gyrator-C active in-
ductors are studied. Section 2.2 investigates the most important figure-of-merits
that quantify the performance of active inductors. These figure-of-merits in-
clude frequency range, inductance tunability, quality factor, noise, linearity,
stability, supply voltage sensitivity, parameter sensitivity, signal sensitivity,
and power consumption. Section 2.3 details the CMOS implementation of
single-ended gyrator-C active inductors. The schematics and characteristics of
floating gyrator-C active inductors are examined in detail in Section 2.4. Class
ABactive inductors are investigated in Section 2.5. The chapter is summarized
in Section 2.6.
2.1 Principles of Gyrator-C Active Inductors
2.1.1 Lossless Single-Ended Gyrator-C Active Inductors
A gyrator consists of two back-to-back connected transconductors. When
one port of the gyrator is connected to a capacitor, as shown in Fig.2.1, the
network is called the gyrator-C network. A gyrator-C network is said to be
lossless when both the input and output impedances of the transconductors of
the network are infinite and the transconductances of the transconductors are
constant.
Consider the lossless gyrator-Cnetwork shown in Fig.2.1(a). The admittance
looking into port 2 of the gyrator-C network is given by
22 CMOS Active Inductors
Y =
I
in
V
2
=
1
s
_
C
G
m1
G
m2
_
. (2.1)
Eq.(2.1) indicates that port 2 of the gyrator-Cnetwork behaves as a single-ended
lossless inductor with its inductance given by
L =
C
G
m1
G
m2
. (2.2)
Gyrator-C networks can therefore be used to synthesize inductors. These
synthesized inductors are called gyrator-C active inductors. The inductance
of gyrator-C active inductor is directly proportional to the load capacitance
C and inversely proportional to the product of the transconductances of the
transconductors of the gyrator. Also, the gyrator-C network is inductive over
the entire frequency spectrum. It should also be noted that the transconductor
in the forward path can be configured with a negative transconductance while
the transconductor in the feedback path has a positive transconductance, as
shown in Fig.2.1(b).
Although the transconductors of gyrator-C networks can be configured
in various ways, the constraint that the synthesized inductors should have
a large frequency range, a low level of power consumption, and a small
silicon area requires that these transconductors be configured as simple as
possible. Fig.2.2 shows the simplified schematics of the basic transconduc-
tors that are widely used in the configuration of gyrator-C active inductors.
Common-gate, common-drain, and differential-pair transconductors all have
a positive transconductance while the common-source transconductor has a
negative transconductance. To demonstrate this, consider the common-gate
transconductor. An increase in v
in
will lead to a decrease in i
D
. Because
i
o
= J − i
D
, i
o
will increase accordingly. So the transconductance of the
common-gate transconductor is positive. Similarly, for the differential-pair
transconductor in Fig.2.2(d). An increase in v
in
will result in an increase in
i
D1
. Since i
D2
= J
3
− i
D1
, i
D2
will decrease. Further i
o
= J
2
−i
D2
, i
o
will
increase. The differential-pair transconductor thus has a positive transconduc-
tance.
2.1.2 Lossless Floating Gyrator-C Active Inductors
An inductor is said to be floating if both the terminals of the inductor are
not connected to either the ground or power supply of the circuits containing
the active inductor. Floating gyrator-C active inductors can be constructed in
a similar way as single-ended gyrator-C active inductors by replacing single-
ended transconductors with differentially-configured transconductors, as shown
in Fig.2.3. Because
Principles of Gyrator-C Active Inductors 23
I
G V
m1
V
2
2
G V
m2 1
V
1
C
L
2
1
2
G V
m1
V
2
2
G V
m2 1
V
1
C
2
1
in
I
in
(a)
(b)
V
in
V
in
V
in
I
in
L
2
V
in
I
in
or
Figure 2.1. Lossless singe-ended gyrator-C active inductors. G
m1
and G
m2
are the transcon-
ductances of transconductors 1 and 2, respectively, and C is the load capacitance at node 1. (a)
Transconductor in the forward path has a positive transconductance while the transconductor
in the feedback path has a negative transconductance; (b) Transconductor in the forward path
has a negative transconductance while the transconductor in the feedback path has a positive
transconductance.
V
+
in1
= −
g
m1
sC
(V
+
in2
−V

in2
),
V

in1
=
g
m1
sC
(V
+
in2
−V

in2
),
I
o2
= g
m2
(V
+
in1
−V

in1
),
(2.3)
we have
I
o2
= −
2g
m1
g
m2
sC
(V
+
in2
−V

in2
). (2.4)
The admittance looking into port 2 of the gyrator-C network is given by
24 CMOS Active Inductors
V
b
v
in
i
o
v
in
i
o
v
in
i
o
v
in
i
o
V
b
in
i
o
V
b
v
M1 M2
M1
M2
(a) (b) (c)
(d) (e)
J
J
J
J
3
J
1
J
2
J
Figure 2.2. Simplifiedschematic of basic transconductors. (a) Common-source transconductor
i
o
= −g
m
v
in
; (b) Common-gate transconductor i
o
= g
m
v
in
; (c) Common-drain (source
follower) transconductor i
o
= g
m
v
in
; (d,e) Differential-pair transconductors i
o
= g
m
v
in
.
Y =
I
in
V
+
in2
−V

in2
=
1
s
_
2C
g
m1
g
m2
_
. (2.5)
Eq.(2.5) reveals port 2 of the gyrator-C network behaves as a floating inductor
with its inductance given by
L =
2C
g
m1
g
m2
. (2.6)
Principles of Gyrator-C Active Inductors 25
C
G V - V ( )
m1
2+ 2-
G V - V ( )
m2 1+ 1-
L
C
2-
2+
1+
1-
2+
2-
I
in
V
in
V
in
I
o2
I
o1
Figure 2.3. Lossless floating gyrator-C active inductors. G
m1
and G
m2
are the transconduc-
tances of transconductors 1 and 2, respectively, and C is the load capacitance at nodes 1+ and
1-.
Floating gyrator-C active inductors offer the following attractive advantages
over their single-ended counterparts : (i) The differential configuration of
the transconductors effectively rejects the common-mode disturbances of the
network, making themparticularly attractive for applications where both analog
and digital circuits are fabricated on the same substrate. (ii) The level of the
voltage swing of floating active inductors is twice that of the corresponding
single-ended active inductors.
2.1.3 Lossy Single-Ended Gyrator-C Active Inductors
When either the input or the output impedances of the transconductors of
gyrator-C networks are finite, the synthesized inductors are no longer lossless.
Also, the gyrator-C networks are inductive only in a specific frequency range.
Consider the gyrator-C network shown in Fig.2.4 where G
o1
and G
o2
denote
the total conductances at nodes 1 and 2, respectively. Note G
o1
is due to the
finite output impedance of transconductor 1 and the finite input impedance
of transconductor 2. To simplify analysis, we continue to assume that the
transconductances of the transconductors are constant. Write KCL at nodes 1
and 2
(sC
1
+G
o1
)V
1
−G
m1
V
2
= 0 (node 1),
−I
in
+ (sC
2
+G
o2
)V
2
−G
m2
(−V
1
) = 0, (node 2).
(2.7)
The admittance looking into port 2 of the gyrator-C network is obtained from
26 CMOS Active Inductors
Y =
I
in
V
2
= sC
2
+G
o2
+
1
s
_
C
1
G
m1
G
m2
_
+
G
o1
G
m1
G
m2
. (2.8)
Eq.(2.8) can be represented by the RLC networks shown in Fig.2.4 with its
parameters given by
R
p
=
1
G
o2
,
C
p
= C
2
,
R
s
=
G
o1
G
m1
G
m2
,
L =
C
1
G
m1
G
m2
.
(2.9)
G V
m1
C R
R
L
p p
s
G
o2
C
2
V
2
2
G V
1
V
1
G
o1
C
1
m2
2
1
2
I
in
V
in
I
in
V
in
C R
R
L
p p
s
2 I
in
V
in
Figure 2.4. lossy single-ended gyrator-C active inductors. C
1
and G
o1
, C
2
and G
o2
denote
the total capacitances and conductances at nodes 1 and 2, respectively.
We comments on the preceding results :
Principles of Gyrator-C Active Inductors 27
When the input and output conductances of the transconductors are consid-
ered, the gyrator-C network behaves as a lossy inductor with its parasitic
parallel resistance R
p
, parallel capacitance C
p
, and series resistance R
s
.
R
p
should be maximized while R
s
should be minimized to low the ohmic
loss. The finite input and output impedances of the transconductors of the
gyrator-C network, however, have no effect on the inductance of the active
inductor.
R
p
and C
p
are solely due to G
o2
and C
2
. G
o1
and C
1
only affect R
s
and L.
The resonant frequency of the RLC networks of the active inductor is given
by
ω
o
=
1
LC
p
=
¸
G
m1
C
1
G
m2
C
2
=

ω
t1
ω
t2
, (2.10)
where
ω
t1,2
=
G
m1,2
C
1,2
(2.11)
is the cut-off frequency of the transconductors. ω
o
is the self-resonant
frequency of the gyrator-C active inductor. This self-resonant frequency is
typically the maximumfrequency at which the active inductor operates. The
self-resonant frequency of an active inductor is set by the cut-off frequency
of the transconductors constituting the active inductor.
The small-signal behavior of a gyrator-C active inductor is fully character-
ized by its RLCequivalent circuit. The RLCequivalent circuit of gyrator-C
active inductors, however, can not be used to quantify the large-signal behav-
ior, such as the dependence of the inductance on the dc biasing condition of
the transconductors and the maximum signal swing of the gyrator-C active
inductors.
When the conductances encountered at nodes 1 and 2 of the gyrator-C active
inductors are zero (lossless), the phase of the impedance of the synthesized
inductor is
π
2
. However, when these conductances are non-zero, the phase
of the impedance of the synthesized inductor will deviate from
π
2
, giving
rise to a phase error. The phase error is due to R
p
and R
s
of the active
inductors. The phase of the impedance of practical active inductors should
be made constant and to be as close as possible to
π
2
.
28 CMOS Active Inductors
The finite input and output impedances of the transconductors constituting
active inductors result in a finite quality factor. For applications such as
band-pass filters, active inductors with a large quality factor are mandatory.
In these cases, Q-enhancement techniques that can offset the detrimental
effect of R
p
and R
s
should be employed to boost the quality factor of the
active inductors.
2.1.4 Lossy Floating Gyrator-C Active Inductors
Lossy floating gyrator-C active inductors can be analyzed in a similar way
as lossy single-ended gyrator-C active inductors. Consider the lossy floating
gyrator-C network shown in Fig.2.5. We continue to assume that the transcon-
ductances of the transconductors are constant. Writing KCL at nodes 1-, 1+,
2-, and 2+ yields
−G
m1
(V
+
2
−V

2
) +
_
sC
1
+G
o1
2
_
(V

1
−V
+
1
) = 0,
I
in
+
_
sC
2
+G
o2
2
_
(V

2
−V
+
2
_
+G
m2
(V
+
1
−V

1
) = 0,
(2.12)
The admittance looking into port 2 of the gyrator-C network is obtained from
Y =
I
in
V
+
2
−V

2
= s
C
2
2
+
G
o2
2
+
1
s
_
C
1
2G
m1
G
m2
_
+
G
o1
2G
m1
G
m2
. (2.13)
Eq.(2.12) can be represented by the RLC network shown in Fig.2.5 with its
parameters given by
R
p
=
2
G
o2
,
C
p
=
C
2
2
,
R
s
=
G
o1
/2
G
m1
G
m2
,
L =
C
1
/2
G
m1
G
m2
.
(2.14)
Characterization of Active Inductors 29
The constant 2 in (2.14) is due to the floating configuration of the active
inductor. The capacitance and conductance at the interface nodes 1+ and 1−
and those at the internal nodes 2+ and 2− will become C
2
/2 and G
o2
/2, and
C
1
/2 and G
o21
/2, respectively. Eq.(2.9) can be used to derive (2.14) without
modification.
G V - V ( )
m1
C R
R
L
p p
s
G
o2
C
2
V
2
2
V
1
2-
1-
2+
G
o2
C
2
G
o1
C
1
G
o1
C
1
1+
2+
2
- +
G V - V ( )
m2
1 1
- +
2-
G
o2
C
2
G
o1
C
1
G
o1
C
1
C
2
G
o2
I
in
V
in
I
in
V
in
Figure 2.5. Lossy floating gyrator-C active inductors. C
1
and G
o1
, C
2
and G
o2
represent the
total capacitances and conductances at nodes 1 and 2, respectively.
2.2 Characterization of Active Inductors
In this section, we investigate the most important figure-of-merits that pro-
vide quantitative measures of the performance of active inductors. These
figure-of-merits include frequency range, inductance tunability, quality factor,
noise, linearity, stability, supply voltage sensitivity, parameter sensitivity, signal
sensitivity, and power consumption.
2.2.1 Frequency Range
It was shown in the preceding section that an lossless gyrator-C active induc-
tor exhibits an inductive characteristic across the entire frequency spectrum.
A lossy gyrator-C active inductor, however, only exhibits an inductive charac-
teristic over a specific frequency range. This frequency range can be obtained
by examining the impedance of the RLC equivalent circuit of the lossy active
inductor
30 CMOS Active Inductors
Z =
_
R
s
C
p
L
_
s
L
R
s
+ 1
s
2
+s
_
1
R
p
C
p
+
R
s
L
_
+
R
p
+R
s
R
p
C
p
L
. (2.15)
When complex conjugate poles are encountered, the pole resonant frequency
of Z is given by
ω
p
=
¸
R
p
+R
s
R
p
C
p
L
. (2.16)
Because R
p
R
s
, Eq.(2.16) is simplified to
ω
p

¸
1
LC
p
= ω
o
, (2.17)
where ω
o
is the self-resonant frequency of the active inductor. Also observe
that Z has a zero at frequency
ω
z
=
R
s
L
=
G
o1
C
1
. (2.18)
The Bod´e plots of Z are sketched in Fig.2.6. It is evident that the gyrator-C
network is resistive when ω < ω
z
, inductive when ω
z
< ω < ω
o
, and capacitive
when ω > ω
o
. The frequency range in which the gyrator-C network is inductive
is lower-bounded by ω
z
and upper-bounded by ω
o
. Also observed is that R
p
has no effect on the frequency range of the active inductor. R
s
, however, affects
the lower bound of the frequency range over which the gyrator-C network is
inductive. The upper bound of the frequency range is set by the self resonant
frequency of the active inductor, which is set by the cut-off frequency of the
transconductors constituting the active inductor. For a given inductance L, to
maximize the frequency range, both R
s
and C
p
should be minimized.
2.2.2 Inductance Tunability
Many applications, such as filters, voltage or current controlled oscillators,
and phase-locked loops, require the inductance of active inductors be tunable
with a large inductance tuning range. It is seen from(2.9) that the inductance of
gyrator-C active inductors can be tuned by either changing the load capacitance
or varying the transconductances of the transconductors constituting the active
inductors. Capacitance tuning in standard CMOS technologies is usually done
by using varactors. Two types of varactors exists, namely pn-junction varactors
Characterization of Active Inductors 31
90
-90
0
|Z(j )| w
Z(j ) w
w
w
45 deg./dec.
-90 deg./dec.
Inductive Capacitive Resistive
20 dB/dec.
-20 dB/dec.
w w
z o
w w
z o
(dB)
(Degree)
R R
R +R
s p
p s
Figure 2.6. Bod´ e plots of the impedance of lossy gyrator-C active inductors.
and MOS varactors. The sideviews of pn-junction varactors are shown in
Fig.2.7 for p+/n-well junctions and Fig.2.8 for n+/p-well junctions. Because
p-substrate is connected to the ground, n+/p-well varactors are single-ended.
p+/n-well varactors, on the other hand, are floating varactors. The swing of the
voltages at the nodes of the varactors must ensure that the n+/p-well and p+/n-
well junctions be revise biased all the time such that a junction capacitance
exists. The junction capacitance of an abrupt pn-junction is given by
C
J
=
C
Jo
_
1 +
v
R
φ
o
, (2.19)
where C
Jo
is the junction capacitance at zero-biasing voltage, v
R
is the reverse
biasing voltage of the junction and φ
o
is built-in potential of the junction. It
is seen that C
J
varies with v
R
in a nonlinear fashion. The performance of
junction varactors is affected by the following factors :
Large parasitic series resistance - p+/n-well varactors suffer from a large
series resistance - the resistance of the n-well. As a result, the quality factor
of the varactor quantified by
Q =
1
ωR
n−well
C
, (2.20)
32 CMOS Active Inductors
where R
n−well
is the parasitic series resistance, is small. To minimize this
unwanted resistance, the spacing between p+ and n+ diffusions should be
minimized.
Large parasitic capacitance between n-well and p-substrate - The larger
the capacitance of the varactors, the larger the n-well and subsequently the
larger the n-well/p-substrate junction capacitance.
Small capacitance tuning range - The nonlinear characteristics of C
J
result
in a small capacitance tuning range with a low capacitance tuning ratio.
Stringent voltage swing requirement - As pointed out earlier that the p+/n-
well and n+/p-well junctions must remain in a reverse biasing condition all
the time to ensure the existence of a junction capacitance. This imposes a
stringent constraint on the swing of the voltage across the terminals of the
varactors
p-substrate
n-well
p+ n+
R
n-well
pn-junction
1 2
1
2
C
J
C
J
R
n-well
D
Figure 2.7. Sideview of p+/n-well varactors
The sideview of accumulation-mode MOS varactors is shown in Fig.2.9 for
V
G
< V
S
and Fig.2.10 for V
G
> V
S
. If V
G
< V
S
, the electrons in the n-well
region underneath the gate will be repelled and a depletion region is created.
When V
G
> V
S
, the electrons fromthe n+ diffusion regions will be pulled to the
region underneath the gate, creating an accumulation layer and C
GS
arises to
the gate-oxide capacitance. As pointed out in [68] that
C
max
C
min
can be made from
2.5 to 3 when −1V ≤v
GS
≤1V . A key advantage of accumulation-mode MOS
varactors is the large voltage swing across the terminals of the varactors. They
are the most widely used varactors in voltage/current-controlled oscillators.
Characterization of Active Inductors 33
p-substrate
p-well
n+
p+
R
p-well
1
2
1
2
C
J
C
J
Figure 2.8. Sideview of n+/p-well varactors.
A common drawback of junction varactors and MOS varactors is their small
capacitance tuning range.
n-well
n+
V
G S
V
p-substrate
n+
Depletion region
Figure 2.9. Sideview of accumulation-mode MOS varactors when V
G
< V
S
.
Conductance tuning can be done by varying the dc operating point of the
transconductors. This approach offers a large conductance tuning range, subse-
quently a large inductance tuning range. The conductance tuning range is set by
the constraint that the transconducting transistors of the transconductors must
remain in the saturation. Conductance tuning can be used for the coarse tuning
of the inductance while capacitance tuning can be used for the fine tuning of the
inductance, as shown in Fig.2.12. The conductance of either the transconductor
34 CMOS Active Inductors
n-well
n+
V
G S
V
p-substrate
n+
Gate voltage
induced electrons
Figure 2.10. Sideview of accumulation-mode MOS varactors when V
G
> V
S
.
v
GS
C
min
C
max
C
GS
V
GS
Figure 2.11. Capacitance of accumulation-mode MOS varactors.
with a positive transconductance or that with a negative transconductance can
be tuned. The conductance tuning range is set by the pinch-off condition while
the capacitance tuning range is set by the range of the control voltage of the
varactors.
It is seen from (2.9) that a change in the transconductances of the transcon-
ductors of an active inductor will not affect R
p
and C
p
of gyrator-C active
inductors. It will, however, alter the parasitic series resistance R
s
of the active
inductor. This is echoed with a change in the quality factor of the active induc-
tors. The variation of the quality factor due to the tuning of Lmust therefore be
compensated for such that L and Q are tuned in a truly independent fashion. It
should be noted that the fine tuning of the inductance of active inductors from
the capacitance tuning does not affect R
s
.
Characterization of Active Inductors 35
G V
m1
2
-G V
m2
1
C
2
1
C
x
V
in
Figure 2.12. Inductance tuning of gyrator-C active inductors. Conductance tuning can be
carried out by varying either G
m1
or G
m2
while capacitance tuning is done by varying the
varactor C
x
.
2.2.3 Quality Factor
The quality factor Q of an inductor quantifies the ratio of the net magnetic
energy stored in the inductor to its ohmic loss in one oscillation cycle. For
spiral inductors, the quality factor of these inductors is independent of the
voltage / current of the inductors. This property, however, does not hold
for active inductors as the inductance of these inductors depends upon the
transconductances of the transconductors constituting the active inductors and
the load capacitance. When active inductors are used in applications such
as LC oscillators, the inductance of the active inductors is a strong function
of the swing of the voltage and current of the oscillators. To quantify the
ratio of the net magnetic energy stored in the inductor to its ohmic loss in one
oscillation cycle and relate it to the performance of LCoscillators, in particular,
the phase noise of the oscillators, an alternative definition of the quality factor
that accounts for the swing of the voltage / current of the active inductors is
needed.
Instantaneous Quality Factor
The quality factor Q of an inductor quantifies the ratio of the net magnetic
energy stored in the inductor to its ohmic loss in one oscillation cycle [33, 29]
Q = 2π×
Net magnetic energy stored
Energy dissipated in one oscillation cycle
. (2.21)
For a linear inductor, the complex power of the active inductor is obtained
from
P(jω) = I(jω)V

(jω) = e[Z]|I(jω)|
2
+jm[Z]|I(jω)|
2
, (2.22)
36 CMOS Active Inductors
where e[Z] and m[Z] are the resistance and inductive reactance of the
inductor, respectively, V (jω) and I(jω) are the voltage across and the current
through the inductor, respectively, the superscript ∗ is the complex conjugation
operator, and |.| is the absolute value operator. The first termin (2.22) quantifies
the net energy loss arising fromthe parasitic resistances of the inductor, whereas
the second term measures the magnetic energy stored in the inductor. Eq.(2.21)
in this case becomes
Q =
m[Z]
e[Z]
. (2.23)
Eq.(2.23) provides a convenient way to quantify Qof linear inductors including
active inductors.
Active inductors are linear when the swing of the voltages / currents of
the inductors are small and all transistors of the active inductors are properly
biased. The quality factor of a lossy gyrator-C active inductor can be derived
directly from (2.15) and (2.23)
Q =
_
ωL
R
s
_
R
p
R
p
+R
s
_
1 +
_
ωL
R
s
_
2
_
_
1 −
R
2
s
C
p
L
−ω
2
LC
p
_
. (2.24)
Fig.2.13 shows the frequency dependence of the quality factor of the active
inductor with R
s
= 4Ω, R
p
= 1kΩ, C
p
= 140 fF, and L = 1.6 nH [33]. It is
seen that the first term in (2.24), denoted by
Q
1
=
ωL
R
s
, (2.25)
quantifies the quality factor of the active inductor at low frequencies. The
second term, denoted by
Q
2
=
R
p
R
p
+R
s
_
1 +
_
ωL
R
s
_
2
_
, (2.26)
accounts for the effect of the finite output impedance of deep sub-micron
MOSFETs, whereas the third term, denoted by
Q
3
= 1 −
R
2
s
C
p
L
−ω
2
LC
p
, (2.27)
Characterization of Active Inductors 37
10
8
10
9
10
10
10
11
0
1
2
3
4
5
6
7
8
9
10
Ferquency [Hz]
Q
u
a
l
i
t
y

f
a
c
t
o
r
Q
1
(ω)
Q
2
(ω)
Q
3
(ω)
Q(ω)
Figure 2.13. Frequency dependence of the quality factor of active inductors.
shows that the quality factor vanishes when frequency approaches the cut-off
frequency of the transconductors of the active inductor. Q
2
and Q
3
manifest
themselves at high frequencies only.
The sensitivity of the quality factor of the active inductor with respect to
R
s
and R
p
is investigated in Figs.2.14 and 2.15, respectively. It is seen that
Q
1
dominates the quality of the active inductor and is therefore widely used to
quantify the quality factor of active inductors.
To boost the quality factor of active inductors, R
s
must be minimized. Four
approaches can be used to reduce R
s
:
Approach 1 - Because R
s
=
G
o1
G
m1
G
m2
, R
s
can be lowered by reducing
G
o1
directly. Since G
o1
is typically the output impedance of the transcon-
ductor with a positive transconductance, the use of transconductors with a
large output impedance is critical. As an example, consider the transcon-
ductors shown in Fig.2.16. The transconductance of the transconductor in
Fig.2.16(a) is positive. This is because an increase in v
in
will lead to an
increase in v
GS
, which in turn increases i
D
. Because i
o
= i
D
− J, i
o
will
increase as well. The transconductance of the transconductor in Fig.2.16(b)
is also positive. This is because an increase in v
in
will decrease i
D1
. As
a result, i
D2
= J
1
− i
D1
will increase. Since i
o
= i
D2
− J
2
, i
o
will
increase as well. Although both transconductors have a positive transcon-
38 CMOS Active Inductors
10
8
10
9
10
10
10
11
0
5
10
15
Ferquency [Hz]
Q
u
a
l
i
t
y

f
a
c
t
o
r
R
s
=1Ω
R
s
=10Ω
Figure 2.14. The effect of R
s
on the quality factor of active inductors. R
s
is varied from 1Ω
to 10Ω with step 1Ω.
ductance and both have an infinite input impedance, the output impedance
of the transconductor in Fig.2.16(a) is given by
1
g
m
approximately whereas
that of the transconductor in Fig.2.16(b) is given by r
o2
. Active inductors
constructed using the transconductor in Fig.2.16(b) will have a smaller R
s
,
subsequently a higher Q.
Approach 2 - R
s
can be lowered by increasing G
m1
and G
m2
directly. Since
the transconductances of the transconductors are directly proportional to the
dc biasing currents and the width of the transistors of the transconductors,
R
s
can be lowered by either increasing the dc biasing currents or increasing
the transistor width. The former, however, increases the static power con-
sumption of the active inductors whereas the latter lowers the self-resonant
frequency of the active inductors. Another downside of this approach is
that the inductance of the inductors L will also be affected.
Approach 3 - Reduce G
o1
using advanced circuit techniques, such as cas-
codes. Cascodes are effective in lowering the output conductance and can
be used here to reduce G
o1
, as shown in Fig.2.17. Table 2.1 compares
the minimum supply voltage and output conductance of basic, cascode,
regulated cascode, and multi-regulated cascode transconductors.
Characterization of Active Inductors 39
10
8
10
9
10
10
10
11
0
1
2
3
4
5
6
7
8
9
10
Ferquency [Hz]
Q
u
a
l
i
t
y

f
a
c
t
o
r
R
p
=1 kΩ
R
p
=10 kΩ
Figure 2.15. The effect of R
p
on the quality factor of active inductors. R
p
is varied from 1kΩ
to 10kΩ with step 1kΩ.
M1 M2
V
b
v
in
v
in
i
o
i
o
(a) (b)
J
J
J
1
2
M1 M2
V
b
v
in
i
o
(c)
J
J
1
2
Figure 2.16. Simplified schematics of transconductors with a positive transconductor. (a)
Common-drain transconductor; (b,c) Differential-pair transconductors.
Approach 4 - Use a shunt negative resistor at the output of the positive
transconductor to cancel out the parasitic resistances, both series and par-
allel, of active inductors. It is well known that the series RL network of
the RLC network shown in Fig.2.18 can be replaced with the parallel RL
network shown in the figure with [68]
40 CMOS Active Inductors
M1
v
M1
M2
M1
M2
in
V
b
(a) (b) (c)
M2
V
b
M1
v
in
(d)
i
o
i
o
i
o
i
o
v
in
v
in
v
in
v
o
-A
v
in
v
o
-A
M3
M3
M4
Figure 2.17. Simplified schematics of cascode transconductors. (a) Basic transconductor; (b)
Cascode transconductor; (c) Regulated cascode transconductor; (d) Multi-regulated cascode
transconductor.
R
p
= R
s
(1 +Q
2
)
L
p
= L
s
_
1 +
1
Q
2
_
,
(2.28)
where Q =
ωL
s
R
s
. This is because in order to have the two network to
be equivalent, in other words, to exhibit the same terminal impedance,
Z
s
(jω) = Z
p
(jω) is required.
R
s
+jωL
s
=
jωR
p
L
p
R
p
+jωL
p
. (2.29)
Matching the real and imaginary parts yield
Characterization of Active Inductors 41
Table 2.1. The minimumsupply voltage and output conductance of basic and cascode transcon-
ductors.
Transconductor Min. V
DD
Output impedance
Basic 2V
sat
G
o
=
1
r
o
Cascode 3V
sat
G
o
=
1
r
o1
(r
o2
g
m2
)
Regulated cascode 2V
T
+V
sat
G
o
=
1
r
o1
(r
o2
g
m2
)(r
o3
g
m3
)
Multi-regulated cascode 2V
T
+V
sat
G
o
=
1
r
o1
(r
o2
g
m2
)(r
o3
g
m3
)(r
o4
g
m4
)
.
R
s
R
p
−ω
2
L
s
L
p
= 0,
R
s
L
p
+R
p
L
s
= R
p
L
p
.
(2.30)
Eq.(2.28) follows fromEq.(2.30). Note that (2.28) is valid at all frequencies.
R
L
s
Z
R
L
p
p
s
Z
s
p
C C
Figure 2.18. Transformation of a RL parallel branch to a RL series branch.
Now consider the RLC network of an active inductor shown in Fig.2.19.
The L ∼ R
s
branch of the RLC circuit of the active inductor is replaced
with the parallel
ˆ
L ∼
ˆ
R
p
network with
ˆ
L = L
_
1 +
1
Q
2
_
,
ˆ
R
p
= R
s
(1 +Q
2
).
(2.31)
42 CMOS Active Inductors
Consider two cases :
– Case 1 - If R
s
is negligible, the quality factor of the active inductor is
mainly determined by R
p
. From (2.24) with R
s
= 0, we arrive at
Q =
R
p
ωL
_
1 −
_
ω
ω
o
_
2
_
. (2.32)
At frequencies below ω
o
,
_
ω
ω
o
_
2
≈0 and
Q≈
R
p
ωL
s
(2.33)
follows.
– Case 2 - If R
p
is large, the quality factor is mainly determined by R
s
Q =
ωL
R
s
_
1 −
_
ω
z
ω
o
_
2

_
ω
ω
o
_
2
_
. (2.34)
At frequencies above ω
z
and below ω
o
,
_
ω
z
ω
o
_
2
≈0 and
_
ω
ω
o
_
2
≈0.
Eq.(2.34) is simplified to
Q≈
ωL
R
s
. (2.35)
As shown in Fig.2.19, the total parasitic parallel resistance of the active
inductor becomes
R
total
= R
p
//
ˆ
R
p
. (2.36)
In this case, a negative resistor of resistance R
comp
= −R
total
can be con-
nected in parallel with C
p
to eliminate the effect of both R
p
and R
s
of the
active inductor simultaneously. Note that the resistance of the negative re-
sistor should be made tunable such that a total cancellation can be achieved.
The quality factor of the compensated active inductor at ω
o
is given by
Q(ω
o
) = (R
p
||
ˆ
R
p
||R
comp
)
¸
C
p
+C
comp
ˆ
L
, (2.37)
Characterization of Active Inductors 43
where C
comp
is the input capacitance of the compensating negative resistor.
It should be noted that because R
s
and R
p
are frequency-dependent, R
comp
should be designed in such a way that a total resistance cancellation is
achieved across the frequency range of the active inductor. It should also be
noted that although the negative resistor compensation technique is widely
used to improve the quality factor of spiral inductors, a total compensation in
this case is difficult to achieve. This is because an active negative resistor is
used to cancel out the largely skin-effect induced parasitic series resistance
of spirals.
C R
R
L
p p
s
C R
R
L
p p
p
R
comp
C
comp
Figure 2.19. Q enhancement using a shunt negative resistor.
Average Quality Factors
Active inductors are RLC tanks when R
s
, R
p
, and C
p
are accounted for.
The quality factor of LC tanks is obtained from [69]
Q(ω) =
ω
o
2
∂φ(ω)
∂ω
, (2.38)
where φ(ω) is the phase of the tank impedance. The quality factor of a passive
LC tank at a given frequency is independent of the current of the tank.
Unlike passive LC tanks, the inductance of the active inductors varies with
the current / voltage of the inductors. The effective quality factor defined as
Q(ω) =
ω
2(I
max
−I
min
)
_
I
max
I
min
Q(ω, i)di, (2.39)
where I
min
and I
max
are the minimum and maximum currents of the transcon-
ductors of active inductors, and Q(ω
o
, i) is the instantaneous quality factor
at frequency ω and channel current i provides an effective mean to quantify
the quality factor of active inductors, especially when active inductors are em-
ployed in circuits that are operated in a large-signal mode, such as LC tank
oscillators.
44 CMOS Active Inductors
2.2.4 Noise
Active inductors exhibit a high level of noise as compared with their spiral
counterparts. To analyze the noise of a gyrator-C active inductor, the power
of the input-referred noise-voltage and that of the noise-current generators
of the transconductors constituting the active inductor must be derived first.
Fig.2.20 shows the partial schematics of basic transconductors widely used in
the construction of gyrator-C active inductors. The power of the input-referred
noise-voltage generator, denoted by v
2
n
, and that of the input-referred noise-
current generator, denoted by i
2
n
, of these transconductors can be derived using
conventional noise analysis approaches for 2-port networks [70], and the results
are given in Table 2.2 where
i
2
nD
= 4kT(γ +R
g
g
m
)g
m
Δf (2.40)
represents the sum of the power of the thermal noise generated in the channel
of MOSFETs and the thermal noise of the gate series resistance of MOSFETs,
R
g
is the gate series resistance, γ = 2.5 for deep sub-micron devices, T is the
temperature in degrees Kelvin, and k is Boltzmann constant. The effect of the
flicker noise of MOSFETs, which has a typical corner frequency of a few MHz
[71], is neglected. The thermal noise of other parasitics of MOSFETs, such as
the thermal noise of the bulk resistance of the source and drain diffusions, is
also neglected.
To illustrate how the results of Table 2.2 are derived, consider the common-
gate transconductor. To derive the input-referred noise-voltage generator v
2
n
of the transconductor, we first short-circuit the input of the transconductor, as
shown in Fig.2.21.
The output noise power of the transconductor due to i
nD
is calculated
v
2
no
= r
2
o
i
2
nD
, (2.41)
where r
o
is the output resistance of the transistor. We then remove i
2
nD
and
apply v
n
at the input of the transconductor, as shown in Fig.2.22. The output
noise power of the transconductor is obtained
v
2
no
= (1 +g
m
r
o
)
2
v
2
n
. (2.42)
Equating (2.41) and (2.42) yields
v
2
n
=
r
2
o
(1 +g
m
r
o
)
2
i
2
nD

1
g
2
m
i
2
nD
. (2.43)
Characterization of Active Inductors 45
Common
source
Cascode
Common
gate
Source
follower
Transconductors with noise sources
Transconductors with noise generators
Figure 2.20. Input-referred noise-voltage and noise-current generators of transconductors at
low frequencies.
46 CMOS Active Inductors
Table 2.2. Power of input-referrednoise-voltage andnoise-current generators of basic transcon-
ductors at low frequencies.
Transconductor v
2
n
i
2
n
Common source v
2
n
=
i
2
nD
g
2
m
i
2
n
= 0
Cascode v
2
n
=
i
2
nD1
g
2
m1
+
i
2
nD2
(g
m1
r
o1
g
m2
)
2
i
2
n
= 0
Common gate v
2
n
=
i
2
nD
g
2
m
i
2
n
= 0
Source follower v
2
n
=
i
2
nD
g
2
m
i
2
n
= 0
g v
m
gs
r
o
v
in
i
nD
v
gs
V
b
i
nD
2
v
no
Figure 2.21. Derivation of input-referred noise-voltage generator of a common-gate transcon-
ductor at low frequencies.
Note that we have utilized r
o
g
m
1 in (2.43) to simplify the results.
To derive the noise-current generator of the common-gate transconductor,
consider Fig.2.23 where the input port of the transconductor is open-circuited
and the output noise power of the circuit is calculated. To avoid the difficulty
caused by the floating node 1, we assume that there exists a resistor of resistance
Characterization of Active Inductors 47
g v
m
gs
r
o
v
n
v
gs
V
b
v
n
v
no
Figure 2.22. Derivation of the input-referred noise-voltage generator of a common-gate
transconductor at low frequencies.
g v
m
gs
r
o
v
in
i
nD
v
gs
V
b
i
nD
2
v
no
R
x
1
2
Figure 2.23. Derivation of the input-referred noise-current generator of a common-gate
transconductor at low frequencies.
R
x
between node 1 and the ground. Note that this approach is used in most IC
CAD systems to avoid floating nodes. Writing KCL at nodes 1 and 2 yields
(g
x
+g
m
+g
o
)v
1
−g
o
v
2
= i
nD
(node 1),
−(g
o
+g
m
)v
1
+g
o
v
2
+i
nD
= 0 (node 2).
(2.44)
Solving (2.44) yields
v
2
2
=
1
g
2
o
i
2
nD
. (2.45)
48 CMOS Active Inductors
We then apply the noise-current generator i
n
at the input of the circuit,
remove all the noise sources of the circuit, as shown in Fig.2.24, and compute
the output noise power. Writing KCL at nodes 1 and 2 yields
(g
x
+g
m
+g
o
)v
1
−g
o
v
2
+i
n
= 0 (node 1),
−(g
o
+g
m
)v
1
+g
o
v
2
= 0 (node 2).
(2.46)
Solving (2.46) yields
v
2
2
=
_
1 +
g
m
g
o
_
2
1
g
2
x
i
2
n
. (2.47)
Equating (2.45) and (2.47) yields
i
2
n
=
_
g
x
g
o
_
2
i
2
nD
_
1 +
g
m
g
o
_
2
. (2.48)
Taking the limit R
x
→∞or equivalently g
x
→0 in (2.48), we arrive at
i
2
n
= 0. (2.49)
g v
m
gs
r
o
i
n
v
gs
V
b
i
n
v
no
R
x
1
2
Figure 2.24. Derivation of the input-referred noise-current generator of a common-gate
transconductor at low frequencies.
Once v
2
n
and i
2
n
of the transconductors are available, the power of the input-
referred noise-voltage and noise-current generators of active inductors can be
derived.
Characterization of Active Inductors 49
Consider the active inductor of Fig.2.25(a) where v
2
n1
and v
2
n2
denote the
power of the noise-voltage generators of the transconductors 1 and 2, respec-
tively, and Y
1
and Y
2
are the admittance at ports 1 and 2, respectively. For the
network of Fig.2.25(a), it is trivial to show that
V
2
1
=
_
V
n1
+Y
1
G
m2
V
n2
+Y
2
V
n1
Y
1
Y
2
+G
m1
G
m2
_
2
. (2.50)
For the network of Fig.2.25(b), we have
V
2
1
=
_
V
n
+
Y
1
Y
1
Y
2
+G
m1
G
m2
I
n
_
2
. (2.51)
To ensure that Fig.2.25(a) and Fig.2.25(b) are equivalent, the right hand-side
of (2.50) and that of (2.51) must be the same. To achieve this, we impose
V
n
= V
n1
,
I
n
= Y
2
V
n1
+G
m2
V
n2
.
(2.52)
Because
Z
in
(s) =
Y
1
Y
1
Y
2
+G
m1
G
m2
, (2.53)
we arrive at
V
n
= V
n1
+Z
in
I
n
. (2.54)
For lossy gyrator-C active inductors, we have
Y
1
= G
o1
+sC
1
,
Y
2
= G
o2
+sC
2
,
(2.55)
Eq.(2.52) becomes
V
n
= V
n1
,
I
n
= (G
o2
+sC
2
)V
n1
+G
m2
V
n2
.
(2.56)
50 CMOS Active Inductors
By assuming V
n1
and V
n2
are uncorrelated, we arrive at
V
2
n
= V
2
n1
,
I
2
n
=
¸
¸
¸G
o2
+jωC
2
¸
¸
¸
2
V
2
n1
+G
2
m2
V
2
n2
.
(2.57)
G V
m1
+
G V
m2
-
G V
m1 +
G V
m2 -
v
n1
v
n2
v
n
i
n
2
1
2
1
Y Y
2 1
Y
1
Y
2
Figure 2.25. Noise of single-ended gyrator-C active inductors.
2.2.5 Linearity
The preceding development of gyrator-C active inductors assumes that the
transconductors of the active inductors are linear. This assumption is only
valid if the swing of the input voltage of the transconductors is small. When
the voltage swing is large, the transconductors will exhibit a nonlinear charac-
teristic and the synthesized active inductors are no longer linear. The linearity
constraint of active inductors sets the maximum swing of the voltage of the
active inductors. If we assume that the transconductances of the transistors
of gyrator-C active inductors are constant when the transistors are biased in
the saturation, then the maximum swing of the voltage of the active inductors
can be estimated from the pinch-off condition of the transistors. When the
transistors of active inductors enter the triode region, the transconductances
of the transistors decrease from g
m
(saturation) to g
ds
(triode) in a nonlinear
fashion, as illustrated graphically in Fig.2.26. It should be emphasized that
although the transconductances of the transconductors of gyrator-C networks
drop when the operating point of the transistors of the transconductors moves
from the saturation region to the triode region, the inductive characteristics at
port 2 of the gyrator-C network remain. The inductance of the gyrator-C active
inductors, however, increases from L =
C
G
m1
G
m2
to L =
C
G
ds1
G
ds2
, where
G
m1,m2
are the transconductances of transconductors 1 and 2 respectively when
(b) (a)
Characterization of Active Inductors 51
in the saturation and G
ds1,ds2
are the transconductances of transconductors 1
and 2 respectively when in the triode. Note that we have assumed the load
capacitance C remains unchanged. The inductance thus varies with the swing
of the voltage of the active inductors in a nonlinear fashion. It should also be
noted that the parasitic resistances of active inductors also vary with the voltage
swing of the active inductors.
V
GS1
V
GS2
i
DS
v
DS
Di
DS1
Di
DS2
Di
DS1
DV
GS
g =
m
Di
DS2
DV
GS
g =
ds
Pinch-off
V
DS1
V
DS2
Figure 2.26. Transconductance of MOSFETs in the saturation and triode regions. Because
Δi
DS1
> Δi
DS2
, g
m
> g
ds
follows.
2.2.6 Stability
Gyrator-C active inductors are negative feedback systems. The stability
of active inductors is critical to the overall stability of systems employing
active inductors. In this section, we investigate the stability of gyrator-C active
inductors.
The impedance looking into port 2 of the gyrator-C active inductor shown
in Fig.2.4 is given by
Z =
sC
1
+G
o1
s
2
C
1
C
2
+s(C
1
+C
2
) +G
m1
G
m2
, (2.58)
where we have utilized G
m
G
o
to simply the results. The poles of the system
are given by
p
1,2
=
C
1
+C
2
2C
1
C
2
_
−1±
¸
1 −
4C
1
C
2
G
m1
G
m2
(C
1
+C
2
)
2
_
. (2.59)
52 CMOS Active Inductors
The poles of the gyrator-C active inductor are located in the left half of the
s-plane and the gyrator-C active inductor is a stable system.
The degree of stability can be assessed by evaluating its damping factor,
which is obtained by comparing the denominator of (2.58) with the standard
form of the characteristic equation of second-order systems
s
2
+ 2ω
o
ξs +ω
2
o
= 0, (2.60)
where ξ denotes the damping factor and ω
o
is the pole resonant frequency. The
result is given by
ξ =
1
2

G
m1
G
m2
_
¸
C
2
C
1
+
¸
C
1
C
2
_
. (2.61)
Eq.(2.61) reveals that an increase in G
m1
and G
m2
will lead to a decrease in ξ.
This is echoed with an increase in the level of oscillation in the response of the
active inductor. Also observed from (2.61) is that the ratios
C
1
C
2
and
C
2
C
1
have
a marginal impact on the damping factor simply because these two quantities
vary in the opposite directions when C
1
and C
2
change, and the values of C
1
and C
2
are often close.
If C
1
= C
2
= C and G
m1
= G
m2
= G
m
, we have
p
1,2
=
1
C
_
−1±
_
1 −G
2
m
_
,
ξ =
1
G
m
.
(2.62)
An increase of G
m
will lead to a decrease of ξ. This is echoed with a reduced
level of damping. Because e[p
1,2
] = −
1
C
, the absolute stability margin is
set by the capacitance C and is independent of G
m
. It should be noted that
the preceding analysis is based on the assumption that active inductors are
2nd-order systems. When the parasitics of MOSFETs are accounted for, active
inductors are no longer 2nd-order systems and their stability will deteriorate.
2.2.7 Supply Voltage Sensitivity
The supply voltage sensitivity of the inductance of active inductors is a
figure-of-merit quantifying the effect of the variation of the supply voltage on
the inductance of the active inductors. The fluctuation of the supply voltage of a
mixed analog-digital system is mainly due to the switching noise of the system
Characterization of Active Inductors 53
[14]. Assume that the supply voltage of a mixed-mode system containing an
active inductor varies fromV
DD
and V
DD
+ΔV
DD
, where ΔV
DD
is a random
variable with E[ΔV
DD
] = 0, where E[.] denotes the mathematical mean
operator. For a well designed mixed-mode system, ΔV
DD
V
DD
holds. The
small-signal analysis approach can therefore be employed to analyze the effect
of ΔV
DD
on the inductance of the active inductor. Following the definition of
normalized sensitivity given in [72], the normalized sensitivity of the inductance
of an active inductor to the supply voltage is defined as
S
L
V
DD
=
V
DD
L
∂L
∂V
DD
. (2.63)
The fluctuation of the supply voltage V
DD
affects the inductance of the
active inductor mainly by altering the dc operating point, subsequently the
transconductances of the transconductors constituting the active inductor. By
assuming that the load capacitance C of the gyrator-C active inductor does not
vary with V
DD
and because L =
C
G
m1
G
m2
, we arrive at
∂L
∂V
DD
= −L
_
1
G
m1
∂L
G
m1
+
1
G
m2
∂L
G
m2
_
. (2.64)
The normalized supply voltage sensitivity of the active inductor is given by
S
L
V
DD
= −
_
S
G
m1
V
DD
+S
G
m2
V
DD
_
, (2.65)
where S
G
m1
V
DD
and S
G
m2
V
DD
are the normalized supply voltage sensitivity of G
m1
and G
m2
, respectively. Eq.(2.65) reveals that both S
G
m1
V
DD
and S
G
m2
V
DD
contribute
equally to S
L
V
DD
. To minimize the supply voltage sensitivity of active inductors,
transconductors with a constant G
m
should be used.
2.2.8 Parameter Sensitivity
The minimum feature size of MOS devices in modern CMOS technologies
has been scaled down more aggressively as compared with the improvement in
process tolerance such that the effect of process variation on the characteristics
of circuits becomes increasingly critical. For example, the resistance of poly re-
sistors in a typical 0.18μm CMOS process has an error of ±20%approximately
and that of n-well resistors has an error of ±30% approximately. Analysis of
the effect of parameter spread is vital to ensure that the performance of circuits
meets design specifications once the circuits are fabricated. Active inductors
consist of a number of active devices and their performance is greatly affected
54 CMOS Active Inductors
by the parameter spread of these components. The normalized sensitivity of
the inductance of an active inductor to a parameter x
j
of the inductor defined
as
S
L
x
j
=
x
j
L
∂L
∂x
j
(2.66)
quantifies the effect of the variation of the parameter x
j
on the inductance of
the active inductor. By assuming that the parameters of the active inductor are
Gaussian distributed and uncorrelated, the overall effect of the variation of the
parameters of the active inductor on the inductance of the inductor is obtained
from
σ
2
L
=
N

j=1
_
∂L
∂x
j
_
2
σ
2
x
j
, (2.67)
where σ
L
and σ
x
j
denote the standard deviations of L and x
j
, respectively, and
N is the number of the parameters of the active inductor. For a gyrator-C active
inductor, because
∂L
C
=
1
G
m1
G
m2
,
∂L
G
m1
= −
C
G
2
m1
G
m2
,
∂L
G
m2
= −
C
G
m1
G
2
m2
,
(2.68)
we obtain the normalized spread of the inductance of the active inductor
σ
2
L
L
2
=
σ
2
C
C
2
+
σ
2
G
m1
G
2
m1
+
σ
2
G
m2
G
2
m2
. (2.69)
There are two ways in which circuit designers can analyze the effect of
parameter spread on the inductance of active inductors, namely worst-case
analysis, also known as corner analysis, and Monte Carlo analysis. The former
determines the inductance of active inductors at process corners while the latter
quantifies the degree of the spread of the inductance of active inductors around
the nominal inductance of the inductors. The accuracy of Monte Carlo analysis
increases with an increase in the number of simulation runs and is therefore
Implementation of Single-Ended Active Inductors 55
extremely time consuming. Corner analysis, on the other hand, is time-efficient
but the results obtained from corner analysis are typically over conservative.
Despite of this, corner analysis is the most widely used method to quantify the
effect of process spread.
2.2.9 Signal Sensitivity
Unlike spiral inductors whose inductance is independent of the voltage and
current of the inductors, the inductance of gyrator-Cactive inductors varies with
the voltage and current of the transconductors constituting the active inductors.
This is because the transconductances G
m1
and G
m2
of the transconductors are
signal dependent when signal swing is large. When an active inductor is used in
applications where the voltage of the active inductor experiences a large degree
of variation, such as active inductor LCoscillators, the transconductances of the
transconductors of the active inductor vary with the signal swing. As a result,
the inductance, parasitic resistances, and quality factor of the active inductor
all vary with the signal swing.
2.2.10 Power Consumption
Spiral inductors do not consume static power. Gyrator-C active inductors,
however, consume dc power, mainly due to the dc biasing currents of their
transconductors. The power consumption of gyrator-C active inductors them-
selves is usually not of a critical concern because the inductance of these
inductors is inversely proportional to the transconductances of the transcon-
ductors constituting the inductors. To have a large inductance, G
m1
and G
m2
are made small. This is typically achieved by lowering the dc biasing currents
of the transconductors. When replica-biasing is used to minimize the effect of
supply voltage fluctuation on the inductance of active inductors, as to be seen
shortly, the power consumed by the replica-biasing network must be accounted
for. Also, when negative resistors are employed for boosting the quality factor
of active inductors, their power consumption must also be included. Often the
power consumption of an active inductor is set by that of its replica-biasing and
negative resistor networks.
2.3 Implementation of Single-Ended Active Inductors
The need for a high self-resonant frequency of active inductors requires
that the transconductors of these active inductors be configured as simple as
possible. This also lowers their level of power consumption and reduces the
silicon area required to fabricate the inductors. Most reported gyrator-C active
inductors employ a common-source configuration as negative transconductors,
common-gate, source follower, and differential pair configurations as positive
56 CMOS Active Inductors
transconductors. These basic transconductors have the simplest configurations
subsequently the highest cutoff frequencies and the lowest silicon consumption.
The load capacitor of the transconductors is realized using the intrinsic
capacitance C
gs
of the transistors of the transconductors directly to maximize
the upper bound of the frequency range of the active inductors and to avoid the
use of expensive floating capacitors, which are available only in mixed-mode
CMOS processes. MOS varactors are often added in parallel with C
gs
to tune
the inductance of active inductors.
This section presents a comprehensive treatment of both the circuit imple-
mentation and characteristics of CMOS active inductors. To simplify analysis,
the following assumptions are made in analysis of active inductors and in de-
termination of both their signal swing and the minimum supply voltage : (i)
nMOS and pMOS transistors have the same threshold voltage V
T
. (ii) nMOS
and pMOS transistors have the same pinch-off voltage V
sat
. (iii) Only C
gs
is
considered. C
gd
and parasitic diffusion capacitances are neglected unless oth-
erwise noted explicitly. (iv) The minimum voltage drop across biasing current
sources and current-source loads is V
sat
.
2.3.1 Basic Gyrator-C Active Inductors
Fig.2.27 show the schematic of two basic gyrator-C active inductors. In
Fig.2.27(a), the transconductor with a positive transconductance is common-
gate configured while the transconductor with a negative transconductance is
common-source configured. In Fig.2.27(b), the transconductor with a positive
transconductance is common-drain configured while the transconductor with
a negative transconductance is common-source configured. All transistors
are biased in the saturation. A notable advantage of the active inductor in
Fig.2.27(b) is that all transistors are nMOS, making it attractive for high-
frequency applications.
For Fig.2.27(a), we have C
1
= C
gs2
, G
o1
≈g
o1
, G
m1
= g
m1
, C
2
= C
gs1
,
G
o2
≈g
m1
, and G
m2
= g
m2
, where g
oj
and g
mj
, j = 1, 2, are the output
conductance and transconductance of transistor j, respectively. Using (2.9),
we obtain the parameters of the equivalent RLC network of the active inductor
Implementation of Single-Ended Active Inductors 57
C
p
= C
gs1
,
R
p
=
1
g
m1
,
L =
C
gs2
g
m1
g
m2
,
R
s
=
g
o1
g
m1
g
m2
.
(2.70)
M2
M1
V
M2
M1
(a) (b)
b
J
2
J
1
v
in
v
in
J
1
J
2
1
2
2
1
Figure 2.27. Simplified schematic of basic gyrator-C active inductors.
It is observed from (2.70) that the parasitic parallel resistance R
p
is rather
small, limiting the quality factor of the active inductor. Also, the parasitic
series resistance is large, further lowering the quality factor. In evaluating the
quality factor of this active inductor, the effect of the parasitic series resistance
R
s
is often neglected as R
p
is small. In this case, the quality factor of the active
inductor is obtained from
Q≈
R
p
ωL
=
ω
t2
ω
. (2.71)
The self-resonant frequency of the active inductor is given by
ω
o

1
_
LC
p
=

ω
t1
ω
t2
, (2.72)
where
58 CMOS Active Inductors
ω
tj
=
g
mj
C
gsj
, i = 1, 2 (2.73)
is the cutoff frequency of transconductor j. At the self-resonance frequency of
the active inductor ω
o
=

ω
t1
ω
t2
, the quality factor becomes
Q(ω
o
) =
_
ω
t2
ω
t1
. (2.74)
The frequency of the zero of the active inductor, which is the lower bound of
the frequency range of the active inductor, is given by (2.18)
ω
z
=
g
o1
C
gs2
. (2.75)
Eqs.(2.72) and (2.75) reveal that :
In order to maximize the frequency range of the active inductor, ω
z
should
be minimized. This can be achieved by reducing g
o1
or increasing C
gs2
.
The former is usually preferred as the latter lowers ω
o
.
Because the output impedance of deep sub-micron MOSFETs is small. The
detrimental effect of R
p
=
1
g
m1
on the quality factor of the active inductor
can not be neglected. The effect of R
p
, however, can be eliminated by
connecting a negative resistor of resistance
ˆ
R
p
= −R
p
in parallel with R
p
.
g
m
V
in
I
o
C
in
V
b M2
M1
v
in
J
2
J
1
v
in
(a) (b)
Figure 2.28. Single-ended negative impedance networks. (a) Block diagram. (b) Circuit
implementation.
Implementation of Single-Ended Active Inductors 59
Negative resistors can be realized using transconductors with positive feed-
back, as shown in Fig.2.28 for single-ended negative resistors and Fig.2.29
for differential negative resistors. The positive feedback of the single-ended
negative resistor in Fig.2.28 is depicted as the followings : An increase in
the gate voltage of M
1
will increase the voltage at the source of M
1
. Since
M2 is a common-gate configuration, an increase in the source voltage of
M2 will increase the drain voltage of M2. Apositive feedback is thus estab-
lished. Readers can verify the positive feedback of the differential negative
resistors of Fig.2.29 in a similar manner. For Fig.2.29. It can be shown that
the impedance at low frequencies is given by
Z≈−
_
1
g
m1
+
1
g
m2
_
. (2.76)
To maximize the frequency range over which a constant negative resistance
exists, transconductors synthesizing negative resistors should be configured
as simple as possible.
I
o
+
I
o
-
V
in
g
m
C
in
v
in
M1 M2
J
v
in
(a) (b)
v
in
M1 M2
J
Figure 2.29. Differential negative impedance networks. (a) Block diagram. (b) Circuit imple-
mentation. The tail current source in the differential configuration can be removed, provided
that biasing currents are provided by the circuit connected to the negative resistor. Note that the
removal of the biasing tail current source will also remove the tunability of the resistance of the
negative resistor.
The effect of R
s
can be compensated for in three different ways :
– Use cascodes and regulated cascodes to reduce g
o1
[37, 38, 65, 73]. It
is well known that cascodes and regulated cascodes are effective means
to boost the output impedance of transconductors. The price paid,
however, is the reduced signal swing.
– The reason that the preceding basic active inductors have a low R
p
is
because the input impedance of the positive transistor in Fig.2.27(a)
60 CMOS Active Inductors
is 1/g
m1
and the output impedance of the positive transconductor in
Fig.2.27(b) is 1/g
m2
. The use of transconductors that have both a
large output impedance and a large input impedance will eliminate
this drawback. As an example, the differentially-configured posi-
tive transconductor of Karsilayan-Schaumann active inductors shown
in Fig.2.16(b,c) has an infinite input impedance and a large output
impedance r
o2
[74, 75, 48, 49, 61].
– Because the series RL branch of the RLC equivalent network of the
active inductor can be replaced with a parallel
ˆ
R
ˆ
L branch. The induc-
tance of the parallel
ˆ
R
ˆ
L branch is given by
ˆ
L = L
_
1 +
1
Q
2
_
while the
resistance is given by
ˆ
R
p
= (1 + Q
2
)R
s
. The total parasitic parallel
resistance of the active inductor becomes R
p,total
= R
p
||
ˆ
R
p
. The qual-
ity factor of the active inductor can be made infinite theoretically by
employing a shunt negative resistor whose resistance is −R
p,total
.
Table 2.3 compares the range of the voltage swing and the minimum supply
voltage of the two basic active inductors. It is seen that the active inductor in
Fig.2.27(a) offers a large input voltage swing and requires a lower minimum
supply voltage.
Table 2.3. Comparison of the input voltage swing and the minimum supply voltage of the basic
active inductors in Fig.2.27.
Active inductor Fig.2.27(a) Fig.2.27(b)
Max. input voltage V
DD
−V
T
−V
sat
V
DD
−V
T
−V
sat
Min. input voltage V
sat
V
T
Min. V
DD
V
T
+ 2V
sat
2V
T
+V
sat
2.3.2 Wu Current-Reuse Active Inductors
Fig.2.30 show the schematic of Wu current-reuse active inductors proposed
in [42, 45, 36]. In the nMOS version of the active inductor, the positive
transconductor is common-gate configured while the negative transconductor
is common-source configured. When only C
gs
is considered, C
1
= C
gs2
,
G
o1
≈g
o1
+ g
o2
, G
m1
= g
m1
, C
2
= C
gs1
, G
o2
=
1
g
m1
, and G
m2
= g
m2
. The
parameters of the equivalent RLC network of this active inductor are given by
Implementation of Single-Ended Active Inductors 61
C
p
= C
gs1
,
R
p
=
1
g
m1
,
L =
C
gs2
g
m1
g
m2
,
R
s
=
g
o1
+g
o2
g
m1
g
m2
.
(2.77)
The quality factor of Wu active inductors can be estimated by neglecting the
effect of R
s
and only focusing on R
p
as R
p
is small.
Q≈
R
p
ωL
=
ω
t2
ω
. (2.78)
At the self-resonant frequency ω
o
=

ω
t1
ω
t2
, we have
Q(ω
o
)≈
_
ω
t2
ω
t1
. (2.79)
V
b
V
b
J
1
J
2
J
2
J
1
M1
M2
M1
M2
v
in
v
in
Figure 2.30. Simplified schematic of Wu current reuse active inductors.
It is seen from the preceding analysis that to increase ω
o
, both ω
t1
and ω
t2
need to be increased. Increasing ω
t1
, however, lowers Q(ω
o
). Increasing ω
t1
should therefore be avoided. To boost ω
t2
without increasing ω
t1
, the dc biasing
current of M
1
is kept unchanged while that of M
2
is increased by injecting
an additional current J
2
into M
2
. The additional current source J
2
is used to
boost the transconductance of M
2
such that the upper frequency bound can
be increased without lowering the quality factor. In practical design, J
2
is
62 CMOS Active Inductors
provided by the stage preceding to the inductors and the active inductors are
known as Wu current-reuse active inductors.
2.3.3 Lin-Payne Active Inductors
The Lin-Payne active inductor proposed in [39] and shown in Fig.2.31
requires the minimum supply voltage of only V
T
+ 2V
sat
. Another implemen-
tation of Lin-Payne active inductors is shown in Fig.2.32 [76]. The minimum
supply voltage of the active inductor is also V
T
+2V
sat
. They can be analyzed
in a similar way as Wu current re-use active inductors.
M2
M1
v
in
J
2
J
1
Figure 2.31. Simplified schematic of Lin-
Payne active inductor.
M2
M1
v
in
J
2
J
1
Figure 2.32. Simplified schematic of the
variation of Lin-Payne active inductor.
2.3.4 Ngow-Thanachayanont Active Inductors
Ngow and Thanachayanont proposed the low-voltage active inductor shown
in Fig.2.33[76]. The addition of M
3
branch relaxes the biasing difficulties
encountered in Lin-Payne active inductors. M
1−3
can easily be biased in the
saturation to ensure a stable operation of the active inductors. The analysis of
Ngow-Thanachayanont active inductors is similar to that of Wu current re-use
active inductors and is left as an exercise for readers.
2.3.5 Hara Active Inductors
Although the MESFET implementation of Hara active inductors appeared
two decades ago [77, 78], the CMOS version of Hara active inductors only
emerged a few years ago [41, 40, 79–82]. Hara active inductors shown in
Fig.2.34 employ only a MOSFET and a resistor. They are indeed gyrator-C
active inductors. The feedback operation of nMOS Hara active inductor is as
the followings : An increase of the input current will result in an increase in
the voltage at the input node. Since the gate voltage is kept at V
DD
, v
GS
is
reduced. This in turn lowers the current flowing out of the active inductor.
Implementation of Single-Ended Active Inductors 63
M2
M1
M3
J
2
J
1
J
3
v
in
Figure 2.33. Simplified schematic of low-voltage gyrator-C active inductor.
R
R
V
c
V
c
R
L
s
R
L
s
v
in
v
in
v
in
v
in
Figure 2.34. Hara active inductors. Resistor R can be made variable by implementing it using
MOSFETs biased in the triode.
The input impedance of the nMOS Hara active inductor can be derived from
its small-signal equivalent circuit shown in Fig.2.35
Z≈
_
1
RC
gs
C
gd
_
sRC
gd
+ 1
s
2
+s
g
m
C
gs
+
g
m
RC
gs
C
gd
, (2.80)
where g
m
g
o
and C
gs
C
gd
were utilized to simplify the results.
The self-resonant frequency ω
o
and the frequency of the zero ω
z
of the active
inductor are given by
ω
o
=
¸
g
m
RC
gs
C
gd
=

ω
t
ω
z
,
ω
z
=
1
RC
gd
,
(2.81)
(a) nMOS (b) pMOS
64 CMOS Active Inductors
v
gs
g
o
g v
m
gs
C
gs
C
gd
R
v
in
Figure 2.35. Small-signal equivalent circuit of nMOS Hara active inductors.
where
ω
t
=
g
m
C
gs
. (2.82)
The Bod´e plots of Hara active inductors are the same as those given in Fig.2.6.
The network exhibits an inductive characteristic in the frequency range ω
z
<
ω < ω
o
. When C
gd
, C
sb
, C
sb
, and high-order effects are neglected, the
inductance L and parasitic series resistance R
s
of nMOS Hara active inductors
are given by
R
s
=
g
m

2
C
2
gs
R
g
2
m

2
C
2
gs
, (2.83)
and
L =
C
gs
(g
m
R −1)
g
2
m

2
C
2
gs
. (2.84)
Observe that
g
m
R > 1 (2.85)
is required to ensure L > 0. Under the condition g
m
R1, Eqs.(2.83) and
(2.84) can be written as
R
s
=
1
g
m
+
_
ω
ω
t
_
2
R
1 +
_
ω
ω
t
_
2

1
g
m
, (2.86)
and
Implementation of Single-Ended Active Inductors 65
L =
R
ω
t
_
1 +
_
ω
ω
t
_
2
_

R
ω
t
, (2.87)
where we have utilized
_
ω
ω
t
_
2
≈0. It is evident from (2.87) that the inductance
L is directly proportional to R and can be tuned by varying R. The series
resistance R
s
is largely dominated by g
m
at frequencies below the cut-off
frequency of the transistor.
The dependence of the input impedance of the nMOS Hara active inductor
on the resistance of the resistor R and the width of the transistor is shown in
Fig.2.36 and Fig.2.37, respectively. The active inductor was implemented in
TSMC-0.18μm 1.8V CMOS technology and analyzed using SpectreRF with
BSIM3V3 device models. It is observed that an increase of R will lower both
ω
z
and ω
o
. This agrees with the theoretical results. Increasing the width of the
transistor will lower ω
o
because
ω
t
=
g
m
C
gs

3
2
μ
n
L
2
c
(V
GS
−V
T
), (2.88)
where μ
n
is the surface mobility of free electrons and L
c
is the channel length,
is independent of the width of the transistor. Also, L is nearly independent of
g
m
.
Hara active inductors suffer from the loss of the voltage headroom of at least
V
T
. In [41], a voltage doubler was used to increase the supply voltage for R.
A drawback of this approach is the complexity of the voltage doubler and the
need for a control clock.
2.3.6 Wu Folded Active Inductors
The drawback of Hara active inductors can be eliminated by employing Wu
folded active inductors shown in Fig.2.38[83]. Wu folded active inductors were
initially proposed by Thanachayanont in [84].
To derive the parameters of the RLC equivalent circuit of Wu folded active
inductors, consider the nMOS version of Wu folded active inductors with its
small-signal equivalent circuit shown in Fig.2.39. To simplify analysis, we
neglect C
gd
, g
o
and other parasitic capacitances of the transistor. It can be
shown that the input impedance is given by
Z =
sRC
gs
+ 1
sC
gs
+g
m
. (2.89)
66 CMOS Active Inductors
10
−1
10
0
10
1
10
2
−300
−200
−100
0
100
200
300
400
500
Frequency [GHz]
I
m
p
e
d
a
n
c
e

[
Ω
]
5μm
25μm
5μm
25μm
Im(Z
in
)
Re(Z
in
)
Figure 2.36. Dependence of Z of nMOS Hara active inductors on the width of the transistor.
W is varied from 5μm to 25μm with step 5μm, R = 1kΩ. DC biasing current 0.5mA.
It becomes evident that Z has a zero at frequency ω
z
=
1
RC
gs
and a pole at
frequency ω
p
=
g
m
C
gs
. The network is resistive at low frequencies ω < ω
z
with
resistance R≈
1
g
m
and inductive when ω
z
< ω < ω
p
. Note the behavior of the
network beyond ω
p
can not be quantified by (2.89) due to the omission of C
gd
.
To derive its RLCequivalent circuit, we examine the input admittance of the
network
Y
in
=
sC
gs
+g
m
sRC
gs
+ 1
=
1
R
+
1
s
RC
gs
g
m

1
R
+
1
g
m

1
R
. (2.90)
Eq.(2.90) can be represented by a series RL network in parallel with a resistor
R
p
with
Implementation of Single-Ended Active Inductors 67
10
−1
10
0
10
1
10
2
−300
−200
−100
0
100
200
300
400
500
Frequency [GHz]
I
m
p
e
d
a
n
c
e

[
Ω
]
0.5 kΩ
2.5 kΩ
2.5 kΩ
0.5 kΩ
Re(Z
in
)
Im(Z
in
)
Figure 2.37. Dependence of Z of nMOS Hara active inductors on R. R is varied from 0.5 kΩ
to 2.5 kΩ with step 0.5 kΩ, W = 10μm. DC biasing current 0.5mA.
R
R
V
c
V
c
R
L
s
R
L
s
v
in
v
in
v
in
v
in
R
p
R
p
Figure 2.38. Simplified schematic of Wu folded active inductors.
R
p
= R,
L =
RC
gs
g
m

1
R
,
R
s
=
1
g
m

1
R
.
(2.91)
(a) nMOS (b) pMOS
68 CMOS Active Inductors
v
gs
g v
m
gs
C
gs
R
v
in
R
v
in
g
m
w
z
w
p
w
|z | (j ) w
R
0
Figure 2.39. Small-signal equivalent circuit of the nMOSversion of Wu folded active inductors.
It becomes evident that g
m
>
1
R
is required in order to have L, R
s
> 0. Also,
if g
m

1
R
, we have R
s

1
g
m
and L≈
RC
gs
g
m
=
R
ω
t
. They are the same as those
of the corresponding Hara active inductor investigated earlier.
2.3.7 Karsilayan-Schaumann Active Inductors
The active inductor proposed by Karsilayan and Schaumann is shown in
Fig.2.40 [74, 75, 48, 49, 61]. It was also developed by Yodprasit and Ngarmnil
in [85]. The active inductor consists of a differentially configured transcon-
ductor with a positive transconductance and a common-source transconductor
with a negative transconductance.
A. Lossless Karsilayan-Schaumann Active Inductors
The inductance of the inductor can be derived from the small-signal analysis
of the inductor with the assumption g
ds
= 0. The admittance of the active
inductor is given by
Y = sC
gs1
sC
gs2
+g
m2
s(C
gs1
+C
gs2
) + (g
m1
+g
m2
)
Implementation of Single-Ended Active Inductors 69
M1 M2
M3
V
C
C
gs2
b
Q
C
I
C
gs1
C
gs3
1
3
2
J
3
J
1
J
2
v
in
Figure 2.40. Simplified schematic of Karsilayan-Schaumann active inductor.
+
g
m2
g
m3
sC
gs3
sC
gs1
+g
m1
s(C
gs1
+C
gs2
) + (g
m1
+g
m2
)
. (2.92)
Further assume M
1
and M
2
are perfectly matched, i.e. g
m1
= g
m2
= g
m
and
C
gs1
= C
gs2
= C
gs
. Eq.(2.92) is simplified to
Y =
1
s
_
2C
gs3
g
m
g
m3
_
+s
_
C
gs
2
_
. (2.93)
Eq.(2.93) reveals that the active inductor can be represented by a capacitor in
parallel with an inductor. The capacitance and inductance are given by
C
p
=
C
gs
2
,
L =
2C
gs3
g
m
g
m3
.
(2.94)
It is interesting to note that the preceding results can also be obtained using
the results given in (2.9) directly. The differentially-configured transconduc-
tor with only one of its two input terminals is connected to the input has a
transconductance
g
m
2
. The capacitance encountered at the input node of the
active inductor is given by
C
gs
2
as capacitors of C
gs1
and C
gs2
are connected
in series.
Eq.(2.94) reveals that the inductance of the active inductor can be increased
by increasing the capacitance between the gate and source of M
3
. This can be
70 CMOS Active Inductors
achieved by adding an auxiliary capacitor C
I
in parallel with C
gs3
, as shown
in Fig.2.40. The inductance in this case becomes
L =
2(C
gs3
+C
I
)
g
m
g
m3
. (2.95)
The auxiliary capacitor C
I
can be implemented using MOS varactors. The
inductance of the active inductor can be tuned in this way.
g
o3
g v
m3
gs3
v
gs1
v
sg2
g v
m2
sg2
g
o1
g
o2
g v
m2
gs2
C
4
C
1
C
3
C
2
v
sg3
1
3
2
v
in
Figure 2.41. Small-signal equivalent circuit of Karsilayan-Schaumann active inductor.
B. Lossy Karsilayan-Schaumann Active Inductors
The preceding analysis excludes the effect of g
ds
and other parasitic capaci-
tances of the transistors. Due to the absence of the lossy conductance g
ds
, the
quality factor of the active inductor can not be studied. In what follows, we
investigate the quality factor of the active inductor and its tunability by follow-
ing the approach of Karsilayan and Schaumann. The small-signal equivalent
circuit of the active inductor is shown in Fig.2.41 where C
1
, C
2
, C
3
, and C
4
represent the total capacitances including both intrinsic and parasitic capaci-
tances encountered at or between respective nodes. It was shown in [74] that
the admittance of the active inductor is given by
Y ≈
1

_
2C
3
g
m
g
m3
_
+
G(ω)
g
m
g
m3
, (2.96)
where
G(ω) = g
o2
+ 2g
o4
−ω
2
C
3
_
C
1
+C
2
g
m
_
. (2.97)
Implementation of Single-Ended Active Inductors 71
The active inductor can be represented by an inductor of inductance
L =
2C
3
g
m
g
m3
(2.98)
in series with a resistor of resistance
R
s
=
G(ω)
g
m
g
m3
. (2.99)
The quality factor of the active inductor is obtained from
Q≈
ωL
R
s
=
2ωC
3
G(ω)
. (2.100)
It is observed from (2.100) that if we set G(ω) = 0, i.e.
C
2
=
(g
o2
+ 2g
o4
)g
m
ω
2
C
3
−C
1
, (2.101)
the quality factor of the active inductor will become infinite. To achieve this, an
auxiliary capacitor C
Q
can be added at the source of M
1
and M
2
, as shown in
Fig.2.40. An important observation is that Q is tuned by varying C
Q
, which is
the capacitance encountered at the source of M
1
and M
3
while the inductance
of the active inductor is tuned by varying C
I
, the capacitance of the auxiliary
capacitor added between the gate and source of M
3
. In other words, Q and L
can be tuned independently.
It was demonstrated in [74] that the quality factor of the active inductor was
made nearly 400 and the inductance of the active inductor exceeded 600 nH in
a 0.5μm CMOS implementation of Karsilayan-Schaumann active inductor.
C. Variations of Karsilayan-Schaumann Active Inductors
To increase the speed of the active inductor and to reduce the silicon con-
sumption, it was shown by Xiao and Schaumann in [61] that the preced-
ing Karsilayan-Schaumann active inductor can also be implemented using all
nMOS transistors (excluding biasing current sources), as shown in Fig.2.42.
Implemented in TSMC-0.18μm CMOS technology, this active inductor exhib-
ited a self-resonant frequency of 6.68 GHz and a quality factor of 106.
In [86, 87], the common-source configured transconductor of the preceding
Karsilayan-Schaumann active inductor was replaced with a static inverter to
boost the transconductance of the transconductor from g
m3
to g
m3
+ g
m4
, as
shown in Fig.2.43. The inductance is tuned by varying C
Q
while the quality
72 CMOS Active Inductors
M1 M2
M3
V
C
C
C
gs1
Cgs2
Cgs3
Q
I
b
J
3
J
2
J
1
v
in
1
2
Figure 2.42. Simplified schematic of Karsilayan-Schaumann active inductor using nMOS
transistors only.
factor is tuned by changing C
I
. The input voltage of the static inverter must
satisfy V
IL
≤v
2
≤V
IH
where V
IL
and V
IH
are the lower and upper voltage
bounds of the transition region of the static inverter, respectively, in order to
ensure that M
3
and M
4
are in the saturation. A disadvantage of this design is
the stringent constraint imposed on the voltage swing of node 2 of the active
inductor.
M1 M2
M3
V
C
C
Q
I
b
M4
v
in
v
o
V
OH
V
OL
V
OH
V
IL
V
IH
Transition
region
1
2
J
1
J
2
v
in
Figure 2.43. Simplified schematic of Karsilayan-Schaumann active inductor using a static-
inverter negative transconductor.
2.3.8 Yodprasit-Ngarmnil Active Inductors
It was pointed out earlier that to boost the quality factor, the effect of both
R
s
and R
p
of an active inductor must be compensated for. The L ∼ R
s
series
branch of the RLC equivalent circuit of an active inductor can be replaced with
Implementation of Single-Ended Active Inductors 73
the L ∼
ˆ
R
p
parallel branch shown in Fig.2.44(b) with the inductance and the
shunt resistance given by
ˆ
L = L(1 +
1
Q
2
),
ˆ
R
p
= R
p
(1 +Q
2
).
(2.102)
Both
ˆ
R
p
and R
p
can be combined into a single parallel resistor of resistance
R
p,total
=
ˆ
R
p
||R
p
.
M1 M2
M3
V
b
C -R L
p,total p
R
p
R
p
R
p,total
J
3
J
4
J
2
J
1
v
in
v
in
1
2 3
Figure 2.44. Simplified schematic of Yodprasit-Ngarmnil active inductor.
To boost the quality factor, a negative resistor of resistance −R
p,total
can
be connected in parallel with R
p,total
, as shown in Fig.2.44(b), such that the
net resistive loss of the active inductor vanishes. It was also shown earlier that
negative resistors are realized by using positive feedback. In Fig.2.44(a), the
added electrical connection between the input terminal of the active inductor
and the drain of M
2
forms the needed positive feedback. This is because an
increase in v
in
will result in an increase in i
D1
, subsequently a decrease in i
D2
as i
D1
+ i
D2
= J
2
(constant). This is echoed with an increase in the drain
voltage of M
2
, which will further increase v
in
. The impedance looking into
the gate of M
1
at low frequencies is given by
Z
in
≈−
_
1
g
m1
+
1
g
m2
_
. (2.103)
The preceding analysis reveals that the differential pair offers two distinct
functions simultaneously :
(a)
(b)
74 CMOS Active Inductors
It behaves as a transconductor with a negative transconductance to construct
the gyrator-C active inductor.
It provides the needed negative resistance between the input terminal and
the ground to cancel out the parasitic resistances of the active inductor.
When g
o
is considered, it was shown in [85] that the quality factor of
Yodprasit-Ngarmnil active inductor is given by
Q =
_
g
m3
g
m1
C
gs3
C
gs1
C
gs1
r
o1
+
2C
gs3
r
o3
. (2.104)
Eq.(2.104) shows that Qcan be tuned by either changing g
m1,2
or r
o1,2
. Because
the former also changes the inductance of the inductor, the preferred choice is
therefore to vary r
o
of M
1
and M
2
, which can be achieved by employing the
cascode configuration of the differential-pair transconductor and varying the
gate voltage of M
4
and M
5
, as shown in Fig.2.45. r
o1
and r
o2
in Fig.2.45 now
become (g
m4
r
o4
)r
o1
and (g
m5
r
o5
)r
o2
, respectively. Transconductances g
m4,5
can be tuned by varying the gate voltage of M
4,5
. As demonstrated in [85], Q
was tuned up to 12000 in a 0.6μm implementation of the active inductor.
M1 M2
M3
V
b
M5
V
M4
g
J
1
J
2
J
3
J
4
1
2
3
v
in
Figure 2.45. Simplified schematic of Cascode Yodprasit-Ngarmnil active inductors.
2.3.9 Uyanik-Tarim Active Inductor
The CMOS active inductor proposed by Uyanik and Tarim in [88] with its
simplified schematic shown in Fig.2.46 only has two transistors connected in
Implementation of Single-Ended Active Inductors 75
series between V
DD
and ground rails, making it very attractive for low-voltage
applications. As seen in Fig.2.46, M
1
and J form a transconductor with a
negative transconductance g
m1
. M
2,3,4
form a transconductor with a positive
transconductance
g
m2
g
m4
g
m3
= g
m2
, provided that M
3
and M
4
are identical. To
tune the inductance of the active inductor without affecting the parasitic series
resistance of the active inductor, which controls the quality factor of the active
inductor, a varactor C is added between the gate of M
2
and the ground.
v
in
M1
M2
M3 M4
J
C
Figure 2.46. Simplified schematic of Uyanik-Tarim active inductor.
Implemented in UMC-0.13μm 1.2V CMOS technology, the simulation re-
sults in [88] shows that the active inductor had a wide frequency range from
0.3 GHz to its self-resonant frequency of approximately 7.32 GHz. The quality
factor of the active inductor exceeded 100 in the frequency range 4.8-6.4 GHz
with its phase error less than 1 degree. The maximum quality factor was 3900,
occurring at 5.75 GHz. The minimum number of transistors stacked between
the power and ground rails also enabled the active inductor to have a large input
signal swing of 18 mV. The inductance was from 38 nH to 144 nH.
2.3.10 Carreto-Castro Active Inductors
The BiCMOS active inductors proposed by Carreto-Castro in [89] can also
be implemented in CMOS technologies, as shown in Fig.2.47. Neglecting C
gd
and g
o
of the transistor. The input impedance of nMOS Carreto-Castro active
inductor is given by
Z =
1
g
m
sRC
gs
+ 1
s
C
gs
g
m
+ 1
. (2.105)
The zero of Z(s) is at frequency
76 CMOS Active Inductors
ω
z
=
1
RC
gs
(2.106)
and the pole of Z(s) is at frequency
ω
p
=
g
m
C
gs
. (2.107)
The inductance of Carreto-Castro active inductor can be derived by examining
the input admittance
Y =
sC +g
m
sRC + 1
=
1
R
+
1
s
_
RC
g
m

1
R
_
+
1
g
m

1
R
. (2.108)
It is seen from (2.108) that the parameters of the RLC equivalent circuit of the
active inductor are given by
R
p
= R,
L =
R
2
C
Rg
m
−1
,
R
s
=
R
Rg
m
−1
.
(2.109)
In order to have L > 0 and R
s
> 0,
Rg
m
> 1 (2.110)
is required. This condition also ensures that ω
z
< ω
p
. In the frequency range
ω
z
< ω < ω
p
, the circuit is inductive. Because
ω
p
ω
z
= Rg
m
, (2.111)
for practical applications, Rg
m
1 is usually required to maximize the effective
frequency range. In this case
Implementation of Single-Ended Active Inductors 77
L≈
RC
g
m
=
R
ω
t
,
R
s

1
g
m
.
(2.112)
It is interesting to note that Hara active inductors, Wu folded active inductors,
and Carreto-Castro active inductors all have the same expressions for L and
R
s
.
R
R
(a) nMOS (b) pMOS
v
in
v
in
J
J
Figure 2.47. Simplified schematic of Carreto-Castro active inductors.
Figs.2.48 and 2.49 show the dependence of the resistance and reactance of
a nMOS Carreto-Castro active inductor on R and J. The active inductor was
implemented in TSMC-0.18μm 1.8V CMOS technology and analyzed using
SpectreRF with BSIM3V3 device models.
2.3.11 Thanachayanont-Payne Cascode Active Inductors
It was pointed out earlier that to maximize the frequency range of active
inductors, ω
z
should be minimized and ω
o
should be maximized. Maximizing
ω
o
is rather difficult because ω
o
of active inductors is set by the cut-off frequency
of the transconductors constituting the active inductors. The frequency of the
zero of active inductors given by ω
z
=
g
o1
C
gs2
, on the other hand, can be lowered
by either increasing C
gs2
or decreasing g
o1
. The former is at the cost of lowering
ω
o
and should therefore be avoided. To reduce g
o1
, Thanachayanont and Payne
proposed the cascode active inductors shown in Fig.2.50 [90, 65, 73]. Cascode
can be implemented in either of the two transconductors, as shown in Fig.2.50.
Note that a modification in the polarity of the transconductors is required to
ensure the existence of a negative feedback in the gyrator-C configuration.
The impedance of Thanachayanont-Payne cascode active inductor is given
by
78 CMOS Active Inductors
10
−1
10
0
10
1
10
2
−600
−400
−200
0
200
400
600
800
Frequency [GHz]
I
m
p
e
d
a
n
c
e

[
Ω
]
Im(Z
in
)
Re(Z
in
)
1 kΩ
2 kΩ
3 kΩ
4 kΩ
1 kΩ
2 kΩ
3 kΩ
4 kΩ
Figure 2.48. Dependence of Z of nMOS Carreto-Castro active inductor on R. Circuit param-
eters : L = 0.18μm, W = 10μm, R = 2kΩ.
10
−1
10
0
10
1
10
2
−400
−200
0
200
400
600
Frequency [GHz]
I
m
p
e
d
a
n
c
e

[
Ω
]
100 μA
200 μA
300 μA
400 μA
100 μA
200 μA
300 μA
400 μA
Im(Z
in
)
Re(Z
in
)
Figure 2.49. Dependence of Z of nMOS Carreto-Castro active inductor on J. Circuit param-
eters : L = 0.18μm, W = 10μm, R = 2kΩ.
Implementation of Single-Ended Active Inductors 79
M2
M1
V
M1
M2
M3
b
V
b
(a)
(b)
M2
M1
V
b1
M3
V
b2
(c)
v
in
J
1
J
2
v
in
v
in
J
1
J
1
J
2
J
2
Figure 2.50. Simplified schematic of Thanachayanont-Payne cascode active inductors. (a)
Basic active inductor. (b,c) Thanachayanont-Payne cascode active inductors.
Z≈
_
g
o1
g
o3
C
gs1
C
gs2
g
m3
_
s
_
C
gs2
g
m3
g
o1
g
o3
_
+ 1
s
2
+s
_
g
o1
g
o3
C
gs2
g
m3
+
g
o1
C
gs1
_
+
g
m1
g
m2
C
gs1
C
gs2
, (2.113)
where g
m
g
o
was utilized in simplifying the results. It is seen from (2.113)
that the frequency of the zero of Z is reduced from
ω
z
=
g
o1
C
gs2
(2.114)
without the cascode to
80 CMOS Active Inductors
ω
z
=
_
g
o1
C
gs2
_
1
g
m3
r
o3
(2.115)
with the cascode. The self-resonant frequency of the active inductor given by
ω
o
=
¸
g
m1
g
m2
C
gs1
C
gs2
=

ω
t1
ω
t2
, (2.116)
however, remains unchanged. It should not be surprised to see that cascodes
do not change ω
o
. This is because cascodes are not subject to Miller effect and
has no effect on the bandwidth. Cascode configurations thus can effectively
expend the frequency range of active inductors by lowering the lower bound of
the frequency range of active inductors.
The parameters of the RLCequivalent network of the cascode active inductor
can be obtained by examining the admittance of the cascode inductor and the
results are given by
R
p
=
1
g
o2
,
C
p
= C
gs1
,
R
s
=
_
g
o1
g
m1
g
m2
_
1
g
m3
r
o3
,
L =
C
gs2
g
m1
g
m2
.
(2.117)
It is evident from (2.117) that the cascode active inductor has the same induc-
tance as that of the corresponding non-cascode gyrator-C active inductor. The
parasitic series resistance is reduced from
R
s
=
g
o1
g
m1
g
m2
(2.118)
without the cascode to
R
s
=
g
o1
g
m1
g
m2
_
1
g
m3
r
o3
_
(2.119)
with the cascode. The parallel resistance is increased from
Implementation of Single-Ended Active Inductors 81
R
p
=
1
g
m1
(2.120)
without the cascode to
R
p
=
1
g
o2
. (2.121)
with the cascode. Both improve the quality factor of the active inductor, as is
evident from
Q≈
ωL
R
s
= ωC
gs2
r
o1
(R
s
dominates) (2.122)
or
Q≈
R
p
ωL
=
g
m2
ωC
gs2
(R
p
dominates) (2.123)
without the cascode and
Q≈
ωL
R
s
= ωC
gs2
r
o1
(g
m3
r
o3
) (R
s
dominates) (2.124)
or
Q≈
R
p
ωL
= (r
o2
g
m1
)
g
m2
ωC
gs2
(R
p
dominates) (2.125)
with the cascode. To summarize, the cascode configurations of gyrator-C active
inductors offer the following attractive characteristics :
Frequency range expansion by lowering the lower bound of the frequency
range.
Quality factor improvement by lowering the parasitic series resistance R
s
and increasing the parasitic parallel resistance R
p
.
No reduction in the upper bound of the frequency range.
No reduction in the inductance.
The minimum supply voltage of the active inductor without the cascode
is given by V
T
+ 2V
sat
. For the cascode active inductor of Fig.2.50(b),
V
DD,min
= 2V
T
+ V
sat
. For the cascode active inductor of Fig.2.50(c),
V
DD,min
= V
T
+ 2V
sat
.
82 CMOS Active Inductors
2.3.12 Weng-Kuo Cascode Active Inductors
Weng and Kuo proposed a current-reuse cascode active inductor in [62] to
eliminate the drawback of the preceding Thanachayanont-Payne cascode active
inductors that the inductance and quality factor can not be tuned independently.
The simplified schematic of Weng-Kuo active inductor is shown in Fig.2.51.
It is seen that g
m1
is proportional to J
1
+J
3
while g
m3
is only proportional to
J
1
. Because L =
C
gs2
g
m1
g
m2
, R
s
=
g
o1
g
o3
g
m1
g
m2
g
m3
, C
p
= C
gs1
, and R
p
=
1
g
o2
, we
have
ω
o
=
¸
g
m1
g
m2
C
gs1
C
gs2
,
Q(ω
o
) =
ω
o
L
R
s
=
g
m3
g
o1
g
o3
¸
g
m1
g
m2
C
gs2
C
gs1
.
(2.126)
ω
o
can be tuned by varying g
m1
and g
m2
while Q can be tuned by varying g
m3
only. So the tuning of Qcan be made independent of ω
o
. Note that because the
tuning of ω
o
, however, will affect Q, an adjustment of Q is therefore required
after each tuning of ω
o
.
M1
M3
V
J
b
1
J
2
J
3
M2
v
in
Figure 2.51. Simplified schematic of Weng-Kuo active inductor.
2.3.13 Manetakis Regulated Cascode Active Inductors
The performance of the preceding cascode active inductors can be improved
by further reducing R
s
. This is achieved by using regulated cascodes and
multi-regulated cascodes, as shown in Fig.2.52 [91]. The parameters of the
Implementation of Single-Ended Active Inductors 83
RLC equivalent circuit of the regulated cascode gyrator-C active inductor are
given by
G
o1
=
1
g
o1
(r
o3
g
m3
)(r
o4
g
m4
)
,
C
1
= C
gs2
,
G
o2
= g
o2
,
C
2
= C
gs1
,
(2.127)
from which we obtain
R
p
=
1
G
o2
,
C
p
= C
2
,
R
s
=
G
o1
g
m1
g
m2
,
L =
C
1
g
m1
g
m2
.
(2.128)
Because ω
z
=
G
o1
C
1
and ω
o
=
1
_
LC
p
, the lower bound of the frequency is
reduced while the upper bound of the frequency range remains unchanged.
The parameters of the RLCequivalent circuit of the multi-regulated cascode
gyrator-C active inductor are given by
G
o1
=
1
g
o1
(r
o3
g
m3
)(r
o4
g
m4
)(r
o5
g
m5
)
,
C
1
= C
gs2
,
G
o2
= g
o2
,
C
2
= C
gs1
.
(2.129)
It is evident from (2.129) that the lower bound of the frequency is further
reduced while the upper bound of the frequency range remains unchanged.
84 CMOS Active Inductors
M1
M2
M3
M4
M1
M2
M3
M4
M5
V
b
v
in
v
in
J
1
J
3
J
2
J
2
J
3
J
1
-A
v
in
v
o
v
in
v
o
-A
Figure 2.52. Simplified schematic of Manetakis regulated and multi-regulated cascode active
inductors. (a) Regulated cascode active inductors. (b) Multi-regulated cascode active inductors.
The simulation results of cascode active inductors, regulated cascode active
inductors and multi-regulated cascode active inductors given in [91] demon-
strated that regulated cascode reduced the lower frequency bound of active
inductors by one decade while multi-regulated cascode further reduced the
lower frequency bond by more than one decade.
2.3.14 Hsiao Feedback Resistance Cascode Active
Inductors
It was shown earlier that cascode active inductors offer a large frequency
range and a high quality factor. To further improve the quality factor, feedback
resistance active inductors shown in Fig.2.53 were proposed by Hsiao et al. in
[92]. This type of active inductors was further investigated in [93, 94]. By
assuming that the biasing current source transistors M
n
and M
p
are ideal, the
parameters of the RLC equivalent circuit of the active inductor are given by
C
p
= C
gs3
,
R
p
=
R
f
g
o2
+ 1
2g
o2
+R
f
g
2
o2
,
R
s
=
g
m1
g
o2
g
o3

2
[g
m2
C
2
gs1
−g
m1
C
gs1
C
gs2
(R
f
g
o2
+ 1)]
g
2
m1
g
m2
g
m3

2
g
m2
g
m3
C
2
gs1
,
L =
g
m1
g
m2
C
gs1

2
C
2
gs1
C
gs2
(R
f
g
o2
+ 1)
g
2
m1
g
m2
g
m3

2
g
m2
g
m3
C
2
gs1
.
(2.130)
(b) (a)
Implementation of Single-Ended Active Inductors 85
It is seen from (2.130) that the added feedback resistor R
f
lowers R
s
and
increases Lsimultaneously. Both boost the quality factor of the active inductor.
The resistance of the feedback resistor R
f
can be tuned by connecting a nMOS
transistor in parallel with a poly resistor, as shown in the figure. Note R
f
does
not consume any static power. For the special case where R
f
= 0, Eq.(2.130)
becomes
C
p
= C
gs3
,
R
p

1
g
o2
,
R
s

g
o2
g
o3
g
m1
g
m2
g
m3
,
L =
C
gs2
g
m1
g
m2
.
(2.131)
It should be noted that the self-resonant frequency of the active inductor with
the feedback resistors R
f
is reduced due to the increase of L. It was shown in
[94] that the decrease of the self-resonant frequency of the feedback resistance
active inductors can be compensated for by varying the biasing voltage V
p
of the
current-source transistor M
p
. The simulation results of a feedback resistance
active inductor implemented in a 0.18μm CMOS technology showed that the
inductance of the inductor was 15 nH with the quality factor exceeding 50 and
the self-resonant frequency of several GHz [92–94].
The preceding Hsiao feedback resistance cascode active inductors were
further developed by Liang et al. where the cascode portion of the active
inductor is replaced with a regulated cascode branch, as shown in Fig.2.54
so that the advantages of the regulated cascodes active inductor investigated
earlier can be utilized [50].
2.3.15 Abdalla Feedback Resistance Active Inductors
Karsilayan-Schaumann active inductors studied earlier offer the key advan-
tage of the independent tuning of their inductance and quality factor. In [95],
Abdalla et al. modified Karsilayan-Schaumann active inductors by adding a
feedback resistor between the two transconductors of the active inductor to
improve the quality factor of the inductor, as shown in Fig.2.55. The added
feedback resistor increases the inductance of the active inductor and at the same
time lowers the parasitic series resistance of the active inductor, thereby boost-
ing the quality factor of the active inductor. C
I
and C
Q
are MOS varactors to
tune the inductance and quality factor of the active inductor, respectively.
86 CMOS Active Inductors
M1
M2
M3
V
b
R
M1
M2
(a) Feedback resistance
active inductor
R
V
f
(b) Cascode feedback
resistance active inductor
Mp
p
Mp
p
Mn Mn
V
V
V
n
V
n
f
f
v
in
v
in
Figure 2.53. Simplified schematic of Hsiao feedback resistance active inductors.
M1
M2
M3
R
f
v
in
M4
J
1
J
2
J
3
Figure 2.54. Simplified schematic of Liang feedback resistance regular cascode active induc-
tors.
2.3.16 Nair Active Inductors
Wei et al. proposed a MESFET-version high-Q active inductor with loss
compensated by a feedback network in [96]. This active inductor was modified
and implemented in CMOS by Nair et al. in [97] and was used in design of a
low-power low-noise amplifier for ultra wideband applications. The simplified
schematic of Nair active inductor is shown in Fig.2.56. It is a cascode active
inductor with a negative feedback network consisting of R
1
, R
2
, and C
2
. It
was shown in [97] that the parameters of this active inductor are given by
Implementation of Single-Ended Active Inductors 87
M1 M2
M3
V
C
C
Q
I
b
J
3
J
2
J
1
v
in
1
2
R
f
Figure 2.55. Simplified schematic of single-ended Abdalla feedback resistance active induc-
tors.
L =
C
3
+C
gs3
g
m1
g
m3
,
ω
o
=
¸
g
m1
g
m3
C
gs1
(C
3
+C
gs3
)
,
Q(ω
o
) =
¸
g
m1
g
m3
(C
3
+C
gs3
)
C
gs1
g
2
o1
.
(2.132)
It is seen from (2.132) that C
3
is used to boost the inductance. It also increases
the quality factor and decreases the self-resonant frequency of the active induc-
tor. The RC network consisting of R
1
, R
2
−C
2
is a negative feedback network
that reduces the parasitic resistances of the active inductor.
2.3.17 Active Inductors with Low Supply-Voltage
Sensitivity
It was pointed out earlier that the parameters of gyrator-C active inductors,
such as the inductance and parasitic resistances, are sensitive to the fluctuation
of the supply voltage of the active inductors. This is because a V
DD
fluctuation
not only alters the dc operating point of the transconductors constituting the ac-
tive inductors, it also changes the junction capacitances of the active inductors.
Replica-biasing is an effective means to reduce the effect of supply voltage
fluctuation on the inductance of active inductors. Fig.2.57 shows the config-
uration of Wu current-reuse active inductors with replica biasing [66]. The
88 CMOS Active Inductors
v
in
M2
R
1
J
1
M3
J
2
M1
R
2
C
1
C
3
C
2
Figure 2.56. Simplified schematic of Nair active inductors.
replica-biasing section consists of a sensing circuit made of M
4,5,6
and an aux-
iliary voltage amplifier. An increase in V
DD
will lead to an increase in v
SG3,6
,
subsequently the channel current of M
3,6
. The voltage of the non-inverting
terminal of the amplifier will also increase. The output of the amplifier will
increase the gate voltage of M
3,6
, which ensures that v
SG3,6
is kept unchanged
approximately, minimizing the effect of V
DD
fluctuation. The width of the
transistors in the replica-biasing section should be the same as that of the active
inductor section so that both will sense the same voltage change caused by the
variation of V
DD
.
M1
b1
V
b2
V
M2
M3
M4
b1
V
M5
M6
v
in
Figure 2.57. Simplified schematic of Wu current reuse active inductor (nMOS) with replica
biasing.
Implementation of Differential Active Inductors 89
2.4 Implementation of Differential Active Inductors
2.4.1 Lu Floating Active Inductors
The schematic of the active inductor proposed by Lu et al. is shown in
Fig.2.58 [60]. It is a differentially configured gyrator-C active inductor. Tran-
sistors M
5
and M
6
are biased in the triode region and behave as voltage-
controlled resistors whose resistances are controlled by the gate voltage V
b
. All
other transistors are biased in the saturation.
The negative feedback of the active inductor is as the followings : an increase
in v
1+
and a decrease in v
1−
will result in an increase in v
2+
and a decrease
in v
2−
due to the common-gate operation of M
1,2
. The source followers of
M
3,4
ensure that v
1+
and v
1−
will be reduced accordingly by approximately
the same amount.
To find out the parameters of the RLCequivalent circuit of the active inductor,
we represent M
5
and M
6
with channel conductances g
ds5
and g
ds6
, respectively,
in the small-signal equivalent circuit of the active inductor. It can be shown
that the differential input impedance of the inductor is given by
Z =
2[s(C
gs1
+C
gs3
) −g
m1
+g
ds5
]
g
ds5
[g
m1
+g
m3
+s(C
gs1
+C
gs3
)]
. (2.133)
The parameters of the RLC equivalent circuit of the active inductor are given
by
R
p
=
2
g
ds5
,
R
s
=
2(g
ds5
−g
m1
)
g
ds5
(2g
m1
+g
m3
−g
ds5
)
,
L =
2(C
gs1
+C
gs3
)
g
ds5
(2g
m1
+g
m3
−g
ds5
)
.
(2.134)
Note that 2g
m1
+ g
m3
> g
ds5
and g
ds5
> g
m1
are required in order for R
s
and L to have a positive value. The preceding development reveals that the
inductance of the active inductor can be tuned by varying V
b
, subsequently
g
ds5
. An increase in V
b
will push M
5
and M
6
from the triode region towards
the saturation region, lowering g
ds5
and g
ds6
. This is echoed with an increase
in the inductance of the inductor.
2.4.2 Gr¨ ozing Floating Active Inductors
As pointed out earlier that floating gyrator-C active inductors can be con-
structed using a pair of differential transconductors. One of the simplest differ-
90 CMOS Active Inductors
M1 M2
V
b
M3 M4
M5 M6
R
R
L
p
s
v
in+
v
in-
1+
1-
2+
2-
v
in+
v
in-
J
1
J
2
Figure 2.58. Simplified schematic of Lu floating active inductor.
entially configured transconductors is the basic differential pair. The floating
active inductor proposed in [43, 44] and shown in Fig.2.59 employs two basic
differential-pair transconductors. Two negative resistors are connected across
the output nodes of the transconductors to cancel out the parasitic resistances
of the active inductor. The inductance of the active inductor is tuned by varying
the transconductances of the transconductors. This is done by adjusting the tail
currents of the differential pairs J
1,2
. The quality factor of the active inductor
is tuned by varying the resistances of the negative resistors. This is attained by
changing the tail current sources J
3,4
of the negative resistors. It was shown
in [43, 44] that in a 0.18μm implementation of the active inductor, the quality
factor of the active inductor was 600 at 2 GHz while the self-resonant frequency
of the active inductor was 5.6 GHz
2.4.3 Thanachayanont Floating Active Inductors
The floating active inductor proposed in [37, 38, 98, 99, 65, 73] is shown
in Fig.2.60. It consists of two cascode-configured gyrator-C active inductors
investigated earlier. Transistors M
4,5
form a negative resistor to cancel out the
parasitic resistances of the active inductor so as to boost its quality factor. Note
that the resistance of the negative resistor can not be tuned in this implementa-
tion. The inductance of the active inductor is tuned by varying J
2
. The active
inductor implemented in a 0.35μm CMOS technology offered an inductance of
70 nH and a self-resonant frequency of 2.8 GHz. The quality factor exceeded
100 over the frequency range from 0.83 GHz to 1.33 GHz with its maxima of
1970 at 1.08 GHz.
Implementation of Differential Active Inductors 91
M1
M2
M5
M6 v
-R
-R
V
b1
V
b2
in+
v
in-
J
1
J
2
J
3,4
M3
M4
M7
M8
Figure 2.59. Simplified schematic of Gr¨ ozing floating active inductor.
V
b1
M1
M3
V
b2
M2 M2 M1
M3
J
1
J
1
J
2
J
2
V
b1
M4
M5
v
in
Figure 2.60. Simplified schematic of Thanachayanont floating active inductor.
2.4.4 Mahmoudi-Salama Floating Active Inductors
The floating active inductor proposed by Mahmoudi and Salama was used in
the design of quadrature down converter for wireless applications [56, 100]. The
schematic of Mahmoudi-Salama floating active inductor is shown in Fig.2.61.
It consists of a pair of differential transconductors and a pair of negative resistors
92 CMOS Active Inductors
at the output of the transconductors. M
8,16
are biased in the triode and behave as
voltage-controlled resistors. They are added to the conventional cross-coupled
configuration of negative resistors to provide the tunability of the resistance of
the negative resistors without using a tail current source.
V
b1
V
b2
M1
M4 M5
M2 M3
V
b1
V
b2
V
b3
b3
V
M6 M7
M8
2C 2C
2C 2C
v
in+
v
in-
A
B
M9
M10 M11
M12 M13
M14 M15
M16
Figure 2.61. Simplified schematic of Mahmoudi-Salama floating active inductor.
The small-signal equivalent circuit of the tunable negative resistor is shown
in Fig.2.62 where a test voltage source V
x
is added for the derivation of the
equivalent resistance of the negative resistor. R represents the resistance of
M
8
. Writing KCL at nodes 1 and 2 yields
g
m1
V
2
+
V
x
R
−I
x
= 0 (node 1),
g
m1
(V
x
+V
2
) −
V
x
R
+I
x
= 0 (node 2).
(2.135)
The resistance of the negative resistors at low frequencies is obtained from
(2.135)
Implementation of Differential Active Inductors 93
ht
g v
m2
1
2
R
1
g v
m1 2
V
x
v
2
v
1
I
R
1 2
M1 M2
x
Figure 2.62. Small-signal equivalent circuit of Mahmoudi-Salama floating active inductor at
low frequencies. g
o
of the transistors is neglected.
Z =
V
x
I
x
= −
R
_
1
g
m1
+
1
g
m2
_
R −
_
1
g
m1
+
1
g
m2
_
= R//
_

_
1
g
m1
+
1
g
m2
_
_
. (2.136)
The inductance of the active inductor is given by L =
C
G
m1
G
m2
, where
2C is the total capacitance encountered at each of the output nodes of the
transconductor, G
m1
and G
m2
are the transconductances of the differential
transconductors 1 and 2, respectively. By assuming that nodes A and B are the
virtual ground and neglecting C
gd
and the diffusion junction capacitances, we
have C≈
C
gs2,3,10,11
2
and G
m
= g
m2,3,10,11
.
2.4.5 Feedback Resistance Floating Active Inductors
The feedback resistance technique studied earlier was also employed in the
design of floating active inductors by Akbari-Dilmaghani et al. in [101] to
improve the performance of these inductors. A similar approach was used by
Abdalla et al. in design of high-frequency phase shifters [102]. This section
investigates these active inductors.
The schematic of the feedback resistance floating active inductor proposed
by Akbari-Dilmaghani et al. is shown in Fig.2.63. It consists of two basic
differential-pair transconductors and two feedback resistors. The functionality
of the added feedback resistors is the same as that of the single-ended active
94 CMOS Active Inductors
inductors investigated earlier, i.e. lowering the parasitic series resistance and
increasing the inductance.
V
b1
V
b2
M4 M5
M2 M3
V
b1
V
b2
R
f
R
f
M1
M6
M9 M10
M8
M7
v
in+
v
in-
Figure 2.63. Simplified schematic of Akbari-Dilmaghani feedback resistance floating active
inductor.
The simplified schematic of Abdalla differential feedback resistance floating
active inductors is shown in Fig.2.64, M
11,12
are biased in the triode and behave
as voltage-controlled resistors. It was shown in [102] that the inductance and
the parasitic series resistance of the floating active inductor are given by
L =
C +C
gs4,5
_
1 +
R
f
R
T
_
g
m1,2
g
m4,5
,
R
s
=
1
R
T
−ω
2
C
gs4,5
CR
f
g
m1,2
g
m4,5
,
(2.137)
where
Class AB Active Inductors 95
C = C
gd7,8
+C
db1,2
+C
db7,8
+C
gs1,2
,
R
T
= R
f
||R
ds11,12
||r
o1,2
||r
o7,8
.
(2.138)
It is evident from (2.137) that R
f
boosts L and lowers R
s
simultaneously.
Both improve the performance of the floating active inductor. Also seen from
(2.137) and (2.138) that M
11,12
control the series resistance R
s
of the active
inductor. By adjusting V
b1
, R
s
can be minimized.
V
b2
M7 M8
M1 M2
V
b2
R
f
R
f
M3
M6
M9 M10
M5 M4
M12
M11
V
b1
V
b1
v
in+
v
in-
Figure 2.64. Simplified schematic differential Abdalla feedback resistance floating active in-
ductors.
2.5 Class AB Active Inductors
The active inductors investigated up to this point fall into the category of
class A active inductors. These class A active inductors suffer from a common
drawback of a small voltage swing at the terminals of the active inductors,
mainly due to the constraint that the input transistors of the transconductors of
the active inductors should be biased and operated in the saturation. Note that
96 CMOS Active Inductors
although the input transistors of the transconductors of active inductors can be
pushed into the triode region while still exhibiting an inductive characteristic,
the nonlinearity of the active inductors will increase.
To increase the voltage swing of the active inductors without sacrificing
linearity, class ABconfigurations can be employed. Aclass ABactive inductor
can be constructed from a nMOS-configured class A active inductor and a
pMOS-configured class A active inductor, as shown in Fig.2.65 [65]. When
the input voltage is high, only the nMOS class A active inductor is activated.
When the input voltage is low, only the pMOS class A active inductor is
activated, i.e. the two active inductors are operated in an interleave manner,
depending upon the swing of the input voltage. The network thus exhibits
an inductive characteristic over a large input voltage range. The swing of
the input voltage and the minimum supply voltage of the class A and class
AB active inductors are compared in Table 2.4. It is seen that the need for
a large minimum supply voltage of class AB active inductors makes them
less attractive for low-voltage applications. It should also be noted that when
V
T
+V
sat
≤v
in
≤V
DD
−(V
T
+V
sat
), both the nMOS and pMOS class A active
inductors are activated.
V
b1
V
b2
V
b1
V
b2
(a) Class A active inductors (b) Class AB active inductor
v
in
v
in
v
in
M1
M2
J
1
J
2
J
2
J
1
M1
M2
M1
M2
M3
M4
J
1
J
2
Figure 2.65. Simplified schematic of Thanachayanont-Ngow class AB active inductor.
Chapter Summary 97
Table 2.4. A comparison of the input voltage swing and the minimum supply voltage of class
A and class AB active inductors.
Class A (nMOS) Class A (pMOS) Class AB
V
in,min
V
sat
V
T
+V
sat
0
V
in,max
V
DD
−(V
T
+V
sat
) V
DD
−V
sat
V
DD
V
DD,min
V
T
+ 2V
sat
V
T
+ 2V
sat
2V
T
+ 2V
sat
Class AB active inductors can also be configured in cascodes to increase
the frequency range and to boost the quality factor, as shown in Fig.2.66.
Floating cascode class AB active inductors can also be constructed in a similar
manner as that of non-cascode floating active inductors, as shown in Fig.2.67.
Negative resistance compensation techniques can be employed to improve the
performance of class AB active inductors.
V
b1
V
b1
V
b1
V
b2
(a) Cascode class A active inductors (b) Cascode class AB active inductor
V
b3
V
b4
V
b2
V
b2
v
in
v
in
v
in
J
1
J
2
M1
M2
M3
M1
M2
M3
J
1
J
2
M1
M2
M3
M4
M5
M6
Figure 2.66. Simplified schematic of Thanachayanont-Ngow cascode class ABactive inductor.
98 CMOS Active Inductors
V
b1
V
b2
V
b3
V
b4
V
b3
V
b4
V
b1
V
b2
J
2
J
1
J
1
J
2
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
M13 M14
M15 M16
Figure 2.67. Simplified schematic of Thanachayanont-Ngow cascode floating class AB active
inductors.
2.6 Chapter Summary
An in-depth examination of the principles, topologies, characteristics, and
implementation of gyrator-C active inductors in CMOS technologies has been
presented. We have shown that both single-ended and floating (differential) ac-
tive inductors can be synthesized using gyrator-C networks. Lossless gyrator-C
networks yield lossless active inductors while lossy gyrator-C networks syn-
thesize lossy active inductors. To provide a quantitative measure of the perfor-
mance of active inductors, a number of figure-of-merits have been introduced.
These figure-of-merits include frequency range, inductance tunability, quality
factor, noise, linearity, stability, supply voltage sensitivity, parameter sensi-
tivity, signal sensitivity, and power consumption. Frequency range specifies
the lower and upper bounds of the frequency in which gyrator-C networks are
inductive. We have shown that the low frequency bound is set by the frequency
of the zero of the gyrator-C networks while the upper frequency bound is set by
the frequency of the pole of the networks. Both are due to the finite input and
output impedances of the transconductors constituting active inductors. One of
the key advantages of active inductors over their spiral counterparts is the large
tunability of their inductance. We have shown that the inductance of gyrator-C
active inductors can be tuned by varying either the transconductances of the
transconductors or the load capacitance. The former is often used for the coarse
control of the inductance whereas the latter is used for the fine control of the
inductance. The distinct sensitivities of the quality factor of active inductors to
their parasitic series and parallel resistances have been investigated. We have
Chapter Summary 99
shown that the parasitic series resistance of an active inductor can be converted
to an equivalent parallel parasitic resistance such that a negative shunt resistor
can be employed to cancel out the effect of both the series and parallel parasitic
resistances of the active inductor simultaneously. An emphasis has been given
to the noise of active inductors as these inductors exhibit a high level of noise
power. The input-referred noise generators of basic transconductors have been
derived using the approach for noise analysis of 2-port networks whereas those
of active inductors have been obtained using the approach for noise analysis
of 1-port networks. The linearity of active inductors has been investigated.
Because active inductors are active networks that are sensitive to both sup-
ply voltage fluctuation and parameter spread. We have used supply voltage
sensitivity and parameter sensitivity to quantify the effect of these unwanted
variations. A distinct characteristic of active inductors is the dependence of
their performance on the swing of their signals. When active inductors are used
in applications such as LC oscillators where signal swing is large, the effect of
this dependence must be accounted for.
The second part of the chapter has focused upon the CMOS implementation
of gyrator-C active inductors. The schematics and characteristics of single-
ended and floating (differential) class Aactive inductors have been investigated
in detail. Class ABgyrator-Cactive inductors have also been studied. The basic
gyrator-C active inductors suffer from a small parasitic parallel resistance and a
large parasitic series resistance due to the small output / input impedance of the
transconductors constituting the active inductors. This in turn lowers the quality
factor of the active inductors. To increase the parallel parasitic resistance and to
lower the parasitic series resistance, transconductors with a large input / output
resistance are critical. This can be achieved by changing the configuration
of transconductors, such as Karsilayan-Schaumann active inductors that have
a large output impedance, employing cascodes, or adding parallel negative
resistors. To increase the tunability of active inductors, additional voltage-
controlled capacitors can be employed at critical nodes of active inductors.
This approach is often preferable over those that vary the transconductances
of the transconductors of active inductors because the latter also alters other
parameters of the active inductors such as inductance with a downside that the
tuning range is rather small. Feedback resistance active inductors, both single-
ended and floating, exhibit improved performance by increasing the inductance
and decreasing the series parasitic resistance simultaneously. Both improve
the quality of the active inductors. To increase the voltage swing of the active
inductors without sacrificing linearity, class AB active inductors constructed
from a pair of nMOS/pMOS-configured class A active inductors have been
presented. Two main drawbacks of class AB active inductors are the circuit
complexity of these inductors and their need for a high supply voltage.
http://www.springer.com/978-0-387-76477-1

22
Y = Iin = V2

CMOS Active Inductors

1 . (2.1) C s Gm1 Gm2 Eq.(2.1) indicates that port 2 of the gyrator-C network behaves as a single-ended lossless inductor with its inductance given by C . (2.2) Gm1 Gm2 Gyrator-C networks can therefore be used to synthesize inductors. These synthesized inductors are called gyrator-C active inductors. The inductance of gyrator-C active inductor is directly proportional to the load capacitance C and inversely proportional to the product of the transconductances of the transconductors of the gyrator. Also, the gyrator-C network is inductive over the entire frequency spectrum. It should also be noted that the transconductor in the forward path can be configured with a negative transconductance while the transconductor in the feedback path has a positive transconductance, as shown in Fig.2.1(b). Although the transconductors of gyrator-C networks can be configured in various ways, the constraint that the synthesized inductors should have a large frequency range, a low level of power consumption, and a small silicon area requires that these transconductors be configured as simple as possible. Fig.2.2 shows the simplified schematics of the basic transconductors that are widely used in the configuration of gyrator-C active inductors. Common-gate, common-drain, and differential-pair transconductors all have a positive transconductance while the common-source transconductor has a negative transconductance. To demonstrate this, consider the common-gate transconductor. An increase in vin will lead to a decrease in iD . Because io = J − iD , io will increase accordingly. So the transconductance of the common-gate transconductor is positive. Similarly, for the differential-pair transconductor in Fig.2.2(d). An increase in vin will result in an increase in iD1 . Since iD2 = J3 − iD1 , iD2 will decrease. Further io = J2 − iD2 , io will increase. The differential-pair transconductor thus has a positive transconductance. L=

2.1.2

Lossless Floating Gyrator-C Active Inductors

An inductor is said to be floating if both the terminals of the inductor are not connected to either the ground or power supply of the circuits containing the active inductor. Floating gyrator-C active inductors can be constructed in a similar way as single-ended gyrator-C active inductors by replacing singleended transconductors with differentially-configured transconductors, as shown in Fig.2.3. Because

Principles of Gyrator-C Active Inductors
Gm1V2

23

I in V in

2

V2 1
Gm2V1

C V1 V in

2 I in L or 2 L I in V in

(a)
Gm1V2

I in V in

2

V2 1
Gm2V1

C V1

(b)
Figure 2.1. Lossless singe-ended gyrator-C active inductors. Gm1 and Gm2 are the transconductances of transconductors 1 and 2, respectively, and C is the load capacitance at node 1. (a) Transconductor in the forward path has a positive transconductance while the transconductor in the feedback path has a negative transconductance; (b) Transconductor in the forward path has a negative transconductance while the transconductor in the feedback path has a positive transconductance.

+ Vin1 = − − Vin1 =

gm1 + − (V − Vin2 ), sC in2 (2.3)

gm1 + − (V − Vin2 ), sC in2

+ − Io2 = gm2 (Vin1 − Vin1 ),

we have Io2 = − 2gm1 gm2 + − (Vin2 − Vin2 ). sC (2.4)

The admittance looking into port 2 of the gyrator-C network is given by

24

CMOS Active Inductors

J v in

J io Vb

io

v in J

io

v in (a) (b)

(c)

J1 vin M1 J3

J2 M2

io Vb v in M1 J M2

io Vb

(d)

(e)

Figure 2.2. Simplified schematic of basic transconductors. (a) Common-source transconductor io = −gm vin ; (b) Common-gate transconductor io = gm vin ; (c) Common-drain (source follower) transconductor io = gm vin ; (d,e) Differential-pair transconductors io = gm vin .

Y

= =

Iin − − Vin2 1 . 2C s gm1 gm2
+ Vin2

(2.5)

Eq.(2.5) reveals port 2 of the gyrator-C network behaves as a floating inductor with its inductance given by L= 2C . gm1 gm2 (2.6)

(ii) The level of the voltage swing of floating active inductors is twice that of the corresponding single-ended active inductors. Lossless floating gyrator-C active inductors.V1. To simplify analysis. the synthesized inductors are no longer lossless. Note Go1 is due to the finite output impedance of transconductor 1 and the finite input impedance of transconductor 2. (node 2). Consider the gyrator-C network shown in Fig. Write KCL at nodes 1 and 2 (sC1 + Go1 )V1 − Gm1 V2 = 0 (node 1).Principles of Gyrator-C Active Inductors 25 2V in 2+ I in Gm1 (V2+.1. respectively. making them particularly attractive for applications where both analog and digital circuits are fabricated on the same substrate. and C is the load capacitance at nodes 1+ and 1-. Floating gyrator-C active inductors offer the following attractive advantages over their single-ended counterparts : (i) The differential configuration of the transconductors effectively rejects the common-mode disturbances of the network.2.) Figure 2.4 where Go1 and Go2 denote the total conductances at nodes 1 and 2. (2. Gm1 and Gm2 are the transconductances of transconductors 1 and 2. The admittance looking into port 2 of the gyrator-C network is obtained from . the gyrator-C networks are inductive only in a specific frequency range. respectively.V 2.) Io1 11+ Io2 C C 2+ V in 2L Gm2 (V1+. we continue to assume that the transconductances of the transconductors are constant.3 Lossy Single-Ended Gyrator-C Active Inductors When either the input or the output impedances of the transconductors of gyrator-C networks are finite.7) −Iin + (sC2 + Go2 )V2 − Gm2 (−V1 ) = 0.3. 2. Also.

respectively. Rs = L= Go1 . We comments on the preceding results : . Go2 Cp = C2 .9) C1 . lossy single-ended gyrator-C active inductors.8) can be represented by the RLC networks shown in Fig.4 with its parameters given by Rp = 1 . C2 and Go2 denote the total capacitances and conductances at nodes 1 and 2. Gm1 Gm2 Vin 2 Iin Rs Gm1V2 Rp Cp L Vin Iin 2 V2 1 Gm2V1 G o2 C2 V1 G o1 C1 Rs Rp Cp L Vin 2 Iin Figure 2.2. C1 Go1 s + Gm1 Gm2 Gm1 Gm2 (2.8) Eq.26 Iin V2 CMOS Active Inductors Y = = sC2 + Go2 + 1 .4. Gm1 Gm2 (2. C1 and Go1 .(2.

2 (2. however. ωo is the self-resonant frequency of the gyrator-C active inductor. however. can not be used to quantify the large-signal behavior. The RLC equivalent circuit of gyrator-C active inductors. and series resistance Rs . have no effect on the inductance of the active inductor.10) ωt1. Rp should be maximized while Rs should be minimized to low the ohmic loss.2 C1. The self-resonant frequency of an active inductor is set by the cut-off frequency of the transconductors constituting the active inductor. C1 C2 ωo = where (2. 2 . giving 2 rise to a phase error. such as the dependence of the inductance on the dc biasing condition of the transconductors and the maximum signal swing of the gyrator-C active inductors. when these conductances are non-zero.11) is the cut-off frequency of the transconductors.2 = Gm1. When the conductances encountered at nodes 1 and 2 of the gyrator-C active inductors are zero (lossless). The phase of the impedance of practical active inductors should π be made constant and to be as close as possible to . The phase error is due to Rp and Rs of the active inductors. the phase 2 π of the impedance of the synthesized inductor will deviate from . Rp and Cp are solely due to Go2 and C2 . The finite input and output impedances of the transconductors of the gyrator-C network. However.Principles of Gyrator-C Active Inductors 27 When the input and output conductances of the transconductors are considered. The resonant frequency of the RLC networks of the active inductor is given by 1 = LCp √ Gm1 Gm2 = ωt1 ωt2 . Go1 and C1 only affect Rs and L. the gyrator-C network behaves as a lossy inductor with its parasitic parallel resistance Rp . The small-signal behavior of a gyrator-C active inductor is fully characterized by its RLC equivalent circuit. This self-resonant frequency is typically the maximum frequency at which the active inductor operates. the phase of the impedance of the synthesized π inductor is . parallel capacitance Cp .

1+.(2.5 with its parameters given by Rp = Cp = 2 . 2 The admittance looking into port 2 of the gyrator-C network is obtained from Y = Iin V2+ − V2− C2 Go2 + + = s 2 2 1 . Q-enhancement techniques that can offset the detrimental effect of Rp and Rs should be employed to boost the quality factor of the active inductors. For applications such as band-pass filters.5.13) Eq.14) Go1 /2 .12) sC2 + Go2 (V2− − V2+ + Gm2 (V1+ − V1− ) = 0. and 2+ yields −Gm1 (V2+ − V2− ) + Iin + sC1 + Go1 (V1− − V1+ ) = 0.1. 2 (2. Rs = Gm1 Gm2 L= C1 /2 . Go2 C2 .12) can be represented by the RLC network shown in Fig. 2. We continue to assume that the transconductances of the transconductors are constant. In these cases. 2 (2.2. Consider the lossy floating gyrator-C network shown in Fig. active inductors with a large quality factor are mandatory.2. Gm1 Gm2 . Go1 C1 s + 2Gm1 Gm2 2Gm1 Gm2 (2.28 CMOS Active Inductors The finite input and output impedances of the transconductors constituting active inductors result in a finite quality factor. Writing KCL at nodes 1-. 2-.4 Lossy Floating Gyrator-C Active Inductors Lossy floating gyrator-C active inductors can be analyzed in a similar way as lossy single-ended gyrator-C active inductors.

signal sensitivity.Characterization of Active Inductors 29 The constant 2 in (2.14) without modification. inductance tunability. and power consumption. Eq.5. we investigate the most important figure-of-merits that provide quantitative measures of the performance of active inductors. C2 and Go2 represent the total capacitances and conductances at nodes 1 and 2.V 2 ) - V2 2+ C1 Go1 1V in C1 G o1 2- I in Rs Rp Cp L G o2 C2 V1 1+ + Gm2 (V 1 . however. noise. These figure-of-merits include frequency range.(2. quality factor.1 Frequency Range It was shown in the preceding section that an lossless gyrator-C active inductor exhibits an inductive characteristic across the entire frequency spectrum. I in 2V in 2+ G o2 C2 + Gm1(V 2 . The capacitance and conductance at the interface nodes 1+ and 1− and those at the internal nodes 2+ and 2− will become C2 /2 and Go2 /2. respectively. 2. stability.2. C1 and Go1 . This frequency range can be obtained by examining the impedance of the RLC equivalent circuit of the lossy active inductor . supply voltage sensitivity. A lossy gyrator-C active inductor.V 1 ) C2 C2 C1 C1 G o2 G o2 G o1 G o1 Figure 2. linearity. and C1 /2 and Go21 /2. only exhibits an inductive characteristic over a specific frequency range.2 Characterization of Active Inductors In this section. 2.9) can be used to derive (2. respectively. parameter sensitivity. Lossy floating gyrator-C active inductors.14) is due to the floating configuration of the active inductor.

and phase-locked loops. the pole resonant frequency of Z is given by Rp + Rs . require the inductance of active inductors be tunable with a large inductance tuning range. and capacitive when ω > ωo .15) When complex conjugate poles are encountered. Rs .2 Inductance Tunability Many applications. Also observe that Z has a zero at frequency ωz = Go1 Rs = . LCp ωp ≈ (2.2. voltage or current controlled oscillators.16) is simplified to 1 = ωo . 2. Rs Rp + Rs 1 s2 + s + + Rp Cp L Rp Cp L (2. to maximize the frequency range. For a given inductance L.9) that the inductance of gyrator-C active inductors can be tuned by either changing the load capacitance or varying the transconductances of the transconductors constituting the active inductors. which is set by the cut-off frequency of the transconductors constituting the active inductor.17) where ωo is the self-resonant frequency of the active inductor. Eq. affects the lower bound of the frequency range over which the gyrator-C network is inductive. both Rs and Cp should be minimized. Also observed is that Rp has no effect on the frequency range of the active inductor. Rp Cp L ωp = Because Rp (2.16) Rs . It is evident that the gyrator-C e network is resistive when ω < ωz . The frequency range in which the gyrator-C network is inductive is lower-bounded by ωz and upper-bounded by ωo . inductive when ωz < ω < ωo . The upper bound of the frequency range is set by the self resonant frequency of the active inductor.6. namely pn-junction varactors . however. L C1 (2. It is seen from (2.18) The Bod´ plots of Z are sketched in Fig.2. Two types of varactors exists. such as filters.(2. Capacitance tuning in standard CMOS technologies is usually done by using varactors.30 s CMOS Active Inductors Z= Rs Cp L L +1 Rs .

vR is the reverse biasing voltage of the junction and φo is built-in potential of the junction. -20 dB/dec.2. The performance of junction varactors is affected by the following factors : Large parasitic series resistance . 31 Z(j w) (Degree) 90 45 deg. n+/p-well varactors are single-ended. -90 deg.the resistance of the n-well. 0 -90 wz wo w Figure 2. the quality factor of the varactor quantified by Q= 1 ωRn−well C (2.2.7 for p+/n-well junctions and Fig. . e and MOS varactors.19) where CJo is the junction capacitance at zero-biasing voltage. The junction capacitance of an abrupt pn-junction is given by CJ = CJo . It is seen that CJ varies with vR in a nonlinear fashion.p+/n-well varactors suffer from a large series resistance .20) ./dec.6. The swing of the voltages at the nodes of the varactors must ensure that the n+/p-well and p+/nwell junctions be revise biased all the time such that a junction capacitance exists. vR 1+ φo (2. on the other hand. As a result. are floating varactors. Because p-substrate is connected to the ground. p+/n-well varactors.8 for n+/p-well junctions. The sideviews of pn-junction varactors are shown in Fig./dec.Characterization of Active Inductors |Z(j w)| (dB) Rs R p R s +Rp wz Resistive Inductive wo Capacitive w 20 dB/dec. Bod´ plots of the impedance of lossy gyrator-C active inductors.

2. As pointed out in [68] that Cmin 2. Sideview of p+/n-well varactors The sideview of accumulation-mode MOS varactors is shown in Fig. A key advantage of accumulation-mode MOS varactors is the large voltage swing across the terminals of the varactors. the larger the n-well and subsequently the larger the n-well/p-substrate junction capacitance.10 for VG > VS .32 CMOS Active Inductors where Rn−well is the parasitic series resistance. the spacing between p+ and n+ diffusions should be minimized.9 for VG < VS and Fig. creating an accumulation layer and C arises to GS Cmax can be made from the gate-oxide capacitance. They are the most widely used varactors in voltage/current-controlled oscillators.7.2.The nonlinear characteristics of C result J in a small capacitance tuning range with a low capacitance tuning ratio. Stringent voltage swing requirement . the electrons in the n-well region underneath the gate will be repelled and a depletion region is created. To minimize this unwanted resistance. Small capacitance tuning range .As pointed out earlier that the p+/nwell and n+/p-well junctions must remain in a reverse biasing condition all the time to ensure the existence of a junction capacitance. is small. If VG < VS . Large parasitic capacitance between n-well and p-substrate .The larger the capacitance of the varactors.5 to 3 when −1V ≤vGS ≤1V . When VG > VS . . This imposes a stringent constraint on the swing of the voltage across the terminals of the varactors 1 D 2 R n-well CJ 1 p+ 2 n+ n-well pn-junction CJ Rn-well p-substrate Figure 2. the electrons from the n+ diffusion regions will be pulled to the region underneath the gate.

VS VG Depletion region n+ n+ n-well p-substrate Figure 2. subsequently a large inductance tuning range. Conductance tuning can be done by varying the dc operating point of the transconductors. The conductance tuning range is set by the constraint that the transconducting transistors of the transconductors must remain in the saturation.12. as shown in Fig. A common drawback of junction varactors and MOS varactors is their small capacitance tuning range.2.Characterization of Active Inductors 33 1 2 1 CJ n+ CJ 2 p+ p-well R p-well p-substrate Figure 2. This approach offers a large conductance tuning range. Sideview of n+/p-well varactors.8. The conductance of either the transconductor .9. Conductance tuning can be used for the coarse tuning of the inductance while capacitance tuning can be used for the fine tuning of the inductance. Sideview of accumulation-mode MOS varactors when VG < VS .

with a positive transconductance or that with a negative transconductance can be tuned. It is seen from (2. The conductance tuning range is set by the pinch-off condition while the capacitance tuning range is set by the range of the control voltage of the varactors. It will.11.10. v GS Capacitance of accumulation-mode MOS varactors. however. The variation of the quality factor due to the tuning of L must therefore be compensated for such that L and Q are tuned in a truly independent fashion.9) that a change in the transconductances of the transconductors of an active inductor will not affect Rp and Cp of gyrator-C active inductors. C GS C max C min VGS Figure 2.34 CMOS Active Inductors VS VG Gate voltage induced electrons n+ n+ n-well p-substrate Figure 2. This is echoed with a change in the quality factor of the active inductors. alter the parasitic series resistance Rs of the active inductor. It should be noted that the fine tuning of the inductance of active inductors from the capacitance tuning does not affect Rs . . Sideview of accumulation-mode MOS varactors when VG > VS .

3 Quality Factor The quality factor Q of an inductor quantifies the ratio of the net magnetic energy stored in the inductor to its ohmic loss in one oscillation cycle. the quality factor of these inductors is independent of the voltage / current of the inductors.21) For a linear inductor. 29] Q = 2π× Net magnetic energy stored . Instantaneous Quality Factor The quality factor Q of an inductor quantifies the ratio of the net magnetic energy stored in the inductor to its ohmic loss in one oscillation cycle [33. This property. the phase noise of the oscillators. (2. in particular. Energy dissipated in one oscillation cycle (2. For spiral inductors. the complex power of the active inductor is obtained from P (jω) = I(jω)V ∗ (jω) = e[Z]|I(jω)|2 + j m[Z]|I(jω)|2 . Inductance tuning of gyrator-C active inductors. however. an alternative definition of the quality factor that accounts for the swing of the voltage / current of the active inductors is needed.12. When active inductors are used in applications such as LC oscillators. the inductance of the active inductors is a strong function of the swing of the voltage and current of the oscillators. Conductance tuning can be carried out by varying either Gm1 or Gm2 while capacitance tuning is done by varying the varactor Cx .2. does not hold for active inductors as the inductance of these inductors depends upon the transconductances of the transconductors constituting the active inductors and the load capacitance. 2.Characterization of Active Inductors 35 V in Gm1V2 2 1 C Cx -Gm2V 1 Figure 2. To quantify the ratio of the net magnetic energy stored in the inductor to its ohmic loss in one oscillation cycle and relate it to the performance of LC oscillators.22) .

denoted by Q1 = ωL . (2.24) Fig. L Q= 1− (2.21) in this case becomes Q= m[Z] .26) accounts for the effect of the finite output impedance of deep sub-micron MOSFETs. whereas the third term. and L = 1.13 shows the frequency dependence of the quality factor of the active inductor with Rs = 4Ω. and |. The second term.2.25) quantifies the quality factor of the active inductor at low frequencies. denoted by Q3 = 1 − 2 Rs Cp − ω 2 LCp .23) Eq. respectively. The quality factor of a lossy gyrator-C active inductor can be derived directly from (2. Eq.27) .(2.22) quantifies the net energy loss arising from the parasitic resistances of the inductor.6 nH [33]. denoted by Q2 = Rp ωL Rp + Rs 1 + Rs 2 .23) provides a convenient way to quantify Q of linear inductors including active inductors.| is the absolute value operator. The first term in (2.23) ωL Rs Rp ωL Rp + Rs 1 + Rs 2 2 R s Cp − ω 2 LCp .15) and (2.(2. V (jω) and I(jω) are the voltage across and the current through the inductor. L (2. Rs (2.24). Cp = 140 fF. whereas the second term measures the magnetic energy stored in the inductor.36 CMOS Active Inductors where e[Z] and m[Z] are the resistance and inductive reactance of the inductor. e[Z] (2. It is seen that the first term in (2. the superscript ∗ is the complex conjugation operator. Active inductors are linear when the swing of the voltages / currents of the inductors are small and all transistors of the active inductors are properly biased. Rp = 1kΩ. respectively.

The transconductance of the transconductor in Fig. shows that the quality factor vanishes when frequency approaches the cut-off frequency of the transconductors of the active inductor.16(b) is also positive.13. Since Go1 is typically the output impedance of the transconductor with a positive transconductance. It is seen that Q1 dominates the quality of the active inductor and is therefore widely used to quantify the quality factor of active inductors. To boost the quality factor of active inductors.16.Because Rs = . Frequency dependence of the quality factor of active inductors. which in turn increases iD . This is because an increase in vin will decrease iD1 . the use of transconductors with a large output impedance is critical. As a result.15. Q2 and Q3 manifest themselves at high frequencies only. iD2 = J1 − iD1 will increase.2.2. Four approaches can be used to reduce Rs : Go1 . consider the transconductors shown in Fig. This is because an increase in vin will lead to an increase in vGS . Rs can be lowered by reducing Gm1 Gm2 Go1 directly. respectively. The sensitivity of the quality factor of the active inductor with respect to Rs and Rp is investigated in Figs. Although both transconductors have a positive transconApproach 1 .2. The transconductance of the transconductor in Fig. Since io = iD2 − J2 . Rs must be minimized. io will increase as well.16(a) is positive. As an example.14 and 2. io will increase as well. Because io = iD − J.Characterization of Active Inductors 10 9 8 7 Quality factor 6 5 4 3 2 1 0 8 10 10 9 37 Q (ω) 1 Q(ω) Q (ω) 3 Q (ω) 2 Ferquency [Hz] 10 10 10 11 Figure 2.2.

2.2. The former. Rs is varied from 1Ω to 10Ω with step 1Ω.Rs can be lowered by increasing Gm1 and Gm2 directly. as shown in Fig.16(b) will have a smaller R . regulated cascode.16(b) is given by ro2 . cascode. increases the static power consumption of the active inductors whereas the latter lowers the self-resonant frequency of the active inductors. Rs can be lowered by either increasing the dc biasing currents or increasing the transistor width. s subsequently a higher Q.Reduce Go1 using advanced circuit techniques. the output impedance 1 approximately whereas of the transconductor in Fig.38 15 CMOS Active Inductors 10 Quality factor Rs=1Ω 5 0 8 10 Rs=10Ω 10 9 10 Ferquency [Hz] 10 10 11 Figure 2. Approach 2 . Table 2. Approach 3 . however. Active inductors constructed using the transconductor in Fig.2. ductance and both have an infinite input impedance. such as cascodes.16(a) is given by gm that of the transconductor in Fig. .17. Cascodes are effective in lowering the output conductance and can be used here to reduce Go1 .14. and multi-regulated cascode transconductors. The effect of Rs on the quality factor of active inductors. Since the transconductances of the transconductors are directly proportional to the dc biasing currents and the width of the transistors of the transconductors.2. Another downside of this approach is that the inductance of the inductors L will also be affected.1 compares the minimum supply voltage and output conductance of basic.

both series and parallel.15. It is well known that the series RL network of the RLC network shown in Fig.18 can be replaced with the parallel RL network shown in the figure with [68] . Simplified schematics of transconductors with a positive transconductor. (a) Common-drain transconductor.Characterization of Active Inductors 10 9 8 7 Quality factor 6 5 4 3 2 1 0 8 10 10 9 39 R =10 kΩ p Rp=1 kΩ 10 Ferquency [Hz] 10 10 11 Figure 2. of active inductors. The effect of Rp on the quality factor of active inductors.c) Differential-pair transconductors. v in io J v in M1 J1 Vb io J2 v in J2 io M2 J1 Vb M2 M1 (a) (b) (c) Figure 2.Use a shunt negative resistor at the output of the positive transconductor to cancel out the parasitic resistances. Rp is varied from 1kΩ to 10kΩ with step 1kΩ. Approach 4 . (b.16.2.

This is because in order to have the two network to Rs be equivalent. Rp + jωLp (2. (d) Multi-regulated cascode transconductor. Rs + jωLs = jωRp Lp .17. Simplified schematics of cascode transconductors. (b) Cascode transconductor. in other words. (c) Regulated cascode transconductor. to exhibit the same terminal impedance. Zs (jω) = Zp (jω) is required.40 CMOS Active Inductors io io vin M1 Vb v in M2 M1 M3 vo vin -A M1 M2 io vin (a) (b) (c) vo Vb M4 M3 v in (d) -A M1 M2 io vin Figure 2. (a) Basic transconductor.29) Matching the real and imaginary parts yield .28) ωLs . Q where Q = (2. Rp = Rs (1 + Q2 ) 1 Lp = Ls 1 + 2 .

28) follows from Eq.1.28) is valid at all frequencies. The L ∼ Rs branch of the RLC circuit of the active inductor is replaced ˆ ˆ with the parallel L ∼ Rp network with 1 ˆ L=L 1+ 2 . ductors.30) Eq. Note that (2. Rs Lp + Rp Ls = Rp Lp . ro1 (ro2 gm2 )(ro3 gm3 )(ro4 gm4 ) Rs Rp − ω 2 Ls Lp = 0.18.19.30). Q ˆ Rp = Rs (1 + Q2 ). Now consider the RLC network of an active inductor shown in Fig.(2. Transformation of a RL parallel branch to a RL series branch. The minimum supply voltage and output conductance of basic and cascode transcon- Transconductor Basic Cascode Regulated cascode Multi-regulated cascode Min. Rs C Zs Ls Zp Rp Lp C Figure 2.Characterization of Active Inductors 41 Table 2.2. VDD 2Vsat 3Vsat 2VT + Vsat 2VT + Vsat Output impedance Go = Go = Go = Go = 1 ro 1 ro1 (ro2 gm2 ) 1 ro1 (ro2 gm2 )(ro3 gm3 ) 1 . (2. (2.(2.31) .

If Rs is negligible. The quality factor of the compensated active inductor at ωo is given by Cp + Ccomp . Q≈ (2. the quality factor is mainly determined by Rs ωL ωz 1− Rs ωo 2 Q= − ω ωo 2 .35) As shown in Fig.34) ≈0 and ω ωo 2 At frequencies above ωz and below ωo . From (2. Case 2 .19.If Rp is large. a negative resistor of resistance Rcomp = −Rtotal can be connected in parallel with Cp to eliminate the effect of both Rp and Rs of the active inductor simultaneously.34) is simplified to ωL . Eq. (2. the total parasitic parallel resistance of the active inductor becomes ˆ Rtotal = Rp //Rp .42 Consider two cases : – CMOS Active Inductors Case 1 .32) ≈0 and Rp ωLs (2.36) In this case. ˆ L ˆ Q(ωo ) = (Rp ||Rp ||Rcomp ) (2.24) with Rs = 0. Rs ωz ωo ≈0. (2.33) Q≈ – follows. 2 (2. we arrive at ω Rp 1− ωL ωo ω ωo 2 2 Q= At frequencies below ωo .(2.2.37) . Note that the resistance of the negative resistor should be made tunable such that a total cancellation can be achieved. . the quality factor of the active inductor is mainly determined by Rp .

Rs Rp Cp L Rp Cp Rp L C comp R comp Figure 2. and Q(ωo . i)di. It should also be noted that although the negative resistor compensation technique is widely used to improve the quality factor of spiral inductors. Average Quality Factors Active inductors are RLC tanks when Rs . This is because an active negative resistor is used to cancel out the largely skin-effect induced parasitic series resistance of spirals. such as LC tank oscillators.39) where Imin and Imax are the minimum and maximum currents of the transconductors of active inductors. It should be noted that because Rs and Rp are frequency-dependent. and Cp are accounted for. Rp .Characterization of Active Inductors 43 where Ccomp is the input capacitance of the compensating negative resistor.38) where φ(ω) is the phase of the tank impedance. i) is the instantaneous quality factor at frequency ω and channel current i provides an effective mean to quantify the quality factor of active inductors. the inductance of the active inductors varies with the current / voltage of the inductors. . Imin (2. The effective quality factor defined as ω 2(Imax − Imin ) Imax Q(ω) = Q(ω. Q enhancement using a shunt negative resistor. especially when active inductors are employed in circuits that are operated in a large-signal mode. a total compensation in this case is difficult to achieve. The quality factor of LC tanks is obtained from [69] Q(ω) = ωo ∂φ(ω) . Rcomp should be designed in such a way that a total resistance cancellation is achieved across the frequency range of the active inductor.19. The quality factor of a passive LC tank at a given frequency is independent of the current of the tank. Unlike passive LC tanks. 2 ∂ω (2.

43) .2. nD (2.2.20 shows the partial schematics of basic transconductors widely used in the construction of gyrator-C active inductors. such as the thermal noise of the bulk resistance of the source and drain diffusions. (2. γ = 2. the power of the input-referred noise-voltage and that of the noise-current generators of the transconductors constituting the active inductor must be derived first.22. as shown in Fig. consider the common2 gate transconductor. To derive the input-referred noise-voltage generator v n of the transconductor. To analyze the noise of a gyrator-C active inductor. is neglected.41) and (2. we first short-circuit the input of the transconductor.42) yields 2 vn = 2 ro 1 i2 ≈ i2 . Rg is the gate series resistance. and that of the input-referred noise2 . 2 nD g 2 nD (1 + gm ro ) m (2. which has a typical corner frequency of a few MHz [71]. The power of the input-referred 2 noise-voltage generator.40) represents the sum of the power of the thermal noise generated in the channel of MOSFETs and the thermal noise of the gate series resistance of MOSFETs. The effect of the flicker noise of MOSFETs. The output noise power of the transconductor due to inD is calculated 2 2 vno = ro i2 .2.41) where ro is the output resistance of the transistor.2 are derived. The output noise power of the transconductor is obtained 2 2 vno = (1 + gm ro )2 vn .2. The thermal noise of other parasitics of MOSFETs.2 where i2 = 4kT (γ + Rg gm )gm Δf nD (2. of these transconductors can be derived using current generator. T is the temperature in degrees Kelvin.21. denoted by vn . and k is Boltzmann constant.44 CMOS Active Inductors 2. is also neglected. denoted by in conventional noise analysis approaches for 2-port networks [70]. Fig.42) Equating (2. as shown in Fig.4 Noise Active inductors exhibit a high level of noise as compared with their spiral counterparts.5 for deep sub-micron devices. To illustrate how the results of Table 2. We then remove i2 and nD apply vn at the input of the transconductor. and the results are given in Table 2.

Input-referred noise-voltage and noise-current generators of transconductors at low frequencies. .Characterization of Active Inductors Transconductors with noise sources 45 Transconductors with noise generators Common source Cascode Common gate Source follower Figure 2.20.

Transconductor 2 vn i2 n Common source 2 vn = i2 nD 2 gm i2 i2 nD1 nD2 + 2 gm1 (gm1 ro1 gm2 )2 i2 nD 2 gm i2 nD 2 gm i2 = 0 n i2 = 0 n Cascode 2 vn = Common gate 2 vn = i2 = 0 n i2 = 0 n Source follower 2 vn = g m vgs Vb 2 inD vno ro inD v gs v in Figure 2.46 CMOS Active Inductors Table 2.23 where the input port of the transconductor is open-circuited and the output noise power of the circuit is calculated.2. Derivation of input-referred noise-voltage generator of a common-gate transconductor at low frequencies. Power of input-referred noise-voltage and noise-current generators of basic transconductors at low frequencies.21. we assume that there exists a resistor of resistance .2. Note that we have utilized ro gm 1 in (2. consider Fig.43) to simplify the results. To avoid the difficulty caused by the floating node 1. To derive the noise-current generator of the common-gate transconductor.

23. Note that this approach is used in most IC CAD systems to avoid floating nodes.45) . Solving (2. Rx between node 1 and the ground. g m vgs Vb 2 inD 2 ro vno inD v gs 1 Rx v in Figure 2. Derivation of the input-referred noise-voltage generator of a common-gate transconductor at low frequencies. 2 go nD (2. Derivation of the input-referred noise-current generator of a common-gate transconductor at low frequencies. Writing KCL at nodes 1 and 2 yields (gx + gm + go )v1 − go v2 = inD (node 1).Characterization of Active Inductors 47 g mv gs Vb vn v gs vn vno ro Figure 2. (2.22.44) −(go + gm )v1 + go v2 + inD = 0 (node 2).44) yields 2 v2 = 1 2 i .

46) (node 2).48) Taking the limit Rx →∞ or equivalently gx →0 in (2.47) yields gx go 2 i2 = n i2 nD gm 1+ go 2. −(go + gm )v1 + go v2 = 0 Solving (2.2. n (2. . (2. remove all the noise sources of the circuit. Derivation of the input-referred noise-current generator of a common-gate transconductor at low frequencies.24. and compute the output noise power. we arrive at i2 = 0.24.47) Equating (2. the power of the inputn referred noise-voltage and noise-current generators of active inductors can be derived.48 CMOS Active Inductors We then apply the noise-current generator in at the input of the circuit. Writing KCL at nodes 1 and 2 yields (gx + gm + go )v1 − go v2 + in = 0 (node 1).49) g mvgs 2 v gs Vb in Rx 1 in ro vno Figure 2.48). 2 Once vn and i2 of the transconductors are available. 2 gx n (2.45) and (2. 2 v2 1 2 i .46) yields gm = 1+ go 2 (2. as shown in Fig.

In = Y2 Vn1 + Gm2 Vn2 .25(a) and Fig.2.51) To ensure that Fig. For lossy gyrator-C active inductors.2. and Y1 and Y2 are the admittance at ports 1 and 2.50) and that of (2. respectively. To achieve this. respectively. Y1 Y2 + Gm1 Gm2 (2. Y2 = Go2 + sC2 . we impose Vn = Vn1 .2.56) . we have Y1 Vn + In Y1 Y2 + Gm1 Gm2 2 V12 = .2.52) (2. Because Zin (s) = we arrive at Vn = Vn1 + Zin In .25(a) where vn1 and vn2 denote the power of the noise-voltage generators of the transconductors 1 and 2.55) (2. (2. In = (Go2 + sC2 )Vn1 + Gm2 Vn2 .25(a).2.25(b) are equivalent. (2.Characterization of Active Inductors 49 2 2 Consider the active inductor of Fig. it is trivial to show that 2 V12 = Vn1 + Y1 Gm2 Vn2 + Y2 Vn1 Y1 Y2 + Gm1 Gm2 .54) Y1 .52) becomes Vn = Vn1 . For the network of Fig.53) (2.51) must be the same.25(b).50) For the network of Fig.(2. the right hand-side of (2. we have Y1 = Go1 + sC1 . Eq. (2.

When the voltage swing is large. This assumption is only valid if the swing of the input voltage of the transconductors is small. the inductive characteristics at port 2 of the gyrator-C network remain. m2 2 G m1V+ vn 1 2 Y1 Y2 in G m1V+ v n1 Y2 1 Y1 v n2 G m2V - Gm2V- (a) Figure 2. as illustrated graphically in Fig. however. increases from L = Gm1 Gm2 Gds1 Gds2 Gm1. The linearity constraint of active inductors sets the maximum swing of the voltage of the active inductors.5 Linearity The preceding development of gyrator-C active inductors assumes that the transconductors of the active inductors are linear. we arrive at 2 2 Vn = Vn1 . It should be emphasized that although the transconductances of the transconductors of gyrator-C networks drop when the operating point of the transistors of the transconductors moves from the saturation region to the triode region. 2 In CMOS Active Inductors = Go2 + jωC2 2 (2. 2.26. If we assume that the transconductances of the transistors of gyrator-C active inductors are constant when the transistors are biased in the saturation. the transconductors will exhibit a nonlinear characteristic and the synthesized active inductors are no longer linear.25. The inductance of the gyrator-C active C C to L = . (b) Noise of single-ended gyrator-C active inductors.m2 are the transconductances of transconductors 1 and 2 respectively when .2.2. then the maximum swing of the voltage of the active inductors can be estimated from the pinch-off condition of the transistors. where inductors. the transconductances of the transistors decrease from gm (saturation) to gds (triode) in a nonlinear fashion.57) 2 Vn1 + 2 G2 Vn2 . When the transistors of active inductors enter the triode region.50 By assuming Vn1 and Vn2 are uncorrelated.

we investigate the stability of gyrator-C active inductors.2 = Go to simply the results. 2C1 C2 (C1 + C2 )2 (2. gm > gds follows. 2.4 is given by Z= sC1 + Go1 . The stability of active inductors is critical to the overall stability of systems employing active inductors. The poles of the system C1 + C2 4C1 C2 Gm1 Gm2 − 1± 1 − . In this section. + s(C1 + C2 ) + Gm1 Gm2 (2.26.2. i DS Pinch-off Di DS1 g m= Di DS1 DV GS VGS2 VGS1 Di DS2 g ds = V DS2 Di DS2 DV GS V DS1 vDS Figure 2. Note that we have assumed the load capacitance C remains unchanged.59) . Because ΔiDS1 > ΔiDS2 .Characterization of Active Inductors 51 in the saturation and Gds1.6 Stability Gyrator-C active inductors are negative feedback systems.58) s2 C1 C2 where we have utilized Gm are given by p1. Transconductance of MOSFETs in the saturation and triode regions. The inductance thus varies with the swing of the voltage of the active inductors in a nonlinear fashion. The impedance looking into port 2 of the gyrator-C active inductor shown in Fig. It should also be noted that the parasitic resistances of active inductors also vary with the voltage swing of the active inductors.ds2 are the transconductances of transconductors 1 and 2 respectively when in the triode.2.

When the parasitics of MOSFETs are accounted for.2 = 1 − 1± 1 − G2 .2. This is echoed with an increase in the level of oscillation in the response of the C2 C1 and have active inductor. (2.58) with the standard form of the characteristic equation of second-order systems 2 s2 + 2ωo ξs + ωo = 0. It should be noted that the preceding analysis is based on the assumption that active inductors are 2nd-order systems. The degree of stability can be assessed by evaluating its damping factor.52 CMOS Active Inductors The poles of the gyrator-C active inductor are located in the left half of the s-plane and the gyrator-C active inductor is a stable system.60) where ξ denotes the damping factor and ωo is the pole resonant frequency. The result is given by 1 ξ= √ 2 Gm1 Gm2 C2 + C1 C1 . If C1 = C2 = C and Gm1 = Gm2 = Gm . This is echoed with a reduced 1 level of damping.7 Supply Voltage Sensitivity The supply voltage sensitivity of the inductance of active inductors is a figure-of-merit quantifying the effect of the variation of the supply voltage on the inductance of the active inductors. The fluctuation of the supply voltage of a mixed analog-digital system is mainly due to the switching noise of the system . ξ= Gm (2.61) Eq.61) reveals that an increase in Gm1 and Gm2 will lead to a decrease in ξ.61) is that the ratios C2 C1 a marginal impact on the damping factor simply because these two quantities vary in the opposite directions when C1 and C2 change.62) An increase of Gm will lead to a decrease of ξ. the absolute stability margin is C set by the capacitance C and is independent of Gm . C2 (2. active inductors are no longer 2nd-order systems and their stability will deteriorate. which is obtained by comparing the denominator of (2. 2. we have p1. m C 1 . Also observed from (2. and the values of C1 and C2 are often close. Because e[p1.2 ] = − .(2.

Eq. subsequently the transconductances of the transconductors constituting the active inductor. where E[.] denotes the mathematical mean operator.Characterization of Active Inductors 53 [14]. Assume that the supply voltage of a mixed-mode system containing an active inductor varies from VDD and VDD + ΔVDD . the resistance of poly resistors in a typical 0. we arrive at vary with VDD and because L = Gm1 Gm2 1 ∂L 1 ∂L ∂L = −L + . For a well designed mixed-mode system. For example.64) The normalized supply voltage sensitivity of the active inductor is given by Gm1 Gm2 L SVDD = − SVDD + SVDD . Analysis of the effect of parameter spread is vital to ensure that the performance of circuits meets design specifications once the circuits are fabricated. The small-signal analysis approach can therefore be employed to analyze the effect of ΔVDD on the inductance of the active inductor.(2. (2. ∂VDD Gm1 Gm1 Gm2 Gm2 (2.63) The fluctuation of the supply voltage VDD affects the inductance of the active inductor mainly by altering the dc operating point. ΔVDD VDD holds.65) reveals that both SVDD and SVDD contribute L equally to SVDD .2. By assuming that the load capacitance C of the gyrator-C active inductor does not C . Following the definition of normalized sensitivity given in [72].8 Parameter Sensitivity The minimum feature size of MOS devices in modern CMOS technologies has been scaled down more aggressively as compared with the improvement in process tolerance such that the effect of process variation on the characteristics of circuits becomes increasingly critical.18μm CMOS process has an error of ±20% approximately and that of n-well resistors has an error of ±30% approximately. 2. respectively. where ΔVDD is a random variable with E[ΔVDD ] = 0. To minimize the supply voltage sensitivity of active inductors. L ∂VDD (2. Active inductors consist of a number of active devices and their performance is greatly affected .65) Gm1 Gm2 where SVDD and SVDD are the normalized supply voltage sensitivity of Gm1 Gm1 Gm2 and Gm2 . transconductors with a constant Gm should be used. the normalized sensitivity of the inductance of an active inductor to the supply voltage is defined as L SVDD = VDD ∂L .

and N is the number of the parameters of the active inductor. The accuracy of Monte Carlo analysis increases with an increase in the number of simulation runs and is therefore . L2 C2 G2 G2 m1 m2 (2.68) (2. because 1 ∂L = .67) where σL and σxj denote the standard deviations of L and xj . Gm2 Gm1 G2 m2 we obtain the normalized spread of the inductance of the active inductor 2 σ2 σ2 σ2 σL = C + Gm1 + Gm2 . namely worst-case analysis. and Monte Carlo analysis.66) quantifies the effect of the variation of the parameter xj on the inductance of the active inductor. respectively. Gm1 Gm1 Gm2 C ∂L =− . By assuming that the parameters of the active inductor are Gaussian distributed and uncorrelated. the overall effect of the variation of the parameters of the active inductor on the inductance of the inductor is obtained from N 2 σL = j=1 ∂L ∂xj 2 2 σx j .54 CMOS Active Inductors by the parameter spread of these components. For a gyrator-C active inductor. also known as corner analysis. (2. C Gm1 Gm2 C ∂L =− 2 .69) There are two ways in which circuit designers can analyze the effect of parameter spread on the inductance of active inductors. The former determines the inductance of active inductors at process corners while the latter quantifies the degree of the spread of the inductance of active inductors around the nominal inductance of the inductors. The normalized sensitivity of the inductance of an active inductor to a parameter xj of the inductor defined as L Sxj = xj ∂L L ∂xj (2.

mainly due to the dc biasing currents of their transconductors. the inductance of gyrator-C active inductors varies with the voltage and current of the transconductors constituting the active inductors. Gyrator-C active inductors. Despite of this. G and Gm2 m1 are made small. their power consumption must also be included. To have a large inductance. corner analysis is the most widely used method to quantify the effect of process spread.10 Power Consumption Spiral inductors do not consume static power. Also. consume dc power. and differential pair configurations as positive .3 Implementation of Single-Ended Active Inductors The need for a high self-resonant frequency of active inductors requires that the transconductors of these active inductors be configured as simple as possible. 2. when negative resistors are employed for boosting the quality factor of active inductors. source follower. This is typically achieved by lowering the dc biasing currents of the transconductors. common-gate. Corner analysis. the inductance. This also lowers their level of power consumption and reduces the silicon area required to fabricate the inductors.9 Signal Sensitivity Unlike spiral inductors whose inductance is independent of the voltage and current of the inductors. is time-efficient but the results obtained from corner analysis are typically over conservative. When replica-biasing is used to minimize the effect of supply voltage fluctuation on the inductance of active inductors. on the other hand. parasitic resistances. and quality factor of the active inductor all vary with the signal swing. such as active inductor LC oscillators. When an active inductor is used in applications where the voltage of the active inductor experiences a large degree of variation. Often the power consumption of an active inductor is set by that of its replica-biasing and negative resistor networks. Most reported gyrator-C active inductors employ a common-source configuration as negative transconductors.2.2. This is because the transconductances Gm1 and Gm2 of the transconductors are signal dependent when signal swing is large. however. 2. as to be seen shortly. the power consumed by the replica-biasing network must be accounted for. The power consumption of gyrator-C active inductors themselves is usually not of a critical concern because the inductance of these inductors is inversely proportional to the transconductances of the transconductors constituting the inductors.Implementation of Single-Ended Active Inductors 55 extremely time consuming. 2. As a result. the transconductances of the transconductors of the active inductor vary with the signal swing.

27 show the schematic of two basic gyrator-C active inductors.56 CMOS Active Inductors transconductors. These basic transconductors have the simplest configurations subsequently the highest cutoff frequencies and the lowest silicon consumption. 2. making it attractive for highfrequency applications. which are available only in mixed-mode CMOS processes. we obtain the parameters of the equivalent RLC network of the active inductor . All transistors are biased in the saturation.3. (ii) nMOS and pMOS transistors have the same pinch-off voltage Vsat . are the output conductance and transconductance of transistor j. 2.2.2. respectively. Go2 ≈gm1 . In Fig. For Fig. the following assumptions are made in analysis of active inductors and in determination of both their signal swing and the minimum supply voltage : (i) nMOS and pMOS transistors have the same threshold voltage VT . the transconductor with a positive transconductance is commongate configured while the transconductor with a negative transconductance is common-source configured. A notable advantage of the active inductor in Fig. j = 1.27(b) is that all transistors are nMOS. In Fig. where goj and gmj . the transconductor with a positive transconductance is common-drain configured while the transconductor with a negative transconductance is common-source configured.2.27(a). Gm1 = gm1 .1 Basic Gyrator-C Active Inductors Fig. To simplify analysis. C2 = Cgs1 . MOS varactors are often added in parallel with Cgs to tune the inductance of active inductors. Using (2. and Gm2 = gm2 . The load capacitor of the transconductors is realized using the intrinsic capacitance Cgs of the transistors of the transconductors directly to maximize the upper bound of the frequency range of the active inductors and to avoid the use of expensive floating capacitors.9). Cgd and parasitic diffusion capacitances are neglected unless otherwise noted explicitly.27(b).2. (iii) Only Cgs is considered. (iv) The minimum voltage drop across biasing current sources and current-source loads is Vsat . Go1 ≈go1 .2. we have C1 = Cgs2 . This section presents a comprehensive treatment of both the circuit implementation and characteristics of CMOS active inductors.27(a).

(b) Simplified schematic of basic gyrator-C active inductors. Also.71) Q≈ The self-resonant frequency of the active inductor is given by ωo ≈ where √ 1 = ωt1 ωt2 . It is observed from (2. limiting the quality factor of the active inductor.72) .70) Cgs2 . In this case. gm1 gm2 J2 1 Vb J1 M1 2 M2 v in v in J1 M2 2 J2 1 M1 (a) Figure 2. LCp (2. Rp = 1 gm1 . the quality factor of the active inductor is obtained from ωt2 Rp = . L= gm1 gm2 Rs = go1 . further lowering the quality factor.Implementation of Single-Ended Active Inductors 57 Cp = Cgs1 .70) that the parasitic parallel resistance R is rather p small. In evaluating the quality factor of this active inductor. the effect of the parasitic series resistance Rs is often neglected as Rp is small. the parasitic series resistance is large.27. ωL ω (2. (2.

go1 . ωz should be minimized. ωt1 (2. The 1 on the quality factor of the active inductor detrimental effect of Rp = gm1 can not be neglected. the quality factor becomes Q(ωo ) = ωt2 .58 ωtj = gmj .74) The frequency of the zero of the active inductor. i = 1. (b) Circuit implementation. is given by (2. .73) is the cutoff frequency of transconductor j. Cgs2 (2. however. The former is usually preferred as the latter lowers ωo . This can be achieved by reducing go1 or increasing Cgs2 .(2.75) J2 vin Io C in Vin vin M1 Vb M2 J1 gm (a) (b) Figure 2. Because the output impedance of deep sub-micron MOSFETs is small. Single-ended negative impedance networks. can be eliminated by ˆ connecting a negative resistor of resistance Rp = −Rp in parallel with Rp . The effect of Rp . 2 Cgsj CMOS Active Inductors (2. which is the lower bound of the frequency range of the active inductor.75) reveal that : In order to maximize the frequency range of the active inductor. At the self-resonance frequency of √ the active inductor ωo = ωt1 ωt2 .28. (a) Block diagram.18) ωz = Eqs.72) and (2.

28 for single-ended negative resistors and Fig. It is well known that cascodes and regulated cascodes are effective means to boost the output impedance of transconductors. Differential negative impedance networks.2. Since M2 is a common-gate configuration. The reason that the preceding basic active inductors have a low R is p because the input impedance of the positive transistor in Fig. Note that the removal of the biasing tail current source will also remove the tunability of the resistance of the negative resistor. vin J I+ o C in V in M1 vin Io J vin M2 M1 M2 gm (a) (b) Figure 2.2.2.28 is depicted as the followings : An increase in the gate voltage of M1 will increase the voltage at the source of M1 .29 for differential negative resistors. provided that biasing currents are provided by the circuit connected to the negative resistor. It can be shown that the impedance at low frequencies is given by 1 gm1 1 gm2 Z≈ − + .29 in a similar manner. is the reduced signal swing. 65.2. The price paid.29. transconductors synthesizing negative resistors should be configured as simple as possible. an increase in the source voltage of M2 will increase the drain voltage of M2.29. The tail current source in the differential configuration can be removed.76) To maximize the frequency range over which a constant negative resistance exists. The positive feedback of the single-ended negative resistor in Fig.27(a) – . 38. however. 73]. For Fig. as shown in Fig.2. The effect of Rs can be compensated for in three different ways : – Use cascodes and regulated cascodes to reduce go1 [37. (a) Block diagram. (2. A positive feedback is thus established.2. Readers can verify the positive feedback of the differential negative resistors of Fig. (b) Circuit implementation.Implementation of Single-Ended Active Inductors 59 Negative resistors can be realized using transconductors with positive feedback.

input voltage Min.27(b) is 1/gm2 .2 Wu Current-Reuse Active Inductors Fig. As an example. and Gm2 = gm2 . Gm1 = gm1 .3 compares the range of the voltage swing and the minimum supply voltage of the two basic active inductors.27(b) VDD − VT − Vsat VT 2VT + Vsat 2. In the nMOS version of the active inductor.c) has an infinite input impedance and a large output impedance ro2 [74. C2 = Cgs1 .3. The Go1 ≈go1 + go2 . 1 .2. The use of transconductors that have both a large output impedance and a large input impedance will eliminate this drawback. Go2 = gm1 parameters of the equivalent RLC network of this active inductor are given by . 36].30 show the schematic of Wu current-reuse active inductors proposed in [42.2. Comparison of the input voltage swing and the minimum supply voltage of the basic active inductors in Fig.2. 75.2. 49.3. The quality factor of the active inductor can be made infinite theoretically by employing a shunt negative resistor whose resistance is −Rp. VDD Fig. input voltage Min. the differentially-configured positive transconductor of Karsilayan-Schaumann active inductors shown in Fig.2. When only Cgs is considered. Table 2.27(a) VDD − VT − Vsat Vsat VT + 2Vsat Fig. The total parasitic parallel ˆ resistance of the active inductor becomes Rp.2. C1 = Cgs2 .2. – Because the series RL branch of the RLC equivalent network of the ˆˆ active inductor can be replaced with a parallel RL branch. It is seen that the active inductor in Fig. 45.total . Table 2. the positive transconductor is common-gate configured while the negative transconductor is common-source configured.16(b.60 CMOS Active Inductors is 1/gm1 and the output impedance of the positive transconductor in Fig.27(a) offers a large input voltage swing and requires a lower minimum supply voltage. 48. Active inductor Max.total = Rp ||Rp .27. The induc1 ˆˆ ˆ tance of the parallel RL branch is given by L = L 1 + 2 while the Q ˆ resistance is given by Rp = (1 + Q2 )Rs . 61].

L= gm1 gm2 Rs = go1 + go2 . lowers Q(ωo ).79) M2 v in J2 M1 Vb v in J1 J2 M1 M2 J1 Vb Figure 2. ωt1 (2. ωL ω √ At the self-resonant frequency ωo = ωt1 ωt2 . In practical design. the dc biasing current of M1 is kept unchanged while that of M2 is increased by injecting an additional current J2 into M2 . Simplified schematic of Wu current reuse active inductors. both ωt1 and ωt2 need to be increased. Rp = 1 gm1 . J is 2 . Increasing ωt1 should therefore be avoided. ωt2 Rp = .78) (2. (2. It is seen from the preceding analysis that to increase ωo .30. Increasing ωt1 . The additional current source J2 is used to boost the transconductance of M2 such that the upper frequency bound can be increased without lowering the quality factor. however. gm1 gm2 The quality factor of Wu active inductors can be estimated by neglecting the effect of Rs and only focusing on Rp as Rp is small. To boost ωt2 without increasing ωt1 .Implementation of Single-Ended Active Inductors 61 Cp = Cgs1 .77) Cgs2 . we have Q≈ Q(ωo )≈ ωt2 .

79–82]. 40.32. Figure 2. the CMOS version of Hara active inductors only emerged a few years ago [41. M1−3 can easily be biased in the saturation to ensure a stable operation of the active inductors. They can be analyzed in a similar way as Wu current re-use active inductors. The minimum supply voltage of the active inductor is also VT + 2Vsat . 78]. 2.32 [76]. . 2. Simplified schematic of the variation of Lin-Payne active inductor.2.3.33[76]. Since the gate voltage is kept at VDD . This in turn lowers the current flowing out of the active inductor.31 requires the minimum supply voltage of only VT + 2Vsat . Simplified schematic of LinPayne active inductor. vGS is reduced. They are indeed gyrator-C active inductors.34 employ only a MOSFET and a resistor.62 CMOS Active Inductors provided by the stage preceding to the inductors and the active inductors are known as Wu current-reuse active inductors.3. The analysis of Ngow-Thanachayanont active inductors is similar to that of Wu current re-use active inductors and is left as an exercise for readers.2.5 Hara Active Inductors Although the MESFET implementation of Hara active inductors appeared two decades ago [77.3 Lin-Payne Active Inductors The Lin-Payne active inductor proposed in [39] and shown in Fig. Another implementation of Lin-Payne active inductors is shown in Fig. 2.3. The feedback operation of nMOS Hara active inductor is as the followings : An increase of the input current will result in an increase in the voltage at the input node. The addition of M3 branch relaxes the biasing difficulties encountered in Lin-Payne active inductors. M2 v in J2 M1 J1 J2 M1 M2 J1 v in Figure 2.4 Ngow-Thanachayanont Active Inductors Ngow and Thanachayanont proposed the low-voltage active inductor shown in Fig.31.2. Hara active inductors shown in Fig.2.

81) 1 .34. Hara active inductors. The input impedance of the nMOS Hara active inductor can be derived from its small-signal equivalent circuit shown in Fig.35 1 RCgs Cgd sRCgd + 1 .33.2. R Vc vin Rs Vc vin vin Rs L R vin L (a) nMOS (b) pMOS Figure 2.Implementation of Single-Ended Active Inductors 63 J2 M3 M1 J3 M2 J1 vin Figure 2. The self-resonant frequency ωo and the frequency of the zero ωz of the active inductor are given by ωo = √ gm = ωt ωz .80) where gm go and Cgs Cgd were utilized to simplify the results. RCgs Cgd (2. ωz = RCgd . Simplified schematic of low-voltage gyrator-C active inductor. Resistor R can be made variable by implementing it using MOSFETs biased in the triode. gm gm s +s + Cgs RCgs Cgd 2 Z≈ (2.

64 CMOS Active Inductors Cgd gm vgs R Cgs v gs vin Figure 2. gm (2.(2. Csb .6.35. When Cgd . Csb . Eqs.84) can be written as 1 ω + g ωt Rs = m ω 1+ ωt and 2 2 2 gm + ω 2 Cgs R .2. and high-order effects are neglected. e The network exhibits an inductive characteristic in the frequency range ω < z ω < ωo . Cgs (2. 2 2 gm + ω 2 Cgs (2. the inductance L and parasitic series resistance Rs of nMOS Hara active inductors are given by Rs = and L= Observe that gm R > 1 is required to ensure L > 0.83) Cgs (gm R − 1) . Under the condition gm R (2. 2 2 gm + ω 2 Cgs (2.84) (2.83) and R ≈ 1 .82) The Bod´ plots of Hara active inductors are the same as those given in Fig. go where ωt = gm . Small-signal equivalent circuit of nMOS Hara active inductors.86) .85) 1.

It is observed that an increase of R will lower both ωz and ωo . Wu folded active inductors were initially proposed by Thanachayanont in [84]. go and other parasitic capacitances of the transistor. L is nearly independent of gm . It is evident from (2. consider the nMOS version of Wu folded active inductors with its small-signal equivalent circuit shown in Fig. sCgs + gm Z= (2.88) where μn is the surface mobility of free electrons and Lc is the channel length.2. Hara active inductors suffer from the loss of the voltage headroom of at least VT .89) . Cgs 2 L2 c ωt = (2.87) L= R ω ωt 1 + ωt 2 where we have utilized ω 2 ≈0. This agrees with the theoretical results. The series resistance Rs is largely dominated by gm at frequencies below the cut-off frequency of the transistor. respectively. To simplify analysis. 2.2.2. A drawback of this approach is the complexity of the voltage doubler and the need for a control clock. To derive the parameters of the RLC equivalent circuit of Wu folded active inductors. we neglect Cgd .3.37.36 and Fig. a voltage doubler was used to increase the supply voltage for R.87) that the inductance ωt L is directly proportional to R and can be tuned by varying R. is independent of the width of the transistor.18μm 1. Also.38[83]. The dependence of the input impedance of the nMOS Hara active inductor on the resistance of the resistor R and the width of the transistor is shown in Fig. The active inductor was implemented in TSMC-0.6 Wu Folded Active Inductors The drawback of Hara active inductors can be eliminated by employing Wu folded active inductors shown in Fig. ωt (2.8V CMOS technology and analyzed using SpectreRF with BSIM3V3 device models.Implementation of Single-Ended Active Inductors 65 ≈ R . In [41].2. It can be shown that the input impedance is given by sRCgs + 1 . Increasing the width of the transistor will lower ωo because gm 3 μn ≈ (VGS − VT ).39.

W is varied from 5μm to 25μm with step 5μm.66 500 400 300 Impedance [Ω] 200 100 0 −100 −200 25μm 5μm Re(Zin) CMOS Active Inductors Im(Z ) in 25μm 5μm −300 −1 10 10 0 Frequency [GHz] 10 1 10 2 Figure 2. 1 It becomes evident that Z has a zero at frequency ωz = and a pole at RCgs gm . The network is resistive at low frequencies ω < ωz with frequency ωp = Cgs 1 and inductive when ωz < ω < ωp .90) can be represented by a series RL network in parallel with a resistor Rp with .5mA. Dependence of Z of nMOS Hara active inductors on the width of the transistor. R = 1kΩ. To derive its RLC equivalent circuit. 1 R (2. Note the behavior of the resistance R≈ gm network beyond ωp can not be quantified by (2.90) Eq. DC biasing current 0.89) due to the omission of Cgd .36. we examine the input admittance of the network Yin = = sCgs + gm sRCgs + 1 1 1 + RCgs 1 R s 1 + gm − R gm − .(2.

5 kΩ with step 0. Rs = . Dependence of Z of nMOS Hara active inductors on R. W = 10μm. R is varied from 0.37.5 kΩ Im(Z ) in 2.Implementation of Single-Ended Active Inductors 500 400 300 Impedance [Ω] 200 100 0 −100 −200 −300 −1 10 0. (b) pMOS Simplified schematic of Wu folded active inductors.38. gm − R 1 gm − 1 R (2. Rp = R.5 kΩ to 2.5 kΩ Frequency [GHz] 10 1 10 2 Figure 2.5 kΩ Re(Zin) 0. DC biasing current 0.5mA. vin R Rs vin Vc Rs Rp Rp L L R vin Vc vin (a) nMOS Figure 2.91) . L= RCgs 1 .5 kΩ 10 0 67 2.5 kΩ.

wz wp w Small-signal equivalent circuit of the nMOS version of Wu folded active inductors.2. we have Rs ≈ and L≈ = . 49. 75. 1 is required in order to have L. The admittance of the active inductor is given by sCgs2 + gm2 s(Cgs1 + Cgs2 ) + (gm1 + gm2 ) Y = sCgs1 . It was also developed by Yodprasit and Ngarmnil in [85].3. The active inductor consists of a differentially configured transconductor with a positive transconductance and a common-source transconductor with a negative transconductance. Also. 48. 61]. 2. Lossless Karsilayan-Schaumann Active Inductors The inductance of the inductor can be derived from the small-signal analysis of the inductor with the assumption gds = 0.7 Karsilayan-Schaumann Active Inductors The active inductor proposed by Karsilayan and Schaumann is shown in Fig.68 v in R Cgs v gs vin R CMOS Active Inductors gm vgs |z(jw)| R gm 0 Figure 2. They are the same as those if gm R gm gm ωt of the corresponding Hara active inductor investigated earlier.39. A. Rs > 0.40 [74. It becomes evident that gm > R 1 1 RCgs R .

sCgs3 s(Cgs1 + Cgs2 ) + (gm1 + gm2 ) (2.(2. The capacitance and inductance are given by Cp = Cgs .92) Further assume M1 and M2 are perfectly matched.(2.Implementation of Single-Ended Active Inductors 69 J1 J3 vin 1 Cgs1 M1 M3 CI C gs3 J2 2 CQ M2 3 Cgs2 Vb Figure 2. i. 2 2Cgs3 . gm1 = gm2 = gm and Cgs1 = Cgs2 = Cgs .93) Eq.40.9) directly.93) reveals that the active inductor can be represented by a capacitor in parallel with an inductor. Eq. 2 (2.(2.94) reveals that the inductance of the active inductor can be increased by increasing the capacitance between the gate and source of M . The differentially-configured transconductor with only one of its two input terminals is connected to the input has a gm . Eq.94) It is interesting to note that the preceding results can also be obtained using the results given in (2. Simplified schematic of Karsilayan-Schaumann active inductor. This can be 3 . The capacitance encountered at the input node of the transconductance 2 Cgs as capacitors of Cgs1 and Cgs2 are connected active inductor is given by 2 in series. L= gm gm3 (2.92) is simplified to Y = 1 2Cgs3 s gm gm3 +s Cgs . + gm2 gm3 sCgs1 + gm1 .e.

gm2 vsg2 v gs1 v in C4 C1 1 g o3 gm3 vgs3 Figure 2. The inductance of the active inductor can be tuned in this way. The small-signal equivalent circuit of the active inductor is shown in Fig.41. as shown in Fig.40.97) 1 G(ω) 2C3 jω + gm gm3 gm gm3 . 2 g o1 3 g o2 gm2 vgs2 vsg3 C3 C 2 vsg2 Small-signal equivalent circuit of Karsilayan-Schaumann active inductor.2. Lossy Karsilayan-Schaumann Active Inductors The preceding analysis excludes the effect of gds and other parasitic capacitances of the transistors. gm gm3 (2. gm (2. and C4 represent the total capacitances including both intrinsic and parasitic capacitances encountered at or between respective nodes. B. C3 .2. we investigate the quality factor of the active inductor and its tunability by following the approach of Karsilayan and Schaumann. It was shown in [74] that the admittance of the active inductor is given by Y ≈ where G(ω) = go2 + 2go4 − ω 2 C3 C1 + C2 .95) The auxiliary capacitor CI can be implemented using MOS varactors.70 CMOS Active Inductors achieved by adding an auxiliary capacitor CI in parallel with Cgs3 . C2 . The inductance in this case becomes L= 2(Cgs3 + CI ) . Due to the absence of the lossy conductance g .41 where C1 . the ds quality factor of the active inductor can not be studied. In what follows.96) . (2.

101) the quality factor of the active inductor will become infinite. It was demonstrated in [74] that the quality factor of the active inductor was made nearly 400 and the inductance of the active inductor exceeded 600 nH in a 0. which is the capacitance encountered at the source of M1 and M3 while the inductance of the active inductor is tuned by varying CI .e.2.43.18μm CMOS technology. Variations of Karsilayan-Schaumann Active Inductors To increase the speed of the active inductor and to reduce the silicon consumption. as shown in Fig. it was shown by Xiao and Schaumann in [61] that the preceding Karsilayan-Schaumann active inductor can also be implemented using all nMOS transistors (excluding biasing current sources). An important observation is that Q is tuned by varying CQ . C2 = (go2 + 2go4 )gm − C1 .2.2. C.40. To achieve this. as shown in Fig. Q and L can be tuned independently. In [86.98) in series with a resistor of resistance Rs = G(ω) .Implementation of Single-Ended Active Inductors 71 The active inductor can be represented by an inductor of inductance L= 2C3 gm gm3 (2. 87].68 GHz and a quality factor of 106. the common-source configured transconductor of the preceding Karsilayan-Schaumann active inductor was replaced with a static inverter to boost the transconductance of the transconductor from gm3 to gm3 + gm4 .100) Q≈ It is observed from (2. this active inductor exhibited a self-resonant frequency of 6. as shown in Fig. The inductance is tuned by varying CQ while the quality . an auxiliary capacitor CQ can be added at the source of M1 and M2 . Implemented in TSMC-0. = Rs G(ω) (2. In other words. gm gm3 (2. i. the capacitance of the auxiliary capacitor added between the gate and source of M3 .100) that if we set G(ω) = 0. ω 2 C3 (2.42.99) The quality factor of the active inductor is obtained from 2ωC3 ωL .5μm CMOS implementation of Karsilayan-Schaumann active inductor.

the effect of both Rs and Rp of an active inductor must be compensated for.42. respectively. Simplified schematic of Karsilayan-Schaumann active inductor using a staticinverter negative transconductor. A disadvantage of this design is the stringent constraint imposed on the voltage swing of node 2 of the active inductor. The L ∼ Rs series branch of the RLC equivalent circuit of an active inductor can be replaced with . Simplified schematic of Karsilayan-Schaumann active inductor using nMOS transistors only. in order to ensure that M3 and M4 are in the saturation.43.8 Yodprasit-Ngarmnil Active Inductors It was pointed out earlier that to boost the quality factor. vo J2 2 vin M4 1 M3 J1 CQ CI M1 M2 Vb VOH Transition region VOL V IL V IH VOH v in Figure 2. The input voltage of the static inverter must satisfy VIL ≤v2 ≤VIH where VIL and VIH are the lower and upper voltage bounds of the transition region of the static inverter.3.72 CMOS Active Inductors J2 J3 v in 1 M1 C gs1 J1 C gs3 Figure 2. 2. 2 M2 CI Vb C gs2 M3 CQ factor is tuned by changing CI .

To boost the quality factor. the added electrical connection between the input terminal of the active inductor and the drain of M2 forms the needed positive feedback.102) J4 M3 v in 1 2 M1 3 M2 J3 v in Vb -R p.2. such that the net resistive loss of the active inductor vanishes.total Rp R p. ˆ Both Rp and Rp can be combined into a single parallel resistor of resistance ˆ Rp.44. In Fig.44(b). (b) Simplified schematic of Yodprasit-Ngarmnil active inductor.total . subsequently a decrease in iD2 as iD1 + iD2 = J2 (constant). The impedance looking into the gate of M1 at low frequencies is given by Zin ≈ − 1 gm1 + 1 gm2 (2. This is because an increase in vin will result in an increase in iD1 .2. This is echoed with an increase in the drain voltage of M2 .44(a). a negative resistor of resistance −Rp.total can be connected in parallel with Rp.2. as shown in Fig. The preceding analysis reveals that the differential pair offers two distinct functions simultaneously : .44(b) with the inductance and the shunt resistance given by 1 ˆ L = L(1 + 2 ).total = Rp ||Rp .total Rp Cp L J1 J2 (a) Figure 2. It was also shown earlier that negative resistors are realized by using positive feedback. which will further increase vin . Q ˆ Rp = Rp (1 + Q2 ).103) .Implementation of Single-Ended Active Inductors 73 ˆ the L ∼ Rp parallel branch shown in Fig. (2.

Q was tuned up to 12000 in a 0. the preferred choice is therefore to vary ro of M1 and M2 . which can be achieved by employing the cascode configuration of the differential-pair transconductor and varying the gate voltage of M4 and M5 .(2. Because the former also changes the inductance of the inductor. 2.2 . Transconductances gm4. As demonstrated in [85].6μm implementation of the active inductor.2.104) Eq. respectively.5 can be tuned by varying the gate voltage of M4.45. Cgs1 2Cgs3 + ro1 ro3 Q= (2.5 . it was shown in [85] that the quality factor of Yodprasit-Ngarmnil active inductor is given by gm3 gm1 Cgs3 Cgs1 .2.3. ro1 and ro2 in Fig.9 Uyanik-Tarim Active Inductor The CMOS active inductor proposed by Uyanik and Tarim in [88] with its simplified schematic shown in Fig.46 only has two transistors connected in . as shown in Fig.2 or ro1.74 CMOS Active Inductors It behaves as a transconductor with a negative transconductance to construct the gyrator-C active inductor. When go is considered.2.45. J3 M3 2 M4 v in 1 J4 3 Vg M5 M2 Vb M1 J1 J2 Figure 2. It provides the needed negative resistance between the input terminal and the ground to cancel out the parasitic resistances of the active inductor.104) shows that Q can be tuned by either changing gm1.45 now become (gm4 ro4 )ro1 and (gm5 ro5 )ro2 . Simplified schematic of Cascode Yodprasit-Ngarmnil active inductors.

3.105) The zero of Z(s) is at frequency . To transconductance gm3 tune the inductance of the active inductor without affecting the parasitic series resistance of the active inductor.8-6. The minimum number of transistors stacked between the power and ground rails also enabled the active inductor to have a large input signal swing of 18 mV.46.4 GHz with its phase error less than 1 degree. As seen in Fig. making it very attractive for low-voltage applications. The inductance was from 38 nH to 144 nH. J M2 C vin M1 M4 M3 Figure 2.10 Carreto-Castro Active Inductors The BiCMOS active inductors proposed by Carreto-Castro in [89] can also be implemented in CMOS technologies. a varactor C is added between the gate of M2 and the ground. occurring at 5.4 form a transconductor with a positive gm2 gm4 = gm2 .47. 2. The maximum quality factor was 3900. The quality factor of the active inductor exceeded 100 in the frequency range 4. Simplified schematic of Uyanik-Tarim active inductor. M1 and J form a transconductor with a negative transconductance gm1 . which controls the quality factor of the active inductor.2V CMOS technology. gm Cgs s +1 gm (2. as shown in Fig. provided that M3 and M4 are identical. The input impedance of nMOS Carreto-Castro active inductor is given by Z= 1 sRCgs + 1 .32 GHz.13μm 1.46.3 GHz to its self-resonant frequency of approximately 7. Implemented in UMC-0.Implementation of Single-Ended Active Inductors 75 series between VDD and ground rails. M2.75 GHz. Neglecting Cgd and go of the transistor.2.3. the simulation results in [88] shows that the active inductor had a wide frequency range from 0.2.

106) gm .76 1 RCgs CMOS Active Inductors ωz = and the pole of Z(s) is at frequency ωp = (2. This condition also ensures that ωz < ωp . Rgm − 1 R . L= R2 C . ωz for practical applications. In the frequency range ωz < ω < ωp . Because ωp = Rgm . 1 R (2. Rgm − 1 (2.111) 1 is usually required to maximize the effective . the circuit is inductive.108) that the parameters of the RLC equivalent circuit of the active inductor are given by Rp = R. Rgm > 1 (2.110) is required.108) It is seen from (2.107) The inductance of Carreto-Castro active inductor can be derived by examining the input admittance = = sC + gm sRC + 1 1 1 + R 1 RC s + 1 gm − R gm − Y . Rgm frequency range.109) Rs = In order to have L > 0 and Rs > 0. Cgs (2. In this case (2.

50 [90. The frequency of the go1 . Note that a modification in the polarity of the transconductors is required to ensure the existence of a negative feedback in the gyrator-C configuration. gm ωt 1 Rs ≈ . on the other hand.18μm 1. can be lowered zero of active inductors given by ωz = Cgs2 by either increasing Cgs2 or decreasing go1 . 2.47. Thanachayanont and Payne proposed the cascode active inductors shown in Fig. ωz should be minimized and ωo should be maximized. The former is at the cost of lowering ωo and should therefore be avoided.112) It is interesting to note that Hara active inductors. and Carreto-Castro active inductors all have the same expressions for L and Rs .11 Thanachayanont-Payne Cascode Active Inductors It was pointed out earlier that to maximize the frequency range of active inductors. J v in R (a) nMOS Figure 2. as shown in Fig.2.48 and 2. 65. v in R J (b) pMOS Simplified schematic of Carreto-Castro active inductors. Wu folded active inductors.Implementation of Single-Ended Active Inductors 77 L≈ RC R = .8V CMOS technology and analyzed using SpectreRF with BSIM3V3 device models.3. Figs.49 show the dependence of the resistance and reactance of a nMOS Carreto-Castro active inductor on R and J. Maximizing ωo is rather difficult because ωo of active inductors is set by the cut-off frequency of the transconductors constituting the active inductors. 73]. Cascode can be implemented in either of the two transconductors. The active inductor was implemented in TSMC-0. To reduce go1 .2. gm (2.2. The impedance of Thanachayanont-Payne cascode active inductor is given by .50.

W = 10μm. Dependence of Z of nMOS Carreto-Castro active inductor on R. R = 2kΩ.49.18μm. R = 2kΩ. . W = 10μm.48. 600 100 μA 400 Impedance [Ω] 200 μA 300 μA 400 μA 200 0 Im(Z ) in 100 μA 200 μA 300 μA 400 μA Re(Z ) in −200 −400 10 −1 10 0 Frequency [GHz] 10 1 10 2 Figure 2.78 800 600 400 Impedance [Ω] 200 Im(Z ) 0 −200 −400 −600 −1 10 in 4 kΩ 3 kΩ 2 kΩ 1 kΩ CMOS Active Inductors Re(Z ) in 4 kΩ 3 kΩ 2 kΩ 1 kΩ 10 0 Frequency [GHz] 10 1 10 2 Figure 2. Circuit parameters : L = 0. Circuit parameters : L = 0.18μm. Dependence of Z of nMOS Carreto-Castro active inductor on J.

113) that the frequency of the zero of Z is reduced from ωz = without the cascode to go1 Cgs2 (2.50.Implementation of Single-Ended Active Inductors 79 J2 M2 Vb M3 v in M1 J1 J2 M2 Vb J1 J2 (a) Vb1 J1 (c) Figure 2.114) .c) Thanachayanont-Payne cascode active inductors. (b. (2. It is seen from (2. M1 v in (b) M2 M1 M3 Vb2 v in go1 go3 Z≈ Cgs1 Cgs2 gm3 s 2 Cgs2 gm3 go1 go3 +1 gm1 gm2 + Cgs1 Cgs2 .113) go1 go1 go3 s +s + Cgs2 gm3 Cgs1 where gm go was utilized in simplifying the results. (a) Basic active inductor. Simplified schematic of Thanachayanont-Payne cascode active inductors.

gm1 gm2 It is evident from (2.119) with the cascode. The self-resonant frequency of the active inductor given by √ gm1 gm2 = ωt1 ωt2 . gm1 gm2 gm3 ro3 (2. The parameters of the RLC equivalent network of the cascode active inductor can be obtained by examining the admittance of the cascode inductor and the results are given by Rp = 1 . The parallel resistance is increased from . It should not be surprised to see that cascodes do not change ωo .115) with the cascode. remains unchanged. Rs = L= go1 1 .80 go1 1 Cgs2 gm3 ro3 CMOS Active Inductors ωz = (2. go2 Cp = Cgs1 . Cgs1 Cgs2 ωo = (2.116) however.118) Rs = (2. Cascode configurations thus can effectively expend the frequency range of active inductors by lowering the lower bound of the frequency range of active inductors.117) that the cascode active inductor has the same inductance as that of the corresponding non-cascode gyrator-C active inductor. The parasitic series resistance is reduced from Rs = without the cascode to go1 1 gm1 gm2 gm3 ro3 go1 gm1 gm2 (2. This is because cascodes are not subject to Miller effect and has no effect on the bandwidth.117) Cgs2 .

124) with the cascode.50(c). For the cascode active inductor of Fig. For the cascode active inductor of Fig. Quality factor improvement by lowering the parasitic series resistance R s and increasing the parasitic parallel resistance Rp . .Implementation of Single-Ended Active Inductors 81 (2.125) ωL = ωCgs2 ro1 (gm3 ro3 ) Rs (Rs dominates) (2.123) ωL = ωCgs2 ro1 Rs (Rs dominates) (2. No reduction in the inductance.121) with the cascode. as is evident from Q≈ or Q≈ gm2 Rp = ωL ωCgs2 (Rp dominates) (2. go2 (2. the cascode configurations of gyrator-C active inductors offer the following attractive characteristics : Frequency range expansion by lowering the lower bound of the frequency range.2. Both improve the quality factor of the active inductor.min = 2VT + Vsat . The minimum supply voltage of the active inductor without the cascode is given by VT + 2Vsat . VDD.122) without the cascode and Q≈ or Q≈ gm2 Rp = (ro2 gm1 ) ωL ωCgs2 (Rp dominates) (2.120) Rp = without the cascode to Rp = 1 gm1 1 .50(b). No reduction in the upper bound of the frequency range.min = VT + 2Vsat .2. VDD. To summarize.

Cgs1 Cgs2 (2.3.51.126) Q(ωo ) = ωo L gm3 = Rs go1 go3 gm1 gm2 Cgs2 . Because L = gm1 gm2 gm1 gm2 gm3 go2 have gm1 gm2 . Rs = . This is achieved by using regulated cascodes and multi-regulated cascodes. an adjustment of Q is therefore required after each tuning of ωo . we J1 .2. and Rp = .52 [91]. Note that because the tuning of ωo .3. Simplified schematic of Weng-Kuo active inductor. So the tuning of Q can be made independent of ωo . J1 M2 Vb vin J2 M3 J3 M1 Figure 2. will affect Q.13 Manetakis Regulated Cascode Active Inductors The performance of the preceding cascode active inductors can be improved by further reducing Rs . The parameters of the . It is seen that gm1 is proportional to J1 + J3 while gm3 is only proportional to Cgs2 go1 go3 1 . however. The simplified schematic of Weng-Kuo active inductor is shown in Fig.51. Cp = Cgs1 .2.12 Weng-Kuo Cascode Active Inductors Weng and Kuo proposed a current-reuse cascode active inductor in [62] to eliminate the drawback of the preceding Thanachayanont-Payne cascode active inductors that the inductance and quality factor can not be tuned independently. Cgs1 ωo = ωo can be tuned by varying gm1 and gm2 while Q can be tuned by varying gm3 only. 2. as shown in Fig.82 CMOS Active Inductors 2.

The parameters of the RLC equivalent circuit of the multi-regulated cascode gyrator-C active inductor are given by Go1 = 1 go1 (ro3 gm3 )(ro4 gm4 )(ro5 gm5 ) . .129) It is evident from (2.Implementation of Single-Ended Active Inductors 83 RLC equivalent circuit of the regulated cascode gyrator-C active inductor are given by Go1 = 1 . C2 = Cgs1 . C2 = Cgs1 . gm1 gm2 (2.127) C1 = Cgs2 . go1 (ro3 gm3 )(ro4 gm4 ) (2. Go2 Cp = C2 . (2.129) that the lower bound of the frequency is further reduced while the upper bound of the frequency range remains unchanged. Go2 = go2 . C1 = Cgs2 . from which we obtain Rp = 1 .128) C1 . Go2 = go2 . gm1 gm2 Go1 1 and ωo = . the lower bound of the frequency is C1 LCp reduced while the upper bound of the frequency range remains unchanged. Rs = L= Because ωz = Go1 .

regulated cascode active inductors and multi-regulated cascode active inductors given in [91] demonstrated that regulated cascode reduced the lower frequency bound of active inductors by one decade while multi-regulated cascode further reduced the lower frequency bond by more than one decade. in [92]. Simplified schematic of Manetakis regulated and multi-regulated cascode active inductors. 2go2 + Rf go2 2 gm1 go2 go3 + ω 2 [gm2 Cgs1 − gm1 Cgs1 Cgs2 (Rf go2 + 1)] . By assuming that the biasing current source transistors Mn and Mp are ideal.53 were proposed by Hsiao et al. Rp = Rs = Rf go2 + 1 2 .52. To further improve the quality factor. 2. 94]. This type of active inductors was further investigated in [93. The simulation results of cascode active inductors.14 Hsiao Feedback Resistance Cascode Active Inductors It was shown earlier that cascode active inductors offer a large frequency range and a high quality factor. 2 2 gm1 gm2 gm3 + ω 2 gm2 gm3 Cgs1 . (a) Regulated cascode active inductors.2. the parameters of the RLC equivalent circuit of the active inductor are given by Cp = Cgs3 . feedback resistance active inductors shown in Fig.130) L= 2 gm1 gm2 Cgs1 + ω 2 Cgs1 Cgs2 (Rf go2 + 1) .3. 2 2 gm1 gm2 gm3 + ω 2 gm2 gm3 Cgs1 (2. (b) Multi-regulated cascode active inductors.84 CMOS Active Inductors J3 M2 J2 vo M4 v in J1 -A M1 M4 v in M3 v in J2 Vb M5 v o J3 M2 M3 -A M1 J1 v in (a) (b) Figure 2.

Note Rf does not consume any static power.131) It should be noted that the self-resonant frequency of the active inductor with the feedback resistors Rf is reduced due to the increase of L.18μm CMOS technology showed that the inductance of the inductor was 15 nH with the quality factor exceeding 50 and the self-resonant frequency of several GHz [92–94]. . CI and CQ are MOS varactors to tune the inductance and quality factor of the active inductor. The simulation results of a feedback resistance active inductor implemented in a 0. where the cascode portion of the active inductor is replaced with a regulated cascode branch. respectively. as shown in Fig.55. Both boost the quality factor of the active inductor.2. 2. Abdalla et al. Eq. go2 go2 go3 .Implementation of Single-Ended Active Inductors 85 It is seen from (2.2. thereby boosting the quality factor of the active inductor.15 Abdalla Feedback Resistance Active Inductors Karsilayan-Schaumann active inductors studied earlier offer the key advantage of the independent tuning of their inductance and quality factor.3. The preceding Hsiao feedback resistance cascode active inductors were further developed by Liang et al. as shown in Fig. as shown in the figure. In [95]. For the special case where Rf = 0.(2. modified Karsilayan-Schaumann active inductors by adding a feedback resistor between the two transconductors of the active inductor to improve the quality factor of the inductor. gm1 gm2 (2.54 so that the advantages of the regulated cascodes active inductor investigated earlier can be utilized [50]. gm1 gm2 gm3 Cgs2 . It was shown in [94] that the decrease of the self-resonant frequency of the feedback resistance active inductors can be compensated for by varying the biasing voltage V of the p current-source transistor Mp . The resistance of the feedback resistor Rf can be tuned by connecting a nMOS transistor in parallel with a poly resistor.130) becomes Cp = Cgs3 . The added feedback resistor increases the inductance of the active inductor and at the same time lowers the parasitic series resistance of the active inductor.130) that the added feedback resistor Rf lowers Rs and increases L simultaneously. Rp ≈ Rs ≈ L= 1 .

J2 J3 M3 M4 M1 Rf M2 vin J1 Figure 2. and C2 . It is a cascode active inductor with a negative feedback network consisting of R1 .16 Nair Active Inductors Wei et al.53. Simplified schematic of Liang feedback resistance regular cascode active induc- 2.56. This active inductor was modified and implemented in CMOS by Nair et al.86 Vf Vp Mp Rf M2 Vb v in M1 Mn Vn CMOS Active Inductors Vp Mp Rf M2 M3 v in M1 Mn Vn (a) Feedback resistance active inductor Figure 2.54. R2 . It was shown in [97] that the parameters of this active inductor are given by .3. The simplified schematic of Nair active inductor is shown in Fig. tors. (b) Cascode feedback resistance active inductor Simplified schematic of Hsiao feedback resistance active inductors. in [97] and was used in design of a low-power low-noise amplifier for ultra wideband applications.2. proposed a MESFET-version high-Q active inductor with loss compensated by a feedback network in [96].

57 shows the configuration of Wu current-reuse active inductors with replica biasing [66]. This is because a V fluctuation DD not only alters the dc operating point of the transconductors constituting the active inductors. it also changes the junction capacitances of the active inductors.3. Simplified schematic of single-ended Abdalla feedback resistance active induc- L= ωo = C3 + Cgs3 . such as the inductance and parasitic resistances. gm1 gm3 gm1 gm3 .132) Q(ωo ) = It is seen from (2. 2.Implementation of Single-Ended Active Inductors 87 Rf J3 v in 1 M1 J2 2 M2 CI Vb M3 J1 CQ Figure 2. It also increases the quality factor and decreases the self-resonant frequency of the active inductor. Replica-biasing is an effective means to reduce the effect of supply voltage fluctuation on the inductance of active inductors. R2 − C2 is a negative feedback network that reduces the parasitic resistances of the active inductor. The . are sensitive to the fluctuation of the supply voltage of the active inductors. 2 Cgs1 go1 (2.132) that C3 is used to boost the inductance.17 Active Inductors with Low Supply-Voltage Sensitivity It was pointed out earlier that the parameters of gyrator-C active inductors.55.2. Fig. The RC network consisting of R1 . tors. Cgs1 (C3 + Cgs3 ) gm1 gm3 (C3 + Cgs3 ) .

The voltage of the non-inverting terminal of the amplifier will also increase.6 and an auxiliary voltage amplifier. M3 M6 M2 v in M1 Vb1 Vb2 M5 Vb1 M4 Figure 2. The width of the transistors in the replica-biasing section should be the same as that of the active inductor section so that both will sense the same voltage change caused by the variation of VDD .6 . Simplified schematic of Wu current reuse active inductor (nMOS) with replica .88 CMOS Active Inductors R2 C2 R1 M2 J1 M3 C3 C 1 M1 J2 vin Figure 2. minimizing the effect of VDD fluctuation.57. subsequently the channel current of M3. Simplified schematic of Nair active inductors. The output of the amplifier will increase the gate voltage of M3. replica-biasing section consists of a sensing circuit made of M 4.56.6 is kept unchanged approximately. An increase in VDD will lead to an increase in vSG3.6 . biasing. which ensures that vSG3.5.6 .

To find out the parameters of the RLC equivalent circuit of the active inductor. One of the simplest differ- . Transistors M5 and M6 are biased in the triode region and behave as voltagecontrolled resistors whose resistances are controlled by the gate voltage V . The preceding development reveals that the inductance of the active inductor can be tuned by varying V . gds5 [gm1 + gm3 + s(Cgs1 + Cgs3 )] (2.134) 2(gds5 − gm1 ) . (2.Implementation of Differential Active Inductors 89 2.4.58 [60]. 2.4. The negative feedback of the active inductor is as the followings : an increase in v1+ and a decrease in v1− will result in an increase in v2+ and a decrease in v2− due to the common-gate operation of M1.4 Implementation of Differential Active Inductors 2. gds5 (2gm1 + gm3 − gds5 ) 2(Cgs1 + Cgs3 ) .133) The parameters of the RLC equivalent circuit of the active inductor are given by Rp = Rs = L= 2 gds5 . subsequently b gds5 . gds5 (2gm1 + gm3 − gds5 ) Note that 2gm1 + gm3 > gds5 and gds5 > gm1 are required in order for Rs and L to have a positive value. It can be shown that the differential input impedance of the inductor is given by Z= 2[s(Cgs1 + Cgs3 ) − gm1 + gds5 ] .2 Grozing Floating Active Inductors ¨ As pointed out earlier that floating gyrator-C active inductors can be constructed using a pair of differential transconductors. respectively. in the small-signal equivalent circuit of the active inductor. It is a differentially configured gyrator-C active inductor. The source followers of M3. we represent M5 and M6 with channel conductances gds5 and gds6 . lowering gds5 and gds6 . is shown in Fig. An increase in Vb will push M5 and M6 from the triode region towards the saturation region.1 Lu Floating Active Inductors The schematic of the active inductor proposed by Lu et al. All b other transistors are biased in the saturation.2. This is echoed with an increase in the inductance of the inductor.2 .4 ensure that v1+ and v1− will be reduced accordingly by approximately the same amount.

4. Transistors M4. 73] is shown in Fig.08 GHz.2 . It was shown in [43. The quality factor exceeded 100 over the frequency range from 0.59 employs two basic differential-pair transconductors. Note that the resistance of the negative resistor can not be tuned in this implementation. 44] that in a 0. The inductance of the active inductor is tuned by varying the transconductances of the transconductors. This is attained by changing the tail current sources J3. the quality factor of the active inductor was 600 at 2 GHz while the self-resonant frequency of the active inductor was 5.5 form a negative resistor to cancel out the parasitic resistances of the active inductor so as to boost its quality factor.2.8 GHz. entially configured transconductors is the basic differential pair. The quality factor of the active inductor is tuned by varying the resistances of the negative resistors. Two negative resistors are connected across the output nodes of the transconductors to cancel out the parasitic resistances of the active inductor.18μm implementation of the active inductor.83 GHz to 1. The floating active inductor proposed in [43. This is done by adjusting the tail currents of the differential pairs J1.33 GHz with its maxima of 1970 at 1.60.35μm CMOS technology offered an inductance of 70 nH and a self-resonant frequency of 2. It consists of two cascode-configured gyrator-C active inductors investigated earlier. 99. 38.3 Thanachayanont Floating Active Inductors The floating active inductor proposed in [37. The active 2 inductor implemented in a 0.2. 98.4 of the negative resistors. 65. Simplified schematic of Lu floating active inductor.6 GHz 2. 44] and shown in Fig.58. .90 Vb M5 M3 2M6 2+ CMOS Active Inductors v in+ M4 Rp L v inRs v in- v in+ M1 1+ J1 M2 1J2 Figure 2. The inductance of the active inductor is tuned by varying J .

Simplified schematic of Thanachayanont floating active inductor. It consists of a pair of differential transconductors and a pair of negative resistors . 2.59.4.4 Mahmoudi-Salama Floating Active Inductors The floating active inductor proposed by Mahmoudi and Salama was used in the design of quadrature down converter for wireless applications [56.60.2. o J1 M4 M5 J1 V b1 M3 M1 M2 Vb2 M3 M2 M1 V b1 J2 vin J2 Figure 2. The schematic of Mahmoudi-Salama floating active inductor is shown in Fig. Simplified schematic of Gr¨zing floating active inductor.4 V b2 M7 M8 -R v in+ M5 M6 v in- J2 Figure 2.61.Implementation of Differential Active Inductors 91 V b1 M3 M1 M4 -R M2 J1 J 3. 100].

92

CMOS Active Inductors

at the output of the transconductors. M8,16 are biased in the triode and behave as voltage-controlled resistors. They are added to the conventional cross-coupled configuration of negative resistors to provide the tunability of the resistance of the negative resistors without using a tail current source.
V b1 M12 M13 V b3 M16 2C M14 M10 M15 M11 A

2C

M9 V b2

V b1 M4 V b3 M8 2C M6 vin+ M2 B V b2 M1 M7 M3 vin2C M5

Figure 2.61.

Simplified schematic of Mahmoudi-Salama floating active inductor.

The small-signal equivalent circuit of the tunable negative resistor is shown in Fig.2.62 where a test voltage source Vx is added for the derivation of the equivalent resistance of the negative resistor. R represents the resistance of M8 . Writing KCL at nodes 1 and 2 yields gm1 V2 + Vx − Ix = 0 R (node 1), (2.135)

Vx + Ix = 0 (node 2). gm1 (Vx + V2 ) − R The resistance of the negative resistors at low frequencies is obtained from (2.135)

Implementation of Differential Active Inductors
Ix 1 R M1 M2 g m1v 2 v2 v1 g m2v 1 2 1 R Vx 2

93

ht
Figure 2.62. Small-signal equivalent circuit of Mahmoudi-Salama floating active inductor at low frequencies. go of the transistors is neglected.

Z =

Vx =− Ix

1 1 + gm1 gm2 1 1 R− + gm1 gm2 R 1 gm1 + 1 gm2 . (2.136)

= R// −

The inductance of the active inductor is given by L =

C , where Gm1 Gm2 2C is the total capacitance encountered at each of the output nodes of the transconductor, Gm1 and Gm2 are the transconductances of the differential transconductors 1 and 2, respectively. By assuming that nodes A and B are the virtual ground and neglecting Cgd and the diffusion junction capacitances, we Cgs2,3,10,11 and Gm = gm2,3,10,11 . have C≈ 2

2.4.5

Feedback Resistance Floating Active Inductors

The feedback resistance technique studied earlier was also employed in the design of floating active inductors by Akbari-Dilmaghani et al. in [101] to improve the performance of these inductors. A similar approach was used by Abdalla et al. in design of high-frequency phase shifters [102]. This section investigates these active inductors. The schematic of the feedback resistance floating active inductor proposed by Akbari-Dilmaghani et al. is shown in Fig.2.63. It consists of two basic differential-pair transconductors and two feedback resistors. The functionality of the added feedback resistors is the same as that of the single-ended active

94

CMOS Active Inductors

inductors investigated earlier, i.e. lowering the parasitic series resistance and increasing the inductance.
V b1 M9 M10

v in+ M7

M8

v in-

M6 V b2 Rf M4 V b1 M5 Rf

M2

M3

V b2

M1

Figure 2.63. inductor.

Simplified schematic of Akbari-Dilmaghani feedback resistance floating active

The simplified schematic of Abdalla differential feedback resistance floating active inductors is shown in Fig.2.64, M11,12 are biased in the triode and behave as voltage-controlled resistors. It was shown in [102] that the inductance and the parasitic series resistance of the floating active inductor are given by Rf RT

L=

C + Cgs4,5 1 + gm1,2 gm4,5

, (2.137)

1 − ω 2 Cgs4,5 CRf RT , Rs = gm1,2 gm4,5 where

Class AB Active Inductors

95

C = Cgd7,8 + Cdb1,2 + Cdb7,8 + Cgs1,2 , RT = Rf ||Rds11,12 ||ro1,2 ||ro7,8 .

(2.138)

It is evident from (2.137) that Rf boosts L and lowers Rs simultaneously. Both improve the performance of the floating active inductor. Also seen from (2.137) and (2.138) that M11,12 control the series resistance Rs of the active inductor. By adjusting Vb1 , Rs can be minimized.
M9 M10

V b1 v in+ M4 M12 v inM5

M6 V b2 R Rf M7 M8

f

V b1 M1 M11 M2

V b2

M3

Figure 2.64. ductors.

Simplified schematic differential Abdalla feedback resistance floating active in-

2.5

Class AB Active Inductors

The active inductors investigated up to this point fall into the category of class A active inductors. These class A active inductors suffer from a common drawback of a small voltage swing at the terminals of the active inductors, mainly due to the constraint that the input transistors of the transconductors of the active inductors should be biased and operated in the saturation. Note that

It is seen that the need for a large minimum supply voltage of class AB active inductors makes them less attractive for low-voltage applications. The network thus exhibits an inductive characteristic over a large input voltage range. It should also be noted that when VT + Vsat ≤vin ≤VDD − (VT + Vsat ).4. class AB configurations can be employed.e. the two active inductors are operated in an interleave manner.96 CMOS Active Inductors although the input transistors of the transconductors of active inductors can be pushed into the triode region while still exhibiting an inductive characteristic. A class AB active inductor can be constructed from a nMOS-configured class A active inductor and a pMOS-configured class A active inductor. J2 M2 V b1 J1 M1 v in V b1 V b2 v in V b2 J1 M1 M2 J2 M2 M1 v in J2 M3 M4 J1 (a) Class A active inductors Figure 2. When the input voltage is low. (b) Class AB active inductor Simplified schematic of Thanachayanont-Ngow class AB active inductor. as shown in Fig. only the nMOS class A active inductor is activated. The swing of the input voltage and the minimum supply voltage of the class A and class AB active inductors are compared in Table 2. the nonlinearity of the active inductors will increase. depending upon the swing of the input voltage. both the nMOS and pMOS class A active inductors are activated. only the pMOS class A active inductor is activated. i.2. .65. To increase the voltage swing of the active inductors without sacrificing linearity. When the input voltage is high.65 [65].

A comparison of the input voltage swing and the minimum supply voltage of class A and class AB active inductors.67. . Negative resistance compensation techniques can be employed to improve the performance of class AB active inductors.min Vin.min Vsat VDD − (VT + Vsat ) VT + 2Vsat Class A (pMOS) VT + Vsat VDD − Vsat VT + 2Vsat Class AB 0 VDD 2VT + 2Vsat Class AB active inductors can also be configured in cascodes to increase the frequency range and to boost the quality factor. (b) Cascode class AB active inductor Simplified schematic of Thanachayanont-Ngow cascode class AB active inductor. as shown in Fig.2. Class A (nMOS) Vin.4.66.Chapter Summary 97 Table 2. Floating cascode class AB active inductors can also be constructed in a similar manner as that of non-cascode floating active inductors.max VDD. J2 M2 V b1 J1 V b2 vin V b1 M1 M3 M2 M1 M3 V b3 v in J2 vin V b1 M1 M3 M2 M5 V b2 M4 M6 V b4 V b2 J1 (a) Cascode class A active inductors Figure 2.2.66. as shown in Fig.

linearity. One of the key advantages of active inductors over their spiral counterparts is the large tunability of their inductance. The former is often used for the coarse control of the inductance whereas the latter is used for the fine control of the inductance. Both are due to the finite input and output impedances of the transconductors constituting active inductors. Frequency range specifies the lower and upper bounds of the frequency in which gyrator-C networks are inductive. To provide a quantitative measure of the performance of active inductors. We have shown that both single-ended and floating (differential) active inductors can be synthesized using gyrator-C networks. signal sensitivity. noise. We have shown that the inductance of gyrator-C active inductors can be tuned by varying either the transconductances of the transconductors or the load capacitance. We have shown that the low frequency bound is set by the frequency of the zero of the gyrator-C networks while the upper frequency bound is set by the frequency of the pole of the networks. supply voltage sensitivity. and power consumption. inductors.98 CMOS Active Inductors J2 V b1 M1 M2 M15 M16 M8 M7 J2 V b1 M3 V b3 Vb3 M9 M6 Vb4 Vb4 M12 M10 V b2 J1 M4 V b2 J1 M5 M13 M14 M11 Figure 2. inductance tunability. characteristics.67. The distinct sensitivities of the quality factor of active inductors to their parasitic series and parallel resistances have been investigated. a number of figure-of-merits have been introduced. We have . Simplified schematic of Thanachayanont-Ngow cascode floating class AB active 2. parameter sensitivity. quality factor. These figure-of-merits include frequency range. stability.6 Chapter Summary An in-depth examination of the principles. Lossless gyrator-C networks yield lossless active inductors while lossy gyrator-C networks synthesize lossy active inductors. and implementation of gyrator-C active inductors in CMOS technologies has been presented. topologies.

such as Karsilayan-Schaumann active inductors that have a large output impedance.Chapter Summary 99 shown that the parasitic series resistance of an active inductor can be converted to an equivalent parallel parasitic resistance such that a negative shunt resistor can be employed to cancel out the effect of both the series and parallel parasitic resistances of the active inductor simultaneously. employing cascodes. To increase the tunability of active inductors. transconductors with a large input / output resistance are critical. or adding parallel negative resistors. A distinct characteristic of active inductors is the dependence of their performance on the swing of their signals. . The linearity of active inductors has been investigated. To increase the voltage swing of the active inductors without sacrificing linearity. This in turn lowers the quality factor of the active inductors. Both improve the quality of the active inductors. the effect of this dependence must be accounted for. We have used supply voltage sensitivity and parameter sensitivity to quantify the effect of these unwanted variations. The basic gyrator-C active inductors suffer from a small parasitic parallel resistance and a large parasitic series resistance due to the small output / input impedance of the transconductors constituting the active inductors. The second part of the chapter has focused upon the CMOS implementation of gyrator-C active inductors. This can be achieved by changing the configuration of transconductors. both singleended and floating. Feedback resistance active inductors. The input-referred noise generators of basic transconductors have been derived using the approach for noise analysis of 2-port networks whereas those of active inductors have been obtained using the approach for noise analysis of 1-port networks. Class AB gyrator-C active inductors have also been studied. This approach is often preferable over those that vary the transconductances of the transconductors of active inductors because the latter also alters other parameters of the active inductors such as inductance with a downside that the tuning range is rather small. Because active inductors are active networks that are sensitive to both supply voltage fluctuation and parameter spread. class AB active inductors constructed from a pair of nMOS/pMOS-configured class A active inductors have been presented. When active inductors are used in applications such as LC oscillators where signal swing is large. Two main drawbacks of class AB active inductors are the circuit complexity of these inductors and their need for a high supply voltage. An emphasis has been given to the noise of active inductors as these inductors exhibit a high level of noise power. exhibit improved performance by increasing the inductance and decreasing the series parasitic resistance simultaneously. The schematics and characteristics of singleended and floating (differential) class A active inductors have been investigated in detail. To increase the parallel parasitic resistance and to lower the parasitic series resistance. additional voltagecontrolled capacitors can be employed at critical nodes of active inductors.

.

com/978-0-387-76477-1 .springer.http://www.

Sign up to vote on this title
UsefulNot useful