Beruflich Dokumente
Kultur Dokumente
) are alternately
charged to the voltage of the previous stage (V
p
) and then boosted by the same voltage
level to charge the load at a higher output voltage. The clock signals of
1
and
2
are
bootstrapped to the same level as V
p
to connect the two capacitors in series.
The voltage doubler of Fig. 1.5 usually consists two latched CMOS pairs in each
stage [5]. The complementary voltage swings of the internal nodes are used to control
the switches of opposite branches. This circuit eliminates the voltage drop at the out-
put switches, reduces the output voltage ripple, and uses only two nonoverlapping phases.
Moreover, the voltage across each transistor is never higher than the power supply voltage
V
DD
. At high output currents, the overdrive voltage decreases causing the output resistance
to rise due to higher switch resistance, thus increasing resistive power losses and reducing
power efciency and driving capability. Moreover, a short-circuit loss from higher voltage
Chapter 1. Introduction 5
Figure 1.3: Simplied schematic of the boosted voltage generator for DRAM word-line
driver [3].
Figure 1.4: Double charge pump [4].
Chapter 1. Introduction 6
nodes to lower voltage node exists during transitions. The resulting short-circuit current
reduces the charge pump efciency and output voltage.
The two series pMOS transistors (P and P
_
1 if i is the node with the lowest number in a separate part of
the closed switch network, and node j belongs to that separate part
0 otherwise
(2.1)
where 1 i j and 1 j n.
A nn capacitance matrix Cdescribes the CP capacitors in terms of their values, node
connections, and parasitics, and can be expressed as
C(i, j) =
_
_
total capacitance connected permanently to node i if i = j
negative of the total capacitance between i and j if i = j
(2.2)
The CP independent voltage source and ground (i.e. grounded switches are connected
to a zero value voltage source) connections are described by an n 2 matrix G, whose
elements are dened as
G(i, h) =
_
_
1 if the h-th voltage source is connected to node i
0 otherwise
(2.3)
where 1 i n and 1 h 2.
The capacitance matrix and the voltage sources matrix do not change as the switches
change states. The node voltages are represented by the (n1) vector v(t
k
), which denes
Chapter 2. Charge Pump Analysis 13
the voltage between the i-th node and ground at switch event t
k
. The charges delivered by
the independent voltage sources are represented by the (21) vector q
I
(t
k
), which denotes
the charge passed through the h-th voltage source from switch event t
k
to switch event t
k+1
.
In each phase, the closure of the switches imposes a set of (n l + 2) KVL equations and
l charge conservation equations, from which we nd the corresponding nodes voltage at
time t
k
and charges delivered by the sources during the interval (t
k
, t
k+1
). For a complete
solution, conservation equations can be compactly expressed as:
_
_
v
I
(t
k
)
S
k
Cv(t
k1
)
_
_
=
_
_
G
T
0
S
k
C+S
T
k
I S
k
G
_
_
v(t
k
)
q
I
(t
k
)
_
_
, (2.4)
where v
I
(t
k
) is a (2 1) vector which represents the independent voltage sources, and
I is the n n identity matrix. The (n + 2) (n + 2) matrix
k
in (2.4) can be rearranged
to obtain a solution for the nodes voltage v(t
k
) and the delivered charges q
I
(t
k
) as follows:
v(t
k
) = A
k
v
I
(t
k
) +B
k
S
k
Cv(t
k1
) (2.5)
and
q
I
(t
k
) = R
k
v
I
(t
k
) +O
k
S
k
Cv(t
k1
), (2.6)
where A
k
and B
k
are the upper-left n2 submatrix and the upper-right nn submatrix
of
1
k
, respectively. R
k
and O
k
are the lower-left 22 submatrix and the lower-right 2n
submatrix of
1
k
, respectively. In the case of a CP operating with two phases, in steady-
state v(t
k1
) = v(t
k+1
) and v(t
k
) = v(t
k+2
), therefore the CP voltage nodes and delivered
charges can be calculated.
Chapter 2. Charge Pump Analysis 14
Figure 2.3: Procedure for evaluating CP gain.
2.2 Charge Pump Gain
CPs achieve capacitive voltage conversion by means of transfer capacitors and switches
driven by nonoverlapping clock phases. Each transfer capacitor is charged to a certain
voltage level and then it is boosted by another voltage level resulting in a voltage increase
at the output terminal. Since CP circuits do not use inductors, they are well suited for
integrated implementations in planar conventional technologies.
2.2.1 Ideal Gain
The voltage gain Ais dened as the ratio between the maximumopen-circuit output voltage
V
O
and the input voltage V
DD
(assumed constant). Since no current is delivered to the load,
dependencies on the switching frequency and capacitances values are eliminated. When
ideal capacitors are assumed, the gain depends only on the number of capacitors N, the
number of phases, and the topology, which, in turn, determines how the transfer capacitors
are interconnected in each phase. The procedure for evaluating the voltage gain includes
disconnecting any load at the output and nding the output voltage as shown in Fig. 2.3.
Chapter 2. Charge Pump Analysis 15
C
Substrate
bC aC
Top Bottom
Figure 2.4: Integrated capacitor model.
2.2.2 Gain with Parasitic Capacitances
A key reason why the gain of a real integrated CP deviates from the ideal is the unavoidable
presence of parasitic capacitances, which share a portion of each charge packet transferred
between transfer capacitors resulting in reduced gain. Parasitic capacitances are expressed
by the technological parameters and , which give the stray parasitic capacitances C
(between bottom plate and substrate) and C (between top plate and substrate) of any
integrated capacitor C as shown in Fig. 2.4. The value of and are determined by the
process and the type of the integrated capacitors used (integrated capacitors are discussed
in detail in Chapter 3). To assess the impact of parasitic capacitances on the voltage gain
A, their values are included in the capacitance matrix Cby modelling the total capacitance
connected permanently to a node as ( + 1)C or ( + 1)C [18]. The gain with parasitic
elements is lower than the ideal gain, because a portion of each charge packet transferred
between stages is shared with the parasitic capacitors and wasted.
2.3 Charge Pump Output Resistance
In the case of ideal linear elements, the procedure for evaluating the output resistance in-
volves turning off the input voltage V
DD
, applying an ideal source V
X
to the output, and
calculating the ratio between the voltage and the average current of the applied source as
shown in Fig. 2.5. In a two-phase CP the output resistance [19] is given by
Chapter 2. Charge Pump Analysis 16
V
x
I
x
= q
x
f
s
V
IN
= 0
Charge
pump
Figure 2.5: Procedure for evaluating CP output resistance.
R
O
=
r
f C
T
, (2.7)
where f is the switching frequency, C
T
is the value of the total capacitance dened as
the sum of the capacitances of all transfer capacitors C
T
=
N
i=1
C
i
, and r is a constant
that depends on circuit topology which can be expressed as
r =
N
i=1
a
2
ci
, (2.8)
where N is the number of capacitors and a
ci
= q
i
/q
X
is the charge multiplier factor,
which is the ratio of the charge q
i
, transferred by capacitor C
i
in a period, and the charge
q
X
delivered to the load. The charge multiplier factors are calculated by applying charge
conservation to the circuit in phase 1 and 2, and by considering that, in steady-state, each
capacitor receives and delivers the same charge in each of the two phases.
2.3.1 Analysis of Output Resistance with Parasitic Capacitances
To evaluate the effect of parasitic elements on the output resistance, we include C
i
and
C
i
in the capacitance matrix, turn off V
DD
, connect a voltage source at the output, apply
the method above one more time, and nd the charge q
X
delivered by the voltage source
during the switching period, the corresponding current, and thus the output resistance. The
output resistance with parasitic elements is lower than the ideal, because it is inversely
Chapter 2. Charge Pump Analysis 17
proportional to the node capacitances that increase with the parasitics.
2.4 Power Losses in Charge Pumps
Charge pumps transfer charge packets from the power supply at a voltage V
DD
to an out-
put terminal at a higher voltage V
O
. In this operation, CPs dissipate a portion of the input
power and may reduce the benet of scaling the supply voltage down. The energy ef-
ciency is dened as the average power delivered to the load divided by the average of input
power. Power losses arise mainly from capacitor charging and discharging losses, resis-
tive conduction losses, and losses due to parasitic capacitances and short-circuit currents.
The highest efciency is achieved in slow switching conditions. In such conditions and in
steady-state, the main power losses are described by a simple model and can be divided
into load dependent losses and load independent losses [20].
2.4.1 Load-Dependent Losses
Load-dependent losses are revealed when the charge pump is connected to a load and the
output voltage decreases in the presence of a load current I
O
> 0. These losses are mod-
elled through a non-zero equivalent output resistance R
O
and the corresponding power
dissipation is
P
LD
= R
O
I
2
O
. (2.9)
This formula indicates that lower load dependent losses can be achieved by reducing
the output resistance R
O
given in (2.7) which is inversely proportional to the product of the
switching frequency f and the total capacitance C
T
.
Chapter 2. Charge Pump Analysis 18
2.4.2 Load-Independent Losses
Load independent losses are revealed when the CP is not connected to any load and it
still dissipates power. These losses mostly arise from charging and discharging parasitic
capacitances and are also called dynamic losses. They are modelled through a non-zero
equivalent input conductance G
I
and the corresponding power dissipation is
P
LI
= G
I
V
2
DD
. (2.10)
The dynamic losses of switch drivers and other auxiliary functions could be incorpo-
rated in the model as well. However, for the present we focus our analysis only on the
charge pump core. In this case, the input conductance is proportional to the product of the
switching frequency and the total capacitance:
G
I
= fC
T
g , (2.11)
where g is a constant that depends on circuit topology and parasitic capacitances. The
procedure for evaluating the input conductance involves disconnecting the output load and
calculating the charge delivered by the source V
DD
to the CP as shown in Fig. 2.3.
The output power of a charge pump (i.e. the power delivered to the load) is
P
O
= V
O
I
O
= (AV
DD
R
O
I
O
) I
O
, (2.12)
Assuming the gain A > 1, the charge pump has a conversion efciency given by
=
P
O
P
O
+P
LD
+P
LI
=
AV
DD
I
O
R
O
I
2
O
AV
DD
I
O
+G
I
V
2
DD
, (2.13)
Chapter 2. Charge Pump Analysis 19
which is maximum when the output current is equal to
I
O
=
G
I
V
DD
A
_
A
2
G
I
R
O
+ 1 1
_
. (2.14)
In this condition, the ratio
P
LI
P
LD
=
A
2
G
I
R
O
1
__
A
2
G
I
R
O
+ 1 1
_
2
, (2.15)
is larger than one for any acceptable value of A, G
I
, and R
O
. In other words, at peak
efciency, load-independent losses are larger than load-dependent losses (i.e. P
LI
> P
LD
at I
O
=
I
O
). In general, load-independent losses dominate at low currents such that
0 I
O
< V
DD
_
G
I
/R
O
, (2.16)
while for higher currents load-dependent losses are larger. Therefore minimizing load
independent losses is a crucial design objective, especially for charge pumps meant to
operate at peak efciency or at low currents.
2.5 Analysis of Single-Sided Charge Pumps
Several single-sided CP structures have been proposed in the literature, each suited to meet
specic application requirements and address process constrains. Single-sided CPs transfer
charge packets to the load once every switching period. Indeed, the differences between
CPs structures correspond to the conguration of their capacitors and switches in each
phase. Exploring different CPs structures is motivated by choosing the appropriate struc-
ture in order to maximize the efciency.
In the Dickson CPs in Fig. 2.6(a) [2], MOS switches controlled by non-overlapping
control phases eliminate the voltage drops associated with the diodes used in the classic
Chapter 2. Charge Pump Analysis 20
conguration [1]. Each transfer capacitor is charged to the voltage of the preceding stage
and then boosted by V
DD
to charge the next stage at a higher voltage. Ideally, a circuit with
N stages has a voltage gain A = N + 1, an output resistance R
O
= N
2
/(fC
T
), and an
input conductance G
I
= 0.
In the heap CP in Fig. 2.6(b) [16], the voltage across each capacitor never exceeds V
DD
making this type of CPs attractive for implementations in low-voltage processes. Aheap CP
with N stages has an ideal voltage gain A = N +1, an output resistance R
O
= N
2
/(fC
T
),
and an input conductance G
I
= 0.
The Fibonacci CP with three capacitors shown in Fig. 2.6(c) [12] has the same ideal
gain as the Dickson and the heap CPs with four capacitors (Figs. 2.6(a) and 2.6(b)). This
two phase CP single-sided structure has the highest attainable gain for a given number of
capacitors [13]. The gain of an ideal Fibonacci CP with N stages is A = F
N+1
, where F
N
is the N-th Fibonacci number, with F
0
= F
1
= 1 and F
i
= F
i1
+ F
i2
for i > 1. In the
case of equal transfer capacitors C
i
= C
T
/N, the charge multiplier factors are a
ci
= F
Ni
for i = 1 to N, the output resistance of this topology is R
O
=
N
fC
T
N
i=1
(F
Ni
)
2
, and the
input conductance G
I
= 0.
2.5.1 Optimization of the Output Resistance
To minimize the output resistance of any CP for a constant total capacitance C
T
, we sub-
stitute C
1
= C
T
C
2
... C
N
in (2.7) and we set the partials with respect to capacitors
C
i
equal to zero, which means
R
O
C
i
=
1
f
_
a
2
ci
C
T
C
2
... C
N
a
2
ci
C
2
i
_
= 0, (2.17)
for i = 2 to N [18].
Since the available silicon area is a critical constraint for a designer, the CPs capacitor
sizes, which are the largest portion of an integrated CP, are optimized to improve CPs per-
Chapter 2. Charge Pump Analysis 21
0
V
DD
1 2 2 1
1 1 2 2 1
C
1
C
2
C
3
C
4
0
bC
2
bC
1
bC
3
bC
4
a(C
1
+C
3
) a(C
2
+C
4
)
V
DD
V
DD
V
O
Load
0
I
O
(a) Four-stage Dickson CP.
1
2
C
1
C
2
C
3
C
4
0 0
2 2 2
1
0 0
1 1
2
V
DD
1 1 1 1 V
O
Load
0
I
O
bC
3
bC
2
bC
1
bC
4
aC
1
aC
2
aC
3
aC
4
(b) Four-stage heap CP.
(c) Three-stage Fibonacci CP.
Figure 2.6: Schematic diagrams of conventional charge pumps with parasitic capacitances.
formance. Considering the three structures (i.e. the Dickson, the heap, and the Fibonacci)
and the calculated charge multiplier factors for each structure, the optimal capacitor sizes
are found. The optimal performance of an N-stage CP is not necessarily obtained when
capacitances are equal, but when they scale as a function of the charge multiplier factor.
Chapter 2. Charge Pump Analysis 22
C
1
C
2
C
3
C
4
C
1
C
2
C
3
C
1
C
2
C
3
C
4
F
i
b
o
n
a
c
c
i
D
i
c
k
s
o
n
H
e
a
p
A = 5
C
1
C
2
C
3
C
4
C
1
C
2
C
3
C
4
C
5
C
6
C
7
C
1
C
2
C
3
C
4
C
5
C
6
C
7
A = 8
F
i
b
o
n
a
c
c
i
D
i
c
k
s
o
n
H
e
a
p
C
1
C
2
C
3
C
4
C
5
C
1
C
2
C
3
C
4
C
5
C
6
C
7
C
8
C
9
C
10
C
12
C
1
C
2
C
3
C
4
C
5
C
6
C
7
C
8
C
9
C
10
C
11
C
12
A = 13
F
i
b
o
n
a
c
c
i
D
i
c
k
s
o
n
H
e
a
p
C
11
Figure 2.7: Sketch of capacitors with optimal size of Dickson, heap, and Fibonacci CPs of
equal area and gain (i.e. left A = 5, centre A = 8, right A = 13).
For instance, the optimal performance of an N-stage Fibonacci CP is when capacitors are
scaled as the Fibonacci sequence with the largest capacitor next to V
DD
and the smallest
next to the load.
When the capacitors are optimized as shown in Fig. 2.7, the three CPs have a simi-
lar performance. In this case, the trade-off between gain A and output resistance can be
expressed as [18]:
R
O
=
(A1)
2
fC
T
. (2.18)
2.5.2 Single-Sided Charge Pumps with Parasitic Capacitances
The design parameters for the CPs shown in Fig. 2.6 are calculated as a function of and
. For the Dickson CP the gain is
A =
N
1 +
+ 1 , (2.19)
the output resistance is
R
O
=
r
fC
T
, (2.20)
with
r =
N
2
(1 +)
, (2.21)
Chapter 2. Charge Pump Analysis 23
0
0.2
0.4
0.6
0.8
1 3 5 7 9 11
g
Voltage Gain A
Dickson
Fibonacci
Heap
Figure 2.8: Normalized input conductance g of Dickson, heap, and Fibonacci CPs as a
function of A, when = 0.1 and = 0.05.
and the input conductance is
G
I
= fC
T
+ +
1 +
, (2.22)
with
g =
+ +
1 +
. (2.23)
The analytical expressions for A, r, and g in the case of the optimized heap and the
Fibonacci CPs are collected in Table 2.1 and 2.2. The performance comparison indicates
that the Dickson CP performs the best since bottom plate parasitics does not contribute
to the gain reduction as in other structures where a signicant portion of charges delivered
to the output is shared with the parasitic capacitances C
i
associated with the bottom plate
of the transfer capacitors. Also, the input conductance of the Dickson CP is independent
of the number of stages. Accordingly, the load-independent losses of Dickson CPs depend
only on and , the total capacitance, the switching frequency, and V
2
DD
. On the other
hand, the heap CP has the worst performance, exhibiting a much lower gain than other
topologies at large number of stages N.
Chapter 2. Charge Pump Analysis 24
0
4
8
12
16
1 2 3 4 5
r
Voltage Gain A
Dickson
Fibonacci
Heap
Figure 2.9: Normalized output resistance r of Dickson, heap, and Fibonacci CPs as a
function of A, when = 0.1 and = 0.05.
Table 2.1: Heap CP Design Parameters.
N Parameters
1 A = 1 +
1
1+
r =
1
1+
g =
++
1+
2 A = 1 +
2+
1++(3+)+
2
r =
2(2++)
(1++(3+)+
2
)
g =
(5++2)(++)
2(1++(3+)+
2
3 A =
(2++)(2++(4+)+
2
)
1+
2
(1+)+(2+)(3+)+(3+2(3+))
r =
3(1++)(3++)
(1+
2
(1+)+(2+)(3+)+(3+2(3+)))
g =
((++)(14+
2
+4(2+)+(14+3)))
(3(1+
2
(1+)+(2+)(3+)+(3+2(3+))))
4 A =
((1++(3+)+
2
)(5+
2
+(5+)+(5+2)))
(
3
(1+)+(3+)(2+3(2+))+
2
(5+3(3+))+(1+)(1+(3+)
2
))
r =
4(2++)(2+
2
+2(2+)+(4+))
(
3
(1+)+(3+)(2+3(2+))+
2
(5+3(3+))+(1+)(1+(3+)
2
))
g =
((++)(14+
2
+4(2+)+(14+3)))
(3(1+
2
(1+)+(2+)(3+)+(3+2(3+))))
Chapter 2. Charge Pump Analysis 25
Table 2.2: Fibonacci CP Design Parameters.
N Parameters
1 A = 1 +
1
1+
r =
1
1+
g =
++
1+
2 A = 1 +
2
1+
r =
4
1+
g =
++
1+
3 A =
(1+)+(2+)(5+2(4+))
(1+)(2++(5+)+2
2
)
r = 4
2(1+)+((2+)(4+5)
(1+)(2++(5+)+
2
g =
(++)(3(1+)+(2+)(7+8))
4(1+)(2++(5+)+2
2
)
4 A =
48+(1+)(8+5)+(4+)(35+6(4+))
(3+2)(2+3)(1+(3+))+
2
(1+)
2
+(1+)(7+(16+7))
r = 7
42+2
2
(1+)+(19+3(12+5))+(109+(80+17))
(3+2)(2+3)(1+(3+))+
2
(1+)
2
+(1+)(7+(16+7))
g =
(++)(108+5
2
(1+)+(49+92+38
2
)+(277+200+42
2
))
(7(1++(3+)+
2
)(6++(13+)+6
2
))
Examples of numerical values of g and r for various values of Aare respectively shown
in Fig. 2.8 and Fig. 2.9, where they are plotted as function of A for = 0.1 and = 0.05. In
the Dickson CP, the normalized input conductance g does not change with the number of
stages, while the voltage gain A and the output resistance does not depend on the value of
, because the circuit can be built so that the bottom plates of all capacitors are alternately
connected to ground and V
DD
without affecting the charge transfer through the CP. For the
Fibonacci and heap CPs, the normalized input conductance g depends on the number of
stages because bottom plate parasitic capacitances share a portion of the charge transferred
to the output and therefore affect performance.
2.6 Analysis of Double Charge Pumps
The output voltage ripple can be reduced by splitting the CP in two parts, each part with half
the total capacitance and feeding the load in a different half period [4]. This conguration,
called double CP, is usually implemented as a parallel connection of stages operating with
opposite phases. Fig. 2.10 shows implementations of double CPs of the circuits seen in the
Chapter 2. Charge Pump Analysis 26
last section. The voltage gain A, the output resistance R
O
, and the input conductance G
I
are the same as the single-sided CPs when the total capacitance is the same. On the other
hand, the voltage ripple dened as the peak to peak variations in the DC output voltage, is
halved with respect to the single-sided charge pump and can be expressed as [21]
V
ripple
=
I
O
2 f C
L
, (2.24)
where C
L
is the load capacitance.
Another CP structure used to achieve a high voltage gain is the exponential CP shown
in Fig. 2.10(d). The gain of an ideal exponential CP with N stages is A = 2
N
. In the case
of equal transfer capacitors C
i
= C
T
/(2N), the charge multiplier factors are a
ci
= 2
2(Ni)
for i = 1 to N, the output resistance is R
O
=
N
fC
T
N
i=1
2
2(Ni)
, and the input conductance
G
I
= 0. The optimal performance of an exponential CP with N stages is obtained using
(2.17), and the minimum output resistance is when capacitors are sized as
C
i
=
2
Ni
2
N
1
C
T
. (2.25)
A comparison between the optimal capacitors sizes of an ideal Dickson and an ideal
exponential CPs with the same performance is shown in Fig. 2.11
2.6.1 Double Charge Pumps Performance with Parasitic Capacitances
The analytical expressions for A, r, and g in the case of the optimized exponential CP are
collected in Table 2.3, while A, r, and g for the double Dickson, double Fibonacci, and
double heap CPs are the same as those of the single-sided implementations. Comparing the
performance of the exponential CP to the double Dickson, again the Dickson CP performs
the better since bottom plate parasitics does not contribute to the gain A reduction as
in other structures where a signicant portion of charges delivered to the output is shared
with transfer capacitances bottom plate parasitics and wasted every clock cycle. Also, the
Chapter 2. Charge Pump Analysis 27
1 1 2
C C C
2
V
DD
1
C' C' C'
0
2
2
2 1
Load
0
I
O
V
O 1
V
DD
0 V
DD
2 1
V
DD
0
0 V
DD
0 V
DD
2 1
V
DD
0
2 1 2 1
2 1
1
C
C'
2
0 V
DD
2 1
V
DD
2 1
0
(a) Four-stage double Dickson CP.
1
1 1 2
C C C
0 0
2
2
2
1
0
1
2
V
DD
1
1
C' C' C'
0 0
2
2
0
1
2
2
1
2 1
Load
0
I
O
V
O
(b) Three-stage double Fibonacci CP.
2
C' C' C' C'
0 0
2
0 0
V
DD
1
2
C C C C
0 0 0 0
2
1
2
2
1
2
2
1
2
2
1
1
2
1
1
2
1
1
2
1
1
1
Load
V
O
0
I
O
(c) Four-stage double heap CP. (d) Three-stage double exponential CP.
Figure 2.10: Schematic diagrams of double charge pumps.
input conductance of the exponential CP depends on the number of stages. Accordingly,
the load-independent losses of exponential CPs are higher, because of the voltage swings
across the parasitic capacitances larger than V
DD
.
2.7 Charge Reuse
At low output current, the conversion efciency is largely set by parasitic capacitances.
In order to reduce dynamic power losses, charge reusing [11] is investigated to mitigate
these losses. If we consider those internal nodes of a conventional charge pump (Fig. 2.6)
that are connected to ground through a switch at every cycle, the parasitic capacitances
associated with them are charged to a certain voltage and then discharged to 0, therefore
the related charge is wasted in every cycle. We can reuse part of that charge (and therefore
Chapter 2. Charge Pump Analysis 28
Figure 2.11: Sketch of capacitors with optimal size of double Dickson and double expo-
nential CPs of area and gain (i.e. left A= 4, right A = 8).
Table 2.3: Exponential CP Design Parameters.
N Parameters
1 A =
2+
1+
r =
1
1+
g =
++
1+
2 A =
2.(2+)
2
(2++(5+)+2
2
)
r =
3(6++3)
(2++(5+)+2
2
)
g =
2(6++3)(++)
3(2++(5+)+2
2
3 A =
4.(2+)
3
2
(1+)+(2+)
2
(1+4)+(2+)(4+5)
r =
7(
2
+6(2+)+7(2+)
2
)
(
2
(1+)+(2+)
2
(1+4)+(8+14+5
2
))
g =
4(++)(
2
+6(2+)+7(2+)
2
)C
T
7(
2
(1+)+(2+)
2
(1+4)+(8+14+5
2
))
Chapter 2. Charge Pump Analysis 29
2
I
O
1
0
V
DD
V
O
Charge
pump 2
1
2
1 2 3
3 3 3
0
Charge
pump 1
0
OUT IN
OUT IN
0 0
Figure 2.12: Charge reuse conguration of a generic double CP.
save charges drawn from the power supply), if pairs of such nodes with complementary
voltage swings (i.e. 180
out of phase) are equalized before each switch event. Double CPs
clocked with opposite phases, have pairs of such nodes in each stage. Thus, charge reuse
can be applied to all stages of any double CP [20] as shown in Fig. 2.12.
Fig. 2.13 describes the charge reuse concept where an equalization switch driven by an
appropriate control signal is used to bring the nodes (X and X
) to an intermediate voltage
level. The time required by this operation is much smaller than the time needed for charging
the transfer capacitors, because only a small fraction (e.g. ) of the capacitance is involved.
Therefore, the time allocated for the equalization has a limited impact on the operating
frequency.
The principle of charge reuse is based on equalizing the voltages of the parasitic ca-
pacitances in each stage. The equalization switch controlled by phase 3 brings both ca-
pacitances to an intermediate voltage before each switch event, therefore the amount of
charges drawn from the power supply for charging parasitic capacitances is less than the
amount needed by conventional CPs. As a consequence, charge reusing reduces the load-
independent losses. As design examples, we consider double Dickson CP, double Fibonacci
CP, and a double heap CP. Applying charge reusing requires splitting the circuits into two
Chapter 2. Charge Pump Analysis 30
(a) Circuit to describe the charge reuse.
(b) Clock phases and internal
nodes voltage waveforms.
Figure 2.13: Description of charge reuse concept in double charge pumps.
symmetrical parts (double CP) driven by complementary control signals and operating in
parallel, as shown in Fig. 2.12. Examples of charge reuse application to the heap and Fi-
bonacci CPs are shown in Fig. 2.14(a) and 2.14(b), respectively. In these cases, charge
reusing not only reduces G
I
, it also increases A and R
O
.
2.8 Simulation Results with Charge Reuse
Three CP types (i.e. the Dickson, the heap, and the Fibonacci charge pumps) were designed
and simulated with Spectre using MOS switches and poly-diffusion capacitors in a standard
0.18-m technology. Fig. 2.15 shows the normalized input conductance g versus the gain
A. The reduction of g (and consequently of P
LI
) for the Dickson CP is 50%. On the other
Chapter 2. Charge Pump Analysis 31
(a) Three-stage double Fibonacci CP with charge reuse.
(b) Four-stage double heap CP with charge reuse.
Figure 2.14: Schematic diagrams of double charge pumps with charge reuse (parasitic
capacitances are omitted for simplicity).
hand, the improvement for the Fibonacci and heap CPs is less than 50% for gains larger
than two and depends on the number of stages. Fig. 2.16 shows the output characteristics
and the conversion efciency of the three CP types. The results are obtained when N = 4
for Dickson and heap CPs and when N = 3 for the Fibonacci CP. The output characteristics
of the Dickson CP is not changed, while the open-circuit gains of the Fibonacci and heap
CPs with charge reusing are improved (i.e. 1.9% and 8.7% increase, respectively), because
parasitic capacitances draw less charge from the primary charge transfer path. More signif-
icantly, charge reusing substantially improves the overall conversion efciency in any CP
type: The maximum efciency increases from 52.5% to 63% for the Dickson CP and from
Chapter 2. Charge Pump Analysis 32
0
0.2
0.4
0.6
0.8
1
1 3 5 7 9 11
g
Voltage Gain A
Dickson
Fibonacci
Heap
Dickson-reuse
Fibonacci-reuse
Heap-reuse
Figure 2.15: Normalized input conductance g versus voltage gain A for the three CP types
in standard conguration and with charge reuse, when = 0.1, and = 0.05.
23% to 31% for the heap CP, and from 43% to 53% for the Fibonacci CP. Reusing wasted
charges reduces the current drawn from the power supply and increases the conversion
efciency.
2.9 Summary
In this chapter, a method of analysis for evaluating integrated charge pumps performance
and optimizing their capacitor sizes is determined. The analysis allows the calculation of
the voltage gain A, the output resistance R
O
, and the input conductance G
I
and conse-
quently the major power losses (resistive and dynamic power losses) of any integrated CP
can be evaluated. Moreover, charge reuse is applied to with the result of reducing the dy-
namic power losses and improving the overall conversion efciency. The technique can
be applied to any double CP. The application of charge reuse results in reduced dynamic
power losses and a signicant portion of wasted charges is recovered every clock cycle.
The Dickson CP has the best performance in terms of the voltage gain and power ef-
ciency. When charge reuse is considered the double Dickson (voltage doubler) CP has a
Chapter 2. Charge Pump Analysis 33
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 100 200 300 400 500 600 700 800
C
o
n
v
e
r
s
i
o
n
E
f
f
i
c
i
e
n
c
y
I
O
Dickson
Fibonacci
Heap
Dickson-reuse
Fibonacci-reuse
Heap-reuse
( A) m
(a) Conversion efciency.
0
1
2
3
4
5
6
7
8
9
10
0 100 200 300 400 500 600 700 800
V
O
I
O
Dickson
Fibonacci
Heap
Dickson-reuse
Fibonacci-reuse
Heap-reuse
( A) m
(b) Output characteristics.
Figure 2.16: Conversion efciency and output characteristics of the three CP types as a
function of the output current I
O
, when n = 4 for Dickson and heap CPs and N = 3 for
the Fibonacci CP, V
DD
= 1.8 V, C
T
= 200 pF, f = 10 MHz, = 0.1, and = 0.05.
better performance.
Chapter 3. Design
Chapter 3
Design
3.1 Introduction
The designer of integrated charge pumps has to face the constraints of the fabrication tech-
nology. Typically, integrated CMOS circuits share a single substrate, thus the chip layout
geometry and the proximity of the process layers to the substrate produce parasitic capac-
itive couplings. The existence of such parasitics limits the charge pump performance and
efciency. Moreover, the performance of a CP depends critically on how its MOS switches
are controlled. First of all, the overdrive voltage applied to turn a switch on determines its
on resistance and drain-to-source voltage drop, which, in turn, affect the conversion ef-
ciency and voltage gain. In addition, the maximum and minimum voltages applied to the
switch gates affect the dynamic power losses and can be constrained by the device voltage
rating. Finally, precision and adjustability in controlling the switch affect the frequency
of operation (which trades off with the silicon area required for meeting design specica-
tions) and can prevent short-circuit currents from nodes at higher voltages to nodes at lower
voltages during transitions (which affect efciency).
Switch bootstrapping improves conduction during the on state by connecting a given
voltage between the gate and source terminals, typically by using a capacitor pre-charged
34
Chapter 3. Design 35
during the off state [22]. Moreover, reusing some of the charges that are normally wasted
for charging and discharging parasitic capacitances at each cycle is a promising approach
for reducing power dissipation in charge pumps [11].
In this chapter, we analyze and discuss the design aspects of integrated voltage doubler.
First, the standard voltage doubler limitations are pointed out. Second, we propose a new
voltage doubler with a switch bootstrapping technique, where the voltages driving the gates
of nMOS and pMOS switches can be controlled both in terms of voltage swing and timing
such that limitations of standard voltage doubler are alleviated. The application of the
technique is demonstrated through the design of various voltage doublers. Also, dynamic
power losses due to parasitic capacitances are addressed and a method for reducing them
through charge reuse is described. Simulations of the various voltage doublers conrm the
effectiveness of the proposed techniques which result in an improved overall performance.
Technology and design constrains are addressed as well, and design trade-offs are discussed
in order to ne tune the circuit components.
3.2 Voltage Doubler
In the bootstrapped Dickson CP [2], switch voltage drop, varying on resistance, and low
conduction are alleviated by using four non-overlapping clock phases, which also prevent
short-circuit currents from nodes at higher voltages to nodes at lower voltages. This imple-
mentation needs the generation of four appropriate clock phases and MOS switches able to
withstand high voltages.
The output voltage ripple can be reduced by splitting the CP in two parts each with
half the total capacitance and feeding the load in a different half period [4] as depicted in
(2.24). This conguration, called double CP, is usually implemented as cascade connection
of voltage doublers [5], [6], which need only two clock phases instead of four. As shown
in Fig. 3.1, each modular stage is made of two latched CMOS pairs (N
i
, P
i
, N
i
, P
i
),
Chapter 3. Design 36
(
(
Figure 3.1: Conventional 2-phases cross-coupled voltage doubler stage.
two transfer capacitors (C
i
, C
i
), and two drivers (N
Di
-P
Di
, N
Di
-P
Di
), and does not need
dedicated bootstrap drivers. V
i
is the output voltage of the i-th stage and V
i
is the input
voltage. The transfer capacitors of each stage are alternately charged to the voltage of the
previous stage and then boosted by V
DD
to charge the next stage at a higher voltage. The
complementary voltage swings on the internal nodes are used to control the switches of
opposite branches. Since the maximum voltage rise from V
i1
to V
i
is V
DD
, the voltage
across each device is never higher than V
DD
and low voltage MOS switches can be used.
In steady state, the operation of the voltage doubler (Fig. 3.1) is as follows; during the
rst half cycle,
1
= V
DD
and
2
= 0, transistors N
i
, N
Di
, P
i
, and P
Di
are on, and transistors
N
i
, N
Di
, P
i
, and P
Di
are off; transfer capacitor C
i
is charged to V
i1
through N
i
and N
Di
,
while transfer capacitor C
i
is boosted to V
i1
+V
DD
through P
i
and P
Di
. During the second
half cycle, and transistors N
i
, N
Di
, P
i
, and P
Di
are turned on, and transistors N
i
, N
Di
, P
i
,
and P
Di
are off; transfer capacitor C
i
is charged to V
i1
, while transfer capacitor C
i
is
boosted to charge next stage to V
i1
+V
DD
.
Chapter 3. Design 37
3.3 Losses and Efciency
In slow-switching conditions, the main power losses of integrated charge pumps can be
simply classied into load-dependent losses, load-independent losses [23], and short-circuit
power losses of phase drivers and main pass transistors.
3.3.1 Load-Dependent Power Losses
When the load current I
O
> 0, the output voltage V
O
of a voltage doubler with N stages
is reduced because of its non-zero equivalent output resistance R
O
and can be expressed as
[1]
V
O
=
_
N
1 +
+ 1
_
V
DD
N I
O
(1 +) 2 f C
i
(3.1)
The voltage rise per stage V for a voltage doubler is
V =
V
DD
1 +
R
O
N
I
O
. (3.2)
From (3.1), the maximum output current (i.e. when V
O
= 0) I
Omax
is limited to
I
Omax
=
N + 1 +
N
2 f C
i
V
DD
. (3.3)
In real implementations, the switches are designed with MOS transistors operating in
triode region. In particular, CMOS switches are turned on with an overdrive V V
t
(i.e.
for simplicity pMOS and nMOS switches are assumed to have the same threshold voltage
V
t
) and their on resistance R
ON
can be approximated as
R
ON
=
1
k (V V
t
)
(3.4)
where k = C
ox
W/L is the switch transconductance parameter. At high output cur-
rents, the overdrive voltage decreases according to (3.2) (i.e. V becomes low). In these
Chapter 3. Design 38
conditions the on resistance of each switch increases and if V V
t
the switches are off.
The CP output resistance given by
R
O
=
N
(1 +) 2 f C
i
coth
_
1
fC
i
R
ON
_
(3.5)
increases as well, thus making the load-dependent losses P
LD
= R
O
I
2
O
larger. Such losses
are particularly signicant at high output currents and low V
DD
, the maximum output cur-
rent I
Omax
in (3.3) is therefore reduced and a new maximum output current limit is ob-
tained. In other words, for the MOS switch to conduct in the triode region, it must satisfy
the relation
V V
t
(3.6)
which imposes an upper bound on I
O
, and the maximum output current becomes
I
Omax
= (V
DD
(1 +)V
t
) 2fC
i
(3.7)
3.3.2 Load-Independent Power Losses
In integrated voltage doublers, load-independent power losses (also called dynamic or
switching losses) can be calculated through the non-zero equivalent input conductance G
I
as explained in Chapter 2, so that the corresponding power dissipation is approximated as:
P
LI
=
+ +
1 +
fC
T
V
2
DD
(3.8)
Accordingly, the load-independent power losses of a voltage doubler depend only on
and , the switching frequency, the total capacitance, and V
2
DD
, and are independent from
the number of stages.
From the analysis in chapter 2, load-independent losses are the major power losses at
Chapter 3. Design 39
Figure 3.2: 2-phases cross-coupled voltage doubler stage.
low currents. Therefore, based on the design specications, minimizing load-independent
losses for voltage doublers meant to operate at maximum efciency or at low currents is
a critical design consideration. From (2.21), (2.22), and (2.16), a limit condition when
load-independent losses in voltage doublers dominate is found and can be expressed as
I
O
_
+ +
fC
T
V
DD
N
. (3.9)
3.3.3 Short-Circuit Power Losses
In the conventional cross-coupled voltage doubler Fig. 3.2, each stage is seen as a CMOS
latch, the gates of switches N
i
, P
i
and N
i
P
i
are driven by the voltage rise on nodes B
i
and B
i
. At this point, three major cases of reversion and short-circuit losses are identied.
First, in the time slot during transitions when the voltage rise across the stage is higher
than the overdrive of pass transistors N
i
or N
i
(i.e. V V
i1
+ V
t
), a reversion current
ows fromC
i
or C
i
back to node V
i1
. Second, in the time slot during transitions when the
Chapter 3. Design 40
voltage rise across the stage is lower than V
i
V
t
(i.e. V V
i
V
t
), pass transistors P
i
or P
i
are partially on, causing a reversion current from node V
i
back to C
i
and C
i
. Third,
the short time slot, when the CMOS pairs N
i
-P
i
and N
i
-P
i
are conducting simultaneously,
generates a short-circuit current from the higher-voltage node V
i
to the lower-voltage node
V
i1
. All these losses can degrade the CP efciency and the output voltage [24]. The short-
circuit power consumption depends mainly on the voltage rise per stage V , the input
transition time , the threshold voltage V
t
, and transfer capacitors (C
i
, C
i
) [25]
P
SC
= P
SC
(k, V
t
, V, , f, C
i
). (3.10)
Short-circuit losses are particularly signicant at low output currents , when V is high
compared to V
t
, while they are negligible when
V 2V
t
(3.11)
A limit condition on the output current I
O
range where short circuit losses are signi-
cant, can be obtained from (3.2) and (3.11)
I
O
2 f C
i
(V
DD
2 (1 +) V
t
) (3.12)
The problem can be alleviated by driving pMOS switches with level shifters generating
nonoverlapping control signals varying from 0 to V
i
[6] or by using two parallel stages
generating control signals varying from V
i1
to V
i
[8]. The problem can be solved by using
four nonoverlapping clock phases and bootstrapping the pMOS pairs [9] or by adding series
switches and using ve phases [26].
Chapter 3. Design 41
Figure 3.3: Proposed bootstrapping technique applied to a voltage doubler stage.
3.4 Proposed Switch Bootstrapping Technique
The problem of the increased MOS on resistance (reduced driving capability) can be solved
by boosting the voltage driving the main CMOS switches with a voltage swing that does
not vary with I
O
[10], a solution that improves the driving capability at low V
DD
, but does
not alleviate short circuit losses.
In order to prevent short-circuit currents and the reduced current driving capability ob-
served in the conventional voltage doubler, a new modular bootstrapping technique that
allows full control on MOS switches is proposed [27] and [28].
The circuit in Fig. 3.3, provides both control on the timing of the switch transitions
(therefore preventing short-circuit losses) and on the gate voltage swings (therefore im-
proving driving capability).
Having same pass transistors, transfer capacitors, drivers, and nonoverlapping phases as
the conventional one, the proposed circuit includes an nMOS cross-coupled clock booster
Chapter 3. Design 42
(N
bi
, N
bi
, C
ni
, C
ni
) driven by
1
and
2
and a pMOS cross-coupled clock booster (P
bi
,
P
bi
, C
pi
, C
pi
) driven by
1
and
2
. Short-circuit losses are prevented because the volt-
ages applied between the gate and source terminals of pairs N
i
- N
i
and N
Di
- N
Di
have
nonoverlapping transition times with both voltages low, whereas the gate-to-source volt-
ages of pairs P
i
- P
i
and P
Di
- P
Di
have complementary transition times with both voltages
high.
The timing of switch transitions and the nonoverlapping slots can be adjusted by con-
trolling the main clock phases. The amplitude of the gate voltage swings does not depend
on output current or number of stages and is controlled by the low and high levels of the
main phases, typically varying from 0 to V
H
= V
DD
. The corresponding voltage control-
ling N
i
and N
i
goes from V
i1
(off) to V
i1
+ V
H
(on) and the voltage controlling P
i
and
P
i
goes from V
i
V
H
(on) to V
i
(off). In steady state, the maximum voltage across any
switch is V
DD
and internal voltages are within the range from 0 to the maximum CP output
voltage.
3.5 Charge Reuse Technique
Since load-independent losses due to parasitic capacitances have a strong impact on con-
version efciency. Dynamic power losses can be reduced by reusing some of the charges
wasted in charging or discharging the parasitic capacitances each cycle [11], [20]. In partic-
ular, if we consider those internal nodes of a voltage doubler that are connected to ground
through a switch at every cycle and have complementary voltage swings, the parasitic ca-
pacitances associated with them are charged to a certain voltage and then discharged to
0, therefore a part of that charge can be reused (and therefore the input conductance is
reduced). This can be accomplished if we redirect some of the charges wasted at falling
nodes to charge parasitic capacitances at rising nodes before each switch event.
To that end, switches driven by appropriate control signals are used to equalize the
Chapter 3. Design 43
voltages of the parasitic capacitances. The time required by this operation is much smaller
than the time needed for charging the transfer capacitors, because only a small fraction
(e.g. , typically 1.5% to 20%) of the capacitance is involved and equalization switches
are sized to complete charge reuse within each nonoverlapping time slot. Therefore, the
time allocated for the equalization has a limited impact on the voltage doubler operation.
Furthermore, the control signal can be generated through a NOR gate directly from the
nonoverlapping control phases that are already needed to avoid short-circuit losses.
3.5.1 Charge Reuse Voltage Doubler Design
The design of a voltage doubler stage with charge reuse is shown in Fig. 3.4. The parasitic
capacitances C
i
and C
i
are alternately charged to V
DD
and discharged to 0. The equal-
ization switch controlled by a NOR circuit brings both capacitances to V
DD
/2 before each
switch event, therefore the amount of charges drawn from the power supply for charging
parasitic capacitances is half the amount needed by the conventional circuit. As a conse-
quence, charge reusing can reduce the load independent losses by a factor two. Circuit
analysis conrms that the input conductance of the voltage doubler CPs with charge reuse
is half that of conventional voltage doubler CPs:
G
I
= fC
T
+ +
2 (1 +)
, (3.13)
while the voltage gain A and the output resistance R
O
are unchanged.
3.6 Design Constrains
The design of efcient and high performance CPs is usually associated with several design
concerns that need to be addressed. Design considerations on MOS switches and boot-
strapping circuits play an important role in the proper operation of the charge pump.
Chapter 3. Design 44
Figure 3.4: Bootstrapped voltage doubler stage with charge reuse.
3.6.1 MOS Switches
The use of MOS transistors as switches requires that switches are designed appropriately.
MOS switches with a large aspect ratio are required mainly for three reasons. First, large
switches (i.e. with lowon resistances R
ON
) reduce resistive power losses. Second, to ensure
a small time constant (i.e. fast transient) of the charge transfer paths, large switches are
needed. Third, charge pumps require large switches if they have to deliver large currents.
In addition, the maximum switching frequencies at which a charge pump can operate
depend on the time constants of the individual stages. Each stage can be viewed as an
RC network, which needs MOS switches to have a relatively low on resistances so that
capacitor voltages can settle within the clock semi-period. Therefore, a number of time
constants within the half clock cycle are required for a complete charge transfer, and the
following relation must hold:
Chapter 3. Design 45
T
ON
>> R
ON
C
i
(3.14)
A frequency increase requires a reduction in the on-resistance of transfer switches,
which can be obtained by increasing the transistors aspect ratios (W/L), which also re-
quires larger drivers to maintain sharp transitions, and call for longer nonoverlapping time
(due to larger gate capacitance and, hence, transition times). This increases the contribu-
tions of the switches parasitic capacitances that adds to the capacitor parasitics, and, hence,
reduces the voltage gain A, increases the dynamic power losses, and reduces the efciency.
3.6.2 Bootstrapping Circuit
A key design issue of the proposed circuit involves sizing the boosting capacitor adequately
to bootstrap the gate of the pass transistors with the required overdrive voltage. The boosted
voltage on the gate of the pass transistor is reduced because loading capacitance C
load
(here we refer to MOS pass transistor capacitances and other parasitic capacitances) share
a portion of the charge. The added bootstrapping circuits are not on the primary charge
transfer path. However, these capacitors must be able to supply sufcient voltage swing to
the gate of the pass transistor and other parasitic capacitances. The boosted voltage can be
expressed as
V
g
= V
i1
+V
DD
C
N
C
N
+C
load
. (3.15)
In this design, the values of the bootstrapping capacitors C
N
are approximately 10 times
C
load
. This ensures that corresponding voltages controlling pass transistors are within the
required range. Furthermore, the precharge transistors (N
bi
, N
bi
,P
bi
, P
bi
) allow bootstrap-
ping capacitors to be charged to the required voltage level. The time required for such
operation (RC delay) is much less than the time required for charging transfer capacitors,
because bootstrapping capacitors are small and depend mainly on the gate size of the pass
Chapter 3. Design 46
transistor. Therefore, the area of precharge transistors is small as well.
3.6.3 Design Trade-Offs
To achieve satisfactory functional and performance results of the proposed design, several
Spectre simulations were performed in the 0.18-m technology. Fine tuning of the voltage
doubler components such as pass transistors and bootstrapping capacitor sizes was done to
maximize efciency and reduce area.
Fig. 3.5 demonstrates the maximum efciency of a one-stage voltage doubler as a func-
tion of the width of the pass transistors. In this design, the width of pMOS pass transistors
is scaled with respect to the width of nMOS pass transistors according to the mobility ratio
n
/
p
, which is about 2.5. At smaller transistors widths, the maximum efciency is limited
by the high on resistance of the switches. On the other hand, increasing the width of the
switches gives rise to dynamic power losses due to the intrinsic parasitic capacitances of
the switches.
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0 20 40 60 80 100 120 140 160 180
M
a
x
E
f
f
i
c
i
e
n
c
y
Pass transistor width ( m) m
Figure 3.5: Maximum efciency versus transistor width for a voltage doubler when N = 1,
V
DD
= 1.8 V, f = 10 MHz, C
T
= 250 pF, = 0.015, and = 0.01.
Fig. 3.6 shows the maximum efciency of a one-stage voltage doubler as a function of
Chapter 3. Design 47
the bootstrapping capacitor size. As the bootstrapping capacitor size increases, the max-
imum efciency increase. Further increase in the bootstrapping capacitor size will result
in an increased area occupation and reduced maximum efciency since larger drivers are
required to drive these capacitors and the associated parasitics.
0
10
20
30
40
50
60
70
80
90
0.1 1 10 100
M
a
x
E
f
f
i
c
i
e
n
c
y
Capacitance (pF)
Figure 3.6: Bootstrapping capacitor size versus the maximum efciency.
3.7 Technology Constrains
In this section, we examine issues related to design and implementation of fully integrated
voltage doublers in standard CMOS process. While in off-chip implementations the critical
design constraints are number of discrete components and board complexity, in on-chip re-
alization the key cost constraint is silicon area occupation. The area allocated to integrated
capacitors and switches depends on technological parameters, design specications, and
layout optimization. However, technology limitations, some of which are pointed out in
this section, determine the integrated devices characteristics.
Chapter 3. Design 48
3.7.1 Integrated Capacitors
In a digital CMOS technology, capacitors are made by superimposition of conductive and
dielectric layers such as polysilicon, metal, or diffused layers and dielectric layers of sili-
con dioxide (SiO
2
) or silicon nitride (Si
3
N
4
). The performance of a charge pump depends
critically on the properties of integrated capacitors, and in particular on parasitic capaci-
tances (expressed by the technological parameters and ), equivalent series resistance,
and capacitance per unit area.
Metal-metal capacitors can be constructed in an interleaved conguration to maximize
the capacitance utilization between available metal layers. To be more specic, capacitors
constructed with metal layers have a much smaller series resistance (hence time constant)
and can therefore operate at higher frequencies [29]. However, the specic capacitance of
metal-metal structures in standard CMOS processes is low, because metal layers for inter-
connections are separated by thick oxide layers (e.g. 1.35 min the considered technology)
to have minimal capacitive couplings. Such constraint limits the capacitance that can be
integrated in a reasonable silicon area and ultimately limits the CP driving capability. In
addition, metal-metal capacitors have high parasitic coupling to the substrate with respect
to the specic capacitance (up to 20%) and the parasitic capacitances C
i
and C
i
are
particularly high when the distance between the utilized metal layers and substrate is low.
MOS capacitors are constructed between a polysilicon layer and a diffused layer. These
layers are separated by a thin oxide layer (e.g. 4.1 nm in the considered technology) and the
capacitance per unit area is very high. The capacitance value that can be built in a specic
area depends on the technological parameter C
ox
(the capacitance per unit area of the gate
oxide) multiplied by the gate area [30]. The voltage dependence of the gate-to-channel
capacitance is limited for the considered design range. The effect of the gate-to-source
voltage on the gate-to-channel capacitance is shown in Fig. 3.7. The stray parasitic capac-
itances C
i
(between doped silicon and substrate) and C
i
(between the polysilicon layer
and the substrate) associated with capacitor C
i
are lower than other available structures. To
Chapter 3. Design 49
Figure 3.7: CV curve of nMOS capacitor (Spectre simulation).
Figure 3.8: Equivalent series resistance of MOS capacitor.
realize a specic capacitance value, both the length L and the width W of the polysilicon
layer are scaled. As a result, the related parasitic resistance, known as equivalent series
resistance, arises from both the gate sheet resistance R
g
and the channel resistance R
ch
.
The ESR of a MOS capacitor is shown in Fig. 3.8 and is given by [31]
ESR
R
ch
4
+R
g
. (3.16)
The polysilicon gate resistance can be expressed as
R
g
= R
W
L
(3.17)
Chapter 3. Design 50
where R
S
, and the capacitor C
S
to keep the bulk of the main pMOS switches
at voltage level V
S
V
O
. This is important when a small load capacitor C
L
is used or a
large output current is delivered, since the voltage ripple will increase according to (2.24).
Also, it should be noted that minimum-sized devices are used because the biasing capacitor
C
S
is not connected to the load.
Chapter 3. Design 51
C C'
P'
P N
N'
V
DD
CL
f
1 f
2
0
0
CS
P
S
P
S
0
R
L
VO
Figure 3.9: 2-phases cross-coupled voltage doubler stage with dynamic bulk biasing for
pMOS switches [6].
3.8 Design of CPs Auxiliary Circuits
3.8.1 Clock Generation Circuit
An external 50%duty-cycle reference clock cannot usually be utilized directly, but it is used
as an input for a clock generator, which produces the nonoverlapping clock signals. The
timing of switch transitions and the nonoverlapping slots depend on the clock generation
circuit shown in Fig. 3.10. Such circuit is simple and includes only cross-coupled NAND
gates, inverters, and even number of delay blocks. Each delay block is conceived by means
of an inverter and a voltage-controlled RC network that consists of a transmission gate and
an nMOS capacitor. Nonoverlapping time slots of the generated phases (
1
and
2
) depend
mainly on the low to high and the high to low propagation delays through the cascaded
delay blocks. For this design, to ensure enough time to control the switches, a longer
nonoverlapping time has been favored in order to prevent short-circuit power losses in the
drivers and main pass transistors. Nonoverlapping time was also controlled via an external
DC voltage to control the resistance of the transmission gate in the RC network and the
corresponding nonoverlapping time.
Chapter 3. Design 52
Figure 3.10: Nonoverlapping clock generation scheme (detailed schematic is shown in
Appendix A).
Figure 3.11: A CMOS inverter driver with tapering factor 4 (detailed schematic is shown
in appendix A).
3.8.2 Inverter Driver Circuit
Large transfer switches and related interconnects present a large capacitive load on the
clock phases path, therefore clock drivers are necessary to maintain sharp transitions and
reduce short circuit losses. To increase the energy efciency of a charge pump, drivers
are designed so that the power dissipation in the driver chain is minimized. Designers
often choose low-power tapered driver chains, which are constructed with cascaded inverter
stages whose sizes increase progressively by a scaling factor S, as an example, stages are
scaled with S = 4 to minimize the power delay product as shown in Fig. 3.11. In each
inverter stage, pMOS transistors are scaled with respect to the nMOS transistors according
to the mobility ratio [32]. As a result, the output transitions of each inverter have equal rise
and fall time delay.
Chapter 3. Design 53
3.9 Summary
This chapter discusses the fundamental design constraints of integrated voltage doublers.
Integrated devices capabilities and associated power losses in conventional designs are ad-
dressed with focus on resistive, dynamic, and short-circuit power losses. With all these
design aspects in mind, a new switch bootstrapping technique is proposed to overcome
these limitations and prevent short-circuit losses, improve driving capability, and enhance
the overall conversion efciency. Furthermore, a charge reuse technique is applied with
the result of reducing the dynamic power losses. Design and technology constrains are
discussed to optimize design parameters.
Chapter 4. Results
Chapter 4
Results
4.1 Introduction
This chapter presents the simulation results, prototype implementation, and experimental
results of the integrated voltage doublers. First, key simulation results of the designed
voltage doublers are shown and discussed at nominal process (TT), supply voltage (1.8 V)
and temperature (27