IEICE TRANS. ELECTRON., VOL.E86–C, NO.3 MARCH 2003
496
PAPER
A New Dynamic D-Flip-Flop Aiming at Glitch and
Charge Sharing Free
Sung-Hyun YANG†a) , Younggap YOU† , Nonmembers,
and Kyoung-Rok CHO† , Regular Member
SUMMARY A dual-modulus (divide-by-128/129) prescaler
has been designed based on 0.25-µm CMOS technology employing new D-flip-flops. The new D-flip-flops are free from glitch
problems due to internal charge sharing. Transistor merging
technique has been employed to reduce the number of transistors and to secure reliable high-speed operation. At the 2.5-V
supply voltage, the prescaler using the proposed dynamic D-flipflops can operate up to the frequency of 2.95-GHz, and consumes
about 10% and about 27% less power than Yuan/Svensson’s and
Huang’s circuits, respectively.
key words: dynamic D-flip-flops, prescalers, low-power circuits,
glitch free
1.
Introduction
Frequency synthesizers have been basic building blocks
in modern communication systems. One of the key elements of the frequency synthesizer is a dual-modulus
prescaler used in a fast phase-locked-loop (PLL).
High performance prescalers and voltage controlledoscillators are frequently used in high-speed applications. Costly bipolar or GaAs technologies have been
the basis of these high-speed parts, which are being replaced with rapidly improving low-cost CMOS technology. CMOS implementations become very attractive
due to their characteristics of high frequency operation,
low-power consumption yielding long battery life.
Dynamic D-flip-flops for high-speed operation and
low-power consumption are essential to high performance frequency synthesizers. Various flip-flops have
been proposed to improve the operating speed of dualmodulus prescalers [1]–[10]. The dynamic D-flip-flop
designs in previous studies found suffer from glitch and
charge-sharing problems [2], [3], which may result in incorrect operations. To alleviate these problems, additional transistors are introduced to some critical nodes
to make these nodes stable [4]–[6]. However, the newly
added transistors limit operating speed and increase
power consumption. In high-frequency operations, ratioed logic can replace ratioless logic without significant
penalty on power consumption [8]–[10]. Ratioed logic,
however, consumes high power in low-frequency operaManuscript received December 28, 2001.
Manuscript revised July 9, 2002.
†
The authors are with the Department of Computer and
Communication Engineering, Chungbuk National University, San 48, Gaesin-dong, Cheongju Chungbuk, Korea.
a) E-mail: shyang@hbt.chungbuk.ac.kr
tion, and their operating frequency range is limited.
In this paper, a new dynamic D-flip-flop without
glitch problems resulting from charge sharing is proposed, which is a basic building block of a high-speed
and low-power CMOS dual-modulus prescaler. The
proposed dynamic flip-flop is to ensure edge-triggering
operation independent of operating frequency. By reducing the charge sharing effect, more reliable operation can be achieved. This paper is organized as follows. In Sect. 2, serious problems of conventional dynamic flip-flops, such as glitch and charge sharing, are
reviewed. Section 3 describes the structure and operation of the proposed D-flip-flops. Simulation and experimental results are summarized in Sect. 4. Finally,
the conclusions are presented in Sect. 5.
2.
Conventional Dynamic D-Flip-Flops
Dynamic or clocked logic gates are used to decrease
circuit complexity, increase operating speed, and lower
power dissipation [12]. Of various dynamic CMOS circuit techniques, a true single-phase-clock (TSPC) dynamic CMOS circuit is operated with one clock signal
that is never inverted. Therefore, no clock skew exists
except for the clock delay problems, and even higher
clock frequency can be achieved [2], [11].
Figure 1 shows a TSPC D-flip-flop for high-speed
operation introduced in [1], [4]–[6]. The flip-flop consists of nine transistors, where the clocked switching
transistors are placed closer to power/ground for higher
speed [6]. The state transition of the flip-flop occurs at
the rising edge of the clock signal, clk.
Figure 2(a) shows the operation: Qb becomes high
Fig. 1
A TSPC D-flip-flop for high-speed operation [1].
YANG et al.: A NEW DYNAMIC D-FLIP-FLOP AIMING AT GLITCH AND CHARGE SHARING FREE
497
(a) D = 0, clk = 0 → 1: Qb → 1
Fig. 3
Simulation results of the D-flip-flop in Fig. 2.
(b) D = 1, clk = 0 → 1: Qb → 0
Fig. 2
Operation of a conventional dynamic D-flip-flop [1].
along clk changing low to high with D = 0. In Fig. 2,
the dotted lines are the conducting paths when clk =
0, and the solid lines when clk = 1. If D = 0 and clk
= 0, MPS1, MP1, and MPS2 are turned on and n1 , y1 ,
and y2 become high. If the signal clk changes low to
high, the node y2 is discharged to low through MN2 and
MNS1, making MP2 be on and Qb high. Figure 2(b)
shows the case with clk changes low to high and D =
1 making Qb be low. The analysis is extended to other
input combinations in the same manner.
2.1 Glitch and Charge Sharing Problems
Edge-triggered flip-flops take incoming data at the edge
of a clock signal. Glitch problems may occur making
the flip-flops fall into wrong states. Consider the circuit
in Fig. 2(a) with clk = 0 and D = 0, where y1 and y2 are
precharged to high voltage. If clk changes low to high,
the node y2 is discharged to low after some delay. In
other words, y2 remains high for a short time, in which
MN3 and MNS2 are turned on and Qb may change
to low. By discharging of y2 , then, Qb returns to the
correct state of high. A glitch may appear at Qb as
indicated (a) in Fig. 3.
Consider the circuit in Fig. 2 to discuss charge
sharing effects. When clk is low, the node y2 is always
precharged high making MN3 on. The nodes, then, Qb
and n3 may share their charges. The high level of Qb
is somewhat lowered by sharing charges with the low
level of n3 as indicated (b) in Fig. 3.
Glitches induced by charge sharing among internal nodes are illustrated in Fig. 3. With clk low and D
Fig. 4
Toggle configuration using D-flip-flop.
high, both n1 and y2 are precharged to the VDD level
and y1 is discharged to ground. Then clk changes to
high making MNS2 turn on and Qb low. Here, clk is
assumed high for a while and D changes high to low
instantly, then MN1 becomes off and MP1 turned on.
Nodes n1 and y1 share their charges through MP1 making y1 rise above the threshold voltage of MN2. With
clk high and MNS1 on, the node y2 discharges slowly,
which leads MP2 to be turned on and Qb to rise to
high. This is shown as (c) and (d) in Fig. 3. At the
point (c), if clk changes to low right after D changes
to low, there is no critical operation due to the small
amount of discharge at y2 and MP2 cannot be turned
on. However, the edge-triggering operation of the flipflop is prevented in the case that the discharging time
of y2 is long as shown at (d) in Fig. 3.
Charge sharing raises more serious problems when
the D-flip-flop is used as a toggle-flip-flop operating in
a lower frequency range. For toggle operations, Qb is
tied to D as shown in Fig. 4. By changing clk low to
high with Qb in a high state, Qb toggles to low. Since
Qb and D are tied together, D is forced to change high
to low right after clk changes low to high. The time
that D stays low is almost half the period of clk. In a
lower operating frequency, y2 would have enough time
to discharge making MP2 turn on strongly and Qb high.
The flip-flop, therefore, loses its edge-triggering characteristics and fails to perform proper operations.
IEICE TRANS. ELECTRON., VOL.E86–C, NO.3 MARCH 2003
498
Fig. 6
A D-flip-flop design proposed by Huang [6].
Fig. 7
Simulation waveforms of Huang’s D-flip-flop.
(a) Simulation waves at clk = 2 GHz
(b) Simulation waves at clk = 3 GHz
Fig. 5
Simulation waveforms of toggle-flip-flop of Fig. 4.
Figure 5 shows the simulation result of the toggleflip-flop. As shown in Fig. 5(a), when the frequency of
2-GHz is applied to clk, the toggle operation is not performed because of the long discharging time of the node
y2 . In Fig. 5(b), however, the flip-flop shows the proper
toggle operation under the sufficiently high frequency of
3-GHz because the discharging time of y2 is small. As
a result, in the low frequency the edge-triggering characteristic of the flip-flop gets worse and the reliability
degrades. The toggle-flip-flop in Fig. 4 operates properly above 2.5-GHz when it is designed using 0.25-µm
CMOS technology. Though the flip-flop violating the
edge-triggering feature can be used as a component of a
divide-by-4/5 synchronous counter in a high-frequency
dual-modulus prescaler, it cannot be used as a flip-flop
alone [7]–[10]. The edge-triggering characteristics are
very important for reliable operation.
2.2 Huang’s D-Flip-Flop
Huang proposed a D-flip-flop circuit shown in Fig. 6
aiming at the elimination of glitches as shown in Fig. 3
[4]–[6]. To alleviate the charge sharing, the small size
transistor MN5 prevents y1 from rising when D changes
high to low during clk high. MN5 and the inverter
INV1 make a pull-down path of y1 as follows. If D
changes high to low during clk = 1, n1 and y1 share
their charges. That is, y1 would rise up, but the output
of INV1 still has a high value turning on MN5. This
makes y1 low and stable resolving the charge sharing
problem.
The same analogy applies Qb having MP3 and
INV2. A transistor MN4 is introduced between MN3
and MNS2 to remove glitches. That is, MN4 driven
by INV1 breaks the pull-down path of Qb. When D
= 0 and clk = 0, nodes y1 and y2 are precharged to
high. If clk changes low to high, y2 discharges to low.
This operation cannot be completed instantaneously:
MN4 prevents Qb from pulling down to ground. Figure 7 shows the simulation waveforms with the clock
frequency of 100 MHz.
Introduction of MN4 in Fig. 6 brings a new problem as the point (a) of Fig. 7. For clk = 0 and D =
0, y1 and y2 are precharged to high. If clk changes
low to high, y2 discharges to low and Qb becomes high.
Though D changes low to high when clk = 1, Qb remains high. At this time, y1 = 0 and clk = 1, n3 and n4
are low because both MN4 and MNS2 are turned on.
If clk changes to low, y2 is precharged to high. Then
Qb shares charges with n3 and n4 through MN3 and
MN4. This makes the voltage of Qb fall down. Even
though MP3 can stabilize Qb, the time for returning to
a correct logic value is somewhat long.
3.
Proposed Dynamic D-Flip-Flop
We now introduce a new dynamic D-flip-flop eliminat-
YANG et al.: A NEW DYNAMIC D-FLIP-FLOP AIMING AT GLITCH AND CHARGE SHARING FREE
499
ing glitches and reducing the number of transistors. It
is based on a ratioed logic design technique and transistor merging.
3.1 Transistor Merging
Transistor merging is to reduce the number of transistors and thereby save both power and silicon area while
suppressing glitch occurrences. Pull-up and pull-down
transistors are combined together yielding a circuit having fewer pull-up and pull-down transistors. Consider
the circuit shown in Fig. 1, where the nodes n1 and y2
have the same potential of VDD during clk = 0. When
clk = 1, its operation is independent of the n1 level and
y2 may stay high or discharge to low. This observation
leads to merge two pull-up transistors of the conventional design. MPS1 and MPS2 in Fig. 1 are merged to
MPS1 as shown in Fig. 8(a). With the same analysis,
MNS1 in Fig. 8(a) replaces MNS1 and MNS2 in Fig. 1.
3.2 Charge Sharing Problems
In Fig. 8(a), with clk = 0, y2 is always high turning on
MN3 and causing charge sharing between n1 and Qb
and resulting in an incorrect value at Qb. It is effective
to introduce a transistor MNS2 driven by clk between
MP2 and MN3 [2]. Figure 8(b) shows the proposed
circuit, where MNS2 blocks charge sharing between Qb
and n2 . The simulation result for the circuit of Fig. 8(b)
(a) Toggle-flip-flop using the transistor merging technique
is shown in Fig. 9. The 2-GHz sinusoidal wave is applied
as the clock. An inverter is connected to the Qb node
to generate the Q signal. Under 1.8-GHz, the toggle
connection flip-flop in Fig. 8(b) ends up wrong operations due to long discharging time of y2 , and yields an
unpredictable state of y2 . While the structure in Fig. 4
fails to operate at 2-GHz, the structure in Fig. 8(b) can
operate at the frequency of 2-GHz. The reason is that
y2 in Fig. 8(b) starts discharging later than y2 in Fig. 4
since inserting MNS2 makes Qb discharge slowly.
Figure 10 shows the proposed D-flip-flop comprising nine transistors, which is free from glitches induced
by charge sharing. The MPS2 transistor driven by clk
as shown in Fig. 10 can effectively reduce charge sharing experienced in the circuit of Fig. 1. MPS2 disturbs
charge distribution path between y1 and y2 preventing MN2 from turning on. This guarantees the correct
edge-triggering operation of the flip-flop and enhances
its reliability. Unfortunately, the critical path to pull
up y1 node is longer and thereby some speed degradation is expected.
The operation of the proposed D-flip-flop shown in
Fig. 10 is as follows. Consider the circuit of Fig. 11(a),
where nodes y1 , n1 , and y2 are precharged high with
clk = 0 and D = 0. During this phase, MNS1 and
MP2 are off, and Qb holds the previous value. Note
that both n2 and n3 are weak high because of y1 and
Fig. 9
Simulation results for the proposed circuit of Fig. 8(b).
(b) Charge sharing protection between Qb and n1 using MNS2
Fig. 8 The proposed toggle-flip-flop designed with transistor
merging technique.
Fig. 10
The proposed D-flip-flop for glitch elimination.
IEICE TRANS. ELECTRON., VOL.E86–C, NO.3 MARCH 2003
500
(a) D = 0, clk = 0 → 1: Qb → 1
Fig. 12
Fig. 10.
Simulation results of the proposed D-flip-flop in
the initial charge of y2 node, Qinitial , is
Qinitial = Cy2 · V DD
(1)
The total charge after charge sharing, Qf inal , is
Qf inal = (Cy2 + Cn1 ) · Vf inal
(b) D = 1, clk = 0 → 1: Qb → 0
Fig. 11 Operations and signal paths of the proposed
D-flip-flop.
According to the charge conservation law, Qf inal =
Qinitial must be satisfied.
(Cy2 + Vf inal ) = Cy2 · V DD
(3)
Thus the final voltage of y2 , Vf inal , is
Vf inal =
y2 being high. Assuming that clk changes low to high,
MPS1 and MPS2 are turned off and MNS1 and MNS2
are on. Since y2 cannot discharge instantly, a pull-down
path is formed consisting of MNS2, MN3, and MNS1.
But n2 and n3 keep weak high from the previous phase
resulting in a small glitch due to the voltage drop of
Qb. As y2 becomes low through MN2 and MNS1 path,
Qb rises high.
Considering that clk = 0 and D = 1 as shown in
Fig. 11(b), y2 is precharged to high but y1 is low. This
makes MN2 be turned off. If clk changes low to high, Qb
discharges low through the path consisting of MNS2,
MN3, and MNS1. If we change D to low when clk =
1, MP1 is turned on, but the charge sharing between
y1 and y2 never occurs due to the blocking transistor
MPS2. This implies that MN2 remains off and the pulldown path of node y2 does not exist.
Though there exists the charge sharing effect between y2 and n1 , it is different from (c) and (d) of
Fig. 3. As shown in Fig. 12(b), after charge sharing y2
has a stable final value, since clk = 1 and MN2 is off.
On the other hand, the charge sharing in Fig. 3 turns
on MN2, and y2 continues to discharge.
The voltage drop of y2 resulting from charge sharing between n1 and y2 is determined by the ratio of the
total capacitance of n1 and y2 as follows. We assume
that y2 and n1 nodes are initially VDD and ground,
respectively. This is the worst case condition. Then
(2)
Cy2 · V DD
V DD
=
Cy2 + Cn1
1 + Cn1 /Cy2
(4)
The parameters of the employed process are as
follows: the junction capacitance Cj at zero bias
is 1 fF/µm2 , the junction capacitance of the sidewall Cjsw,p (Cjsw,n ) is 0.041 (0.083) fF/µm, Cjgate,p
(Cjgate,n ) is 0.587 (0.82) fF/µm, and the gate-oxide capacitance is 6.64 fF/µm2 . From the layout, the parasitic capacitances for interconnection lines are extracted
to the order of 10−1 –10−2 fF, which have a little effect
on the calculation of the parasitic capacitance. With
the source/drain area and their perimeters, Cn1 is calculated to 8.4 fF and Cy2 is calculated to 30.5 fF. Since
(Cn1 /Cy2 ) = 3.63, Vf inal is 1.96 V at VDD = 2.5 V,
which is lower than the threshold voltage of a PMOS
transistor of 0.6 V. If we lower the logic threshold voltage of the inverter consisting of MP2, MNS2, MN3, and
MNS1, this effect can be further alleviated. Note that
the final voltage of y2 after charge sharing, Vf inal , does
not change any more, which is the difference from (c)
and (d) of Fig. 3.
Simulation results in Fig. 12 show successful operations of the proposed design. Nine transistors and
one inverter (i.e., 11 transistors) are used to build the
same logic operation of Qb and Q of Fig. 7 with reliable
edge-triggering operations.
Figure 13 shows the structure and the simulation
result of Yuan/Svensson D-flip-flop [2], [11], which has
no charge sharing problem but has glitch problems [6].
YANG et al.: A NEW DYNAMIC D-FLIP-FLOP AIMING AT GLITCH AND CHARGE SHARING FREE
501
(a) Yuan/Svensson’s D-flip-flop
(a) Yuan/Svensson’s D-flip-flop
(b) Huang’s D-flip-flop
(b) Simulation result
Fig. 13
4.
Yuan/Svensson’s D-flip-flop and the simulation result.
Simulation and Experimental Results
To evaluate the performance of the proposed D-flipflop, a divide-by-16 asynchronous counter and a dualmodulus divide-by-128/129 prescaler have been designed. Huang’s general purpose D-flip-flops [6] of
Fig. 6 and Yuan/ Svensson D-flip-flops [2], [11] of Fig. 13
are used to build an asynchronous counter and a
prescaler for the performance comparison. The circuits
have been resized in 0.25-µm technology. Figure 14
and Fig. 15 show the transistor sizing and the layout for
the three D-flip-flops, respectively. The proposed flipflop occupies almost the same area as Yuan/Svensson’s
flip-flop and about 23% less area than Huang’s flipflop. Yuan/Svensson flip-flop operates at the highest
frequency among the three flip-flops. The proposed flipflop shows the lowest power consumption, the medium
speed, and the medium power-delay product (PDP).
The maximum operating frequency and the PDP of
the proposed circuit is slightly degraded from stacking three PMOS transistors, MPS1, MP1, and MPS2
as shown in Fig. 10. The lowest power consumption
is achieved by merging of the pull-up and pull-down
transistors. The proposed flip-flop in Fig. 10 can be
also obtained by connecting the sources of MP1 and
MN3 to the nodes y2 and n2 in Fig. 13(a), respectively.
Table 1 summarizes the comparison of the number of
transistors and the layout area for three D-flip-flops.
(c) The proposed D-flip-flop
Fig. 14 Transistor sizing for (a) Yuan/Svensson’s D-flip-flop,
(b) Huang’s D-flip-flop, and (c) the proposed D-flip-flop. For all
transistors, the channel length L is 0.25-µm.
4.1 Asynchronous Counter
Figure 16 shows the schematic diagram of a divide-by16 asynchronous counter. In each flip-flop, node Qb
is fed back to the node D to get a toggle operation
at the rising edge of the clock signal. The divide-by16 counter consists of four cascaded toggle flip-flops,
in which clk drives the first flip-flop and its output Q
drives the next flip-flop in turn. The output signal, the
divide-by-16 signal of the main clock, is acquired at the
node Q of the last flip-flop stage. The configuration
provides with a practical load conditions for high speed
operation: it employs the next stage flip-flops as its
load instead of the artificial load capacitance.
The maximum operating frequency and the power-
IEICE TRANS. ELECTRON., VOL.E86–C, NO.3 MARCH 2003
502
(a) Layout of Yuan/Svensson’s D-flip-flop (15.6 × 15.15 µm2 )
Fig. 17 Performance comparison of divide-by-16 asynchronous
counters.
Table 2 Transistor dimensions of the proposed D-flip-flop related to the scale factor, Sf . λ = 2.25 µm × Sf , and the channel
length is 0.25-µm for all transistors.
(b) Layout of Huang’s D-flip-flop (20.4 × 15.15 µm2 )
(c) Layout of the proposed D-flip-flop (15.75 × 15.0 µm2 )
Fig. 15
Table 1
Fig. 16
counter.
Layout of three dynamic D-flip-flops.
Comparison for three D-flip-flops.
Schematic diagram of a divide-by-16 asynchronous
delay product versus the supply voltage are evaluated
through the SPICE simulation. Figure 17 summarizes
the simulation results, where the power-delay products
are measured at the maximum frequency of Huang’s
circuit for an apparent comparison. At the supply voltage of 2.5-V, the maximum operating frequencies of the
proposed circuit, Huang’s circuit, and Yuan/Svensson’s
circuit are 4.06-GHz, 3.07-GHz, and 4.85-GHz, respectively. The power-delay products of them at the maximum frequency of Huang’s counter (i.e., at the frequency of 3.07-GHz) are 3.113-pJ, 4.284-pJ, and 2.744pJ, respectively. The performance of the proposed circuits is better than that of Huang’s circuits in the operating frequency and the power-delay product. At
the frequency of higher than 3-GHz, Huang’s flip-flop
in Fig. 6 suffers from the speed degradation since the
added inverter, INV1, fails to follow the input frequency.
To find out the effect of the supply voltage and
the device size on the divide-by-16 counter, simulations
have been performed scanning the supply voltage and
the device size. Table 2 shows the working transistor
dimensions of the proposed D-flip-flop, where the scale
factor Sf is used to change the transistor width. For
convenience, the parasitic capacitances of each transistor’s source/drain are calculated with the transistor width and the source/drain length without sharing
source/drain areas with the adjacent transistors.
As shown in Fig. 18, the supply voltage has been
considered from 1.0 to 3.5-V and Sf from 0.2 to 2.0.
The counter operates up to 4.06-GHz with Sf of 1.0.
As the supply voltage and Sf are increased, the maximum operating frequency is gradually increased and
finally saturated as shown in Fig. 18(a). The operating
frequency saturation at high Sf is due to the increase
of the parasitic capacitances. The power consumption
illustrated in Fig. 18(b) shows that the power consumption plane becomes steeper along the supply voltage
and the scale factor increase. The optimum values of
YANG et al.: A NEW DYNAMIC D-FLIP-FLOP AIMING AT GLITCH AND CHARGE SHARING FREE
503
(a) Maximum operating frequency versus VDD and scale factor
Fig. 20
Dual-modulus divide-by-128/129 prescaler.
(b) Power consumption versus VDD and scale factor
Fig. 18 Maximum operating frequency and power
consumption features.
Fig. 21
Performance comparison of divide-by-128 prescalers.
factor is 1.0, the rising time of the clock must be faster
than 0.9-ns. However, if we select the scale factor of
0.2, the rising time requirement becomes a slower value
of 2.1-ns. Figure 19 can be used to estimate the driving
capability of clock drivers for the proper operation of
flip-flops.
4.2 Dual-Modulus Prescaler
Fig. 19 Maximum allowable rising time of the clock according
to the scale factors.
the supply voltage and Sf can be deduced from (a)
and (b) of Fig. 18 with a specification on the maximum
operating frequency and the power consumption.
It is important to control the rise/fall rate at
the clock edge in dynamic flip-flops [2], [6], [11]. The
proposed flip-flop requires the clock rising time below
about 1 ns for the correct operation. For the rising time
slower than 1-ns, the precharge and the evaluation may
occur in the middle of a rising edge leading the flip-flop
to the wrong operation [6]. A smaller device size allows a slower edge rate of the clock signal. As shown in
Fig. 19, the maximum allowable rising time is decreased
as the device size increase. For example, when the scale
The block diagram of the dual-modulus divide-by128/129 (64/65) prescaler is shown in Fig. 20. The
dual-modulus prescaler has two counters: a divideby-4/5 synchronous counter and a divide-by-32 asynchronous counter. As the output signal of the divideby-4/5 counter, fso in Fig. 20, is connected to the clock
input of the divide-by-32 counter, the prescaler can divide the main clock signal by 128 or 129. The divideby-4/5 counter is the only part operating at the maximum frequency determining the overall speed of the
prescaler [5]. As several logic gates are inserted between the flip-flops to get a proper operation of the
prescaler, the speed degradation is unavoidable due to
the propagation delay.
The select signal chooses the division ratio of the
prescaler between 64/65 and 128/129. When the select is low, the last toggle flip-flop in the asynchronous
counter is bypassed and the prescaler operates as a
IEICE TRANS. ELECTRON., VOL.E86–C, NO.3 MARCH 2003
504
Fig. 22 The chip photograph of the dual-modulus prescaler.
The chip area including oscillator was measured to 85 µm×54µm.
erate at about 4-GHz with the supply voltage of 3.5 V.
A dual-modulus divide-by-128/129 prescaler has
been fabricated employing 0.25-µm CMOS technology.
The chip photograph is shown in Fig. 22. The area of
the chip including oscillator was 85-µm by 54-µm. Figure 23 shows the measured output from the fabricated
chip at the supply voltage of 2.5-V. In Fig. 22(a), the
measured output frequency of the prescaler was 22.62MHz (T = 44.2 ns), and the input clock frequency is
estimated to 2.895-GHz. The result is very close to the
simulation result. The output waveforms according to
the division ratios are also shown in Fig. 23(b).
5.
(a) Measured output waveform for the divide-by-128 prescaler
Conclusions
A new dynamic D-flip-flop for high-speed operation
and low-power consumption is presented aiming at
glitch free operation. The flip-flop consists of only
nine transistors based on the transistor merging technique, which is smaller than other alternatives. The
fewer transistors in the flip-flop can achieve faster operation and lower power consumption. To evaluate
the proposed flip-flop circuit, a dual-modulus divideby-128/129 prescaler has been designed and fabricated
using 0.25-µm CMOS technology. At the 2.5-V supply
voltage, the prescaler using the proposed dynamic Dflip-flops can operate up to the frequency of 2.95-GHz
consuming 3.621-mW, and shows half the power-delay
product of Huang’s circuits.
Acknowledgement
(b) Output waveform according to the division ratios
Fig. 23
Measured waveforms of the prescaler.
The authors would like to thank Hynix Semiconductor and Integrated Circuit Design Education Center
(IDEC) for the significant support to fabricate the test
chip as well as the reviewers for their valuable comments.
References
divide-by-64/65 counter. When the select is high,
the prescaler operates as a divide-by-128/129 counter.
When the mode signal is low (high), a divide-by-64 (65)
or 128 (129) prescaler is obtained.
The maximum operating frequency and the powerdelay product versus the supply voltage of the
prescalers are also measured. As shown in Fig. 21,
at 2.5-V supply voltage, the maximum operating frequencies in the proposed circuit, Huang’s circuit, and
Yuan/Svensson’s circuit are 2.95-GHz, 2.77-GHz, and
3.12-GHz, showing the power-delay products of 6.443pJ, 11.699-pJ, and 5.353-pJ, respectively. For comparison, the power-delay products are measured at the
maximum frequency of Huang’s circuit (i.e., at the frequency of 2.77-GHz). Each circuit operates at almost
the same frequency, since the speed of prescaler is limited by the propagation delay of the logic gates. Note
that the prescaler using the proposed flip-flops can op-
[1] H. Oguey and E. Vittoz, “CODYMOS frequency dividers
achieve low power consumption and high frequency,” Electron Lett., pp.386–387, Aug. 1973.
[2] J. Yuan and C. Svensson, “High-speed CMOS circuit technique,” IEEE J. Solid-State Circuits, vol.24, no.1, pp.62–70,
Feb. 1989.
[3] R. Rogenmoser, N. Felber, Q. Huang, and W. Fichtner,
“1.16 GHz dual-modulus 1.2-µm CMOS prescaler,” Proc.
IEEE 1993 CICC, pp.27.6.1–27.6.4, San Diego, CA, May
1993.
[4] Q. Huang and R. Rogenmoser, “A glitch-free single-phase
DFF for gigahertz applications,” Proc. IEEE 1994 ISCAS,
vol.4, pp.11–13, London, May 1994.
[5] R. Rogenmoser, Q. Huang, and F. Piaza, “1.57 GHz
asynchronous and 1.4 GHz dual-modulus 1.2-µm CMOS
prescaler,” Proc. IEEE 1994 CICC, pp.387–390, San Diego,
CA, May 1994.
[6] Q. Huang and R. Rogenmoser, “Speed optimization of edgetriggered CMOS circuits for gigahertz single-phase clocks,”
IEEE J. Solid-State Circuits, vol.31, pp.456–465, March
1996.
YANG et al.: A NEW DYNAMIC D-FLIP-FLOP AIMING AT GLITCH AND CHARGE SHARING FREE
505
[7] N. Foroudi and T.A. Kwasniewski, “CMOS high-speed
dual-modulus frequency divider for RF frequency synthesis,” IEEE J. Solid-State Circuits, vol.30, pp.93–100, Feb.
1995.
[8] B. Chang, J. Park, and W. Kim, “A 1.2 GHz CMOS dualmodulus prescaler using new dynamic D-type flip-flops,”
IEEE J. Solid-State Circuits, vol.31, pp.749–752, May 1996.
[9] C.-Y. Yang, G-.-K. Dehng, J.-M. Hsu, and S.-I. Liu, “New
dynamic flip-flops for high-speed dual-modulus prescaler,”
IEEE J. Solid-State Circuits, vol.33, pp.1568–1571, Oct.
1998.
[10] K.-H. Sung and L.-S. Kim, “Comments on new dynamic
flip-flops for high-speed dual-modulus prescaler,” IEEE J.
Solid-State Circuits, vol.35, pp.919–920, June 2000.
[11] M. Afghahi and C. Svensson, “A unified single-phase clocking scheme for VLSI systems,” IEEE J. Solid-State Circuits,
vol.25, pp.225–233, Feb. 1990.
[12] R.J. Baker, H.W. Li, and D.E. Boyce, CMOS Circuit Design, Layout, and Simulation, Ch.14, IEEE Press, New
York, 1998.
Sung-Hyun Yang
received the B.S.
and M.S. degrees in Computer and Communication Engineering from Chungbuk
National University, Cheongju, Korea in
1999 and 2001, respectively. He is currently pursuing the Ph.D. degree in Computer and Communication Engineering at
the Chungbuk National University, Korea.
His research interests are highspeed and low-power circuit, CMOS active pixel image sensor, analog-to-digital
conversion, and continuous-time filter designs.
Younggap You
received the B.S. degree in Electronic Engineering from the
Sogang Jesuit University, Seoul, Korea
and the M.S. and Ph.D. degrees in Electrical Engineering from the University of
Michigan, Ann Arbor, U.S.A., in 1981
and 1986, respectively. From 1975 to
1979, he was with the Agency for Defense Development, Korea, where he was
involved in high speed digital design. He
worked as a principal engineer at LG
Semiconductor, inc., Seoul, Korea, form 1986 to 1988. He is
currently a Professor in Dept. of Computer and Communication
Engineering at Chungbuk National University, Cheongju, Korea.
His research interests are fault tolerant computing, computer architecture, cryptography, cellular system design, and frequency
synthesis technology.
Kyoung-Rok Cho
received the B.S.
degree in Electronic Engineering from
Kyoungpook National University, Taegu,
Korea in 1977, and M.S. and Ph.D. degrees in Electrical Engineering from the
University of Tokyo, Tokyo, Japan, in
1989 and 1992, respectively. From 1979
to 1986, he was with TV research center of Gold Star Company in Korea. He
is currently a Professor in Dept. of Computer and Communication Eng. of Chungbuk National University, Korea, since August 1992. His research
interests are in the field of high-speed and low-power circuit design, and ASIC design for communication system. From 1999
to 2000, he was a visiting scholar at Oregon State University,
OR. He is a member of Institute of Electrical and Electronics
Engineer (IEEE), and Korea Institute Tele-communication Electronics (KITE).