Available via license: CC BY 4.0
Content may be subject to copyright.
Ternary and Quaternary CNTFET Full Adders are
less efficient than the Binary ones for
Carry-Propagate Adders
Daniel Etiemble
LISN
University Paris Saclay
Gif sur Yvette, France
de@lri.fr
Abstract—In Carry Propagate Adders, carry propagation is
the critical delay. The most efficient scheme is to generate Cout0
(Cin=0) and Cout1(Cin =1) and multiplex the correct output
according to Cin. For any radix, the carry output is always 0/1.
We present two versions of ternary adders with Cin=(0V, Vdd/2)
and Cin = (0V, Vdd ) and two versions of quaternary adders with
Cin=(0V, Vdd/3) and Cin = (0V, Vdd ). Using full swing Vdd for
Cin reduces the propagation delays for ternary and quaternary
adders. 6-bit, 4-trit and 3-quit CPAs are then compared.
Index Terms—Ternary adders, Quaternary adders, Binary
adders, Carry-Propagate Adders, CNTFET, propagation delays,
power dissipation, chip area.
I. INTRODUCTION
Carry Propagate Adders (CPAs) are the most simple N-digit
adders. Figure 1 presents a 4-digit CPA. Whatever digit radix
is used (2,3,4...), the carries are always 0/1. In this paper,
we consider both binary, ternary and quaternary CPAs. From
Figure 1, it results that the performance of a CPA is a direct
function of the used 1-digit full adder. More precisely, the
critical delay path of a CPA is related to the carry propagation.
Fig. 1: 4-digit Carry Propagate Adder
A. Carry values in a m-valued adder
The truth table of the ternary full adder is shown in Table
I. The truth table of the quaternary full adder is shown in
Table II. In both cases, the input and output carries are binary
(0,1). This property is valid for any radix. So, ternary 1-trit
full adders have ternary inputs and output (0,1,2) and binary
carries (0,1). Quaternary 1-digit adders have quaternary inputs
and outputs (0,1,2,3) and binary carries (0,1). There are two
techniques to get the different levels:
•Using two power supplies Vdd and Vdd/2 for ternary
circuits and three power supplies Vdd,2Vdd/3 and Vdd /3
for quaternary circuits.
•Using only one power supply (Vdd) and get the interme-
diate values through transistor connected as resistors. In
that case, there is a large static power dissipation resulting
from the direct current flow through the voltage divider
for intermediate levels. This is why we only consider the
option with two or three power supplies.
With this approach, all proposed designs use Vdd/2 as the
voltage when carry=1 for ternary circuits and Vdd/3 for qua-
ternary ones . This raises one question : must the binary
carries mandatory use 0 and 1 ternary/quaternary values or
can they use 0/2 ternary values or 0/3 quaternary values? In
other words, 0 and Vdd/2 or 0 and Vdd in the ternary case and
0 and Vdd/3 or 0 and Vdd when the ternary/quaternary adders
have a Vdd power supply. In this paper, we consider ternary
and quaternary circuits using the two different approaches for
carry levels and we compare them with binary adders. The
comparison is extended to CPAs computing approximately the
same amount of information: 6-bit (Binary Digit) CPAs, 4-trit
(Ternary Digit) CPAs and 3-quit (Quaternary Digit) CPAs. The
paper is organized as follow:
•we present the methodology
•we present the different ternary adders and their perfor-
mance
•we present the different quaternary adders and their
performance
•We present the different binary adders that are used for
comparison with their performance
•we compare the performance of the quaternary, ternary
and binary CPAs computing the same amount of infor-
mation
•we finally conclude.
arXiv:2207.04839v1 [cs.AR] 11 Jul 2022
TABLE I: Truth table of a ternary full adder
Cin=0 Cin =1
A B S0Cout0A B S1Cout1
0 0 0 0 0 0 1 0
0 1 1 0 0 1 2 0
0 2 2 0 0 2 0 1
1 0 1 0 1 0 2 0
1 1 2 0 1 1 0 1
1 2 0 1 1 2 1 1
2 0 2 0 2 0 0 1
2 1 0 1 2 1 1 1
2 2 1 1 2 2 2 1
TABLE II: Truth table of a quaternary adder
A B Cin S0Cout0A B Cin S0Cout1
0 0 0 0 0 0 0 1 1 0
0 1 0 1 0 0 1 1 2 0
0 2 0 2 0 0 2 1 3 0
0 3 0 3 0 0 3 1 0 1
1 0 0 1 0 1 0 1 2 0
1 1 0 2 0 1 1 1 3 0
1 2 0 3 0 1 2 1 0 1
1 3 0 0 1 1 3 1 1 1
2 0 0 2 0 2 0 1 3 0
2 1 0 3 0 2 1 1 0 1
2 2 0 0 1 2 2 1 1 1
2 3 0 1 1 2 3 1 2 1
3 0 0 3 0 3 0 1 0 1
3 1 0 0 1 3 1 1 1 1
3 2 0 1 1 3 2 1 2 1
3 3 0 2 1 3 3 1 3 1
TABLE III: TFAs Comparison
CNTFETs Technique
TFA / Year Count
In [1] 2011 412 Decoders-Binary-Encoder
In [2] 2017 105 Two custom algorithm + TMuxes
In [3] 2017 74 TMUXes
In [4] 2018 89 TMUXes
In [5] 2018 98 TBDD algorithm
In [6] 2019 142 Unary ops +MUXes+Encoder
In [7] 2020 74 Pass transistors + MUXes
In [8] 2020 106 Modified Quine-McCluskey algorithm
In [9] 2021 54 Unary ops + Decoders + Transmission gates
B. Related works
A lot of ternary full adders have been published in the last
decade [1]–[9]. They use different techniques. Transistor count
is not a sufficient criteria to determine the best TFAs. However,
considering Table III and a similar Table comparing Ternary
Half Adders in [10], we may consider that the technique using
unary operators and MUXes is the most efficient one. Several
quaternary full adders with CNTFET simulations have been
published in the last decade [11]–[13]. In a preprint paper
without simulations, several possible implementations based
on the transistor count have been evaluated [14]. This paper
considers CPAs.
C. Contributions
This paper compares efficient implementations of CNTFET
binary, ternary and quaternary adders. The main contributions
of this paper are
•For ternary and quaternary adders, we use both the carry
swing corresponding to 0 and 1 usual carry swing and
the full Vdd carry swing. Using full carry swing reduces
the carry propagation delays.
•While ternary and quaternary adders use a Vdd power
supply, we consider binary adders with both Vdd and
Vdd/2 power supplies. Reduced voltage swings for binary
adders drastically reduce power dissipation and the Power
Delay Product (PdP).
•We compare the performance of 6-bit CPAs, 4-trit CPAs
and 3-quit CPA that computes the same or approximately
the same amount of information.
II. METHODOLOGY
The significant figures to compare circuit designs include
switching times, power dissipation, chip area, etc. The com-
parison is realized by using HSpice simulations and evaluating
the chip area according to transistor sizes.
A. CNTFET technology
All simulations are done with the 32nm CNTFET parame-
ters of Stanford library [15]. We use CNTFET technology for
two main reasons:
•Simulations parameters for the most recent FinFET tech-
nologies are not available
•Most of papers presenting designs of ternary or qua-
ternary circuits in the last period use simulations with
this 32 nm CNTFET technology. This allows our results
to be compared with all published results on ternary or
quaternary circuits.
One advantage of CNTFET technology is that the threshold
levels of gates only depend on the diameter of individual
transistors, which facilitates the design of m-valued circuits.
B. Propagation delays
Generally, propagation delays are presented as an average
of the delays corresponding to all combinations of input
transitions. This presentation could be confusing. For the CPA
presented in Figure 1, Ai,Biand C0inputs are simultaneously
available. The important information is the propagation delay
corresponding to the critical paths, i.e. from C0(or A0/B0)
to C4and S3. When the 4-digit CPA is used to build larger
CPAs, the critical path is from C0to C4. We will only present
the propagation delays corresponding to the critical paths.
C. Power and Energy dissipation
Power and PDP (Power Delay Product) directly depends on
the duration of the input signals. It is important to use the
same input signal for all designs. For all simulations, we use
the input waveforms shown in Fig. 2. We have verified that the
delays for 0-2 or 2-0 ternary transitions are always less than
ternary transitions 0-1, 1-2, 2-1 or 1-0. The situation is similar
for quaternary transitions. We use these waveforms to compute
the worst-case delays from Input (A or B) to Sum/Cout and
from Cin to Sum/Cout.
2
(a) Ternary waveforms
(b) Quaternary waveforms
(c) Binary waveforms
Fig. 2: Input waveforms for all simulations
TABLE IV: Transistor diameters
n Diameter (nm) |V th|(V)
8 0.626 0.696
10 0.783 0.557
13 1.018 0.428
19 1.487 0.293
29 2.27 0.192
37 2.896 0.150
D. Chip area
Without drawing the layout of the circuits, there is no
technique to evaluate the chip area. We use a rough evaluation
of the chip area by summing the diameters of all the used
transistors by each circuit. This rough evaluation is a little
bit better than the transistor count. In this paper, we use the
diameter values presented in Table IV.
E. Circuit styles
Many techniques have been proposed to design full adders.
We only consider techniques with the following properties:
•No static power dissipation
•The circuit outputs have full swing. Reduced swings
degrade noise margins and can degrade the operation of
cascaded circuits, such as CPAs
•The circuits should have a sufficient driving capability.
This point is outlined in subsection II-G.
F. Temperature
All the simulations are done with a 25°C temperature. The
ternary adders (section IV), the quaternary one (section IV)
and one of the three binary ones (section V) use the same
circuit style. The same CNTFET parameters are used. There
are very few opportunities that different temperatures would
change the results of the comparisons between the different
adders.
G. Carry propagation in Carry Propagate Adders (CPAs)
As previously mentioned, carry propagation delay is the
critical one in CPAs. This delay must be minimized, either
for quaternary FAs or for ternary FAs or for binary FAs. One
technique is illustrated in Figure 3: Cout =Cout0when Cin=0
and Cout =Cout1when Cin=1. The correct Cout is obtained
via a multiplexer implemented with transmission gates. This
technique is used in many published binary full adders, such
as Transmission Gate Adders (TGAs) and other ones quoted
in [16]. This approach has a major drawback for CPAs. When
there is a direct propagation from the first to the last full
adders, there is a RC line effect (Figure 4) associated to the
capacitive loads that significantly degrade the carry delays. To
get a minimal delay without degrading the switching times,
the Cout signal must be restored by an inverter gate, as
shown in Figure 5. It means that Cout0and Cout1should be
computed, transmitted to Cout through the multiplexer and the
final inverter generates Cout. This technique will be used in
most circuits that we present.
Fig. 3: Cin to Cout carry propagation in a full adder
Fig. 4: RC effect with series of transmission gates
Fig. 5: Cin to Cout carry improved propagation with capacitive
loads in a full adder
III. THE TE RNA RY FUL L ADDERS
We consider two different ternary full adders based on the
MUX approach which common scheme is presented in Fig.
6. The difference between the first one (called TFA1) and
the second one (TFA2) are detailed after the presentation of
the MUX approach. For TFA1 and TFA2, two versions are
considered that differ by the carry voltage swing (0-Vdd/2)
and (0-Vdd).
3
Fig. 6: 1-trit Full Adder (Mux approach)
TABLE V: Functions A1and A2
A1A2
0 1 2
1 2 0
2 0 1
A. The MUX approach
According to Table I, When Cin=0
•When B=0 then Sum=A
•When B=1 then Sum = (A+1) mod(3) quoted as A1
•When B=2 then Sum = (A+2) mod(3) quoted as A2
•When B=0 then Carry=0
•When B=1 then Carry=1 when A= 2 else 0
•When B=2 then Carry=1 when A > 0else 0
When Cin=1
•When B=0 then Sum=A1
•When B=1 then Sum=A2
•When B=2 then Sum= A
•When B=0 then Carry=1 when A= 2 else 0
•When B=1 then Carry=1 when A > 0else 0
•When B=2 then Carry=1
Functions A1and A2are presented in Table V
B. The ternary full adders
TFA1 and TFA2 use the same threshold detectors (Fig. 7).
They implement the NI (Negative Inverter) and PI (Positive
Inverter) functions presented in Table VI. The operators A1
and A2are derived from the threshold detectors as shown in
Fig. 8. Anis the output of a negative inverter, Apis the output
of a positive inverter. Anb and Apb are the outputs of binary
inverters with inputs Anand Ap.
TFA1 and TFA2 differ by the implementation of the MUX
operators. TFA1 has a specific implementation of sum MUX
and carry MUX [17] in the ternary adder (Fig. 9). TFA2 uses
TABLE VI: NI and PI binary functions
NI PI
0 2 2
1 0 2
2 0 0
the MUX3 operators shown in Fig. 10. TFA1 and TFA2 use
the same MUX2 implementation.
There are few differences between 0.45V and 0.9V carry
versions. TFA1 uses a NTI inverter to get Cin and the final
carry inverter has a 0.45V power supply. For TFA2, Cin and
Cout use 0.9V inverters. TFA1 Σ(Di)= 72 nm (for carry
swing = Vdd/2 = 0.45V) and Σ(Di)= 73 nm (for carry swing
=Vdd= 0.9V). TFA2 Σ(Di)= 111 nm for carry swing = 0.45V
and Σ(Di)= 112 nm for carry swing = 0.9V.
Fig. 7: Threshold detectors
Fig. 8: A¹ and A² circuits
C. Performance with a 2 fF capacitive load
Fig. 11 presents the Input to Cout/Sum performance with
aCL= 2 fF capacitive load. Fig. 12 presents the Cin to
Cout/Sum performance with the same load.
The following remarks can be made
•TFA2 has a larger Σ(Di)(x1.5).
•There is a huge difference for Input to Cout/Sum. The
only difference comes from the implementation of ternary
MUXes. MUX3 implementation of TFA2 is more effi-
cient.
•There is little difference in Cin to Cout delay for TFA1
and TFA2. This is not surprising as both uses similar
MUX2 + Inverter designs for this propagation.
•TFA1 and TFA2 with 0.9V carry swing have Cin to Cout
delay roughly two times faster than TFA1 and TFA2 with
0.45V carry swing. The 0.9V inverters have more driving
capability than the 0.45V inverters.
While TFA2 has 50% more Σ(Di), the huge difference in
Input to Sum delay for the last stage of a CPA makes TFA2
the best ternary adder either with 0.45V or 0.9V carry swing.
D. Delays and power according to capacitive load
With a log-log scale (except for CL= 0 fF), Fig. 13 presents
the input to outputs delays according to CL. Fig. 14 presents
4
Fig. 9: Specific implementation of ternary MUXes for Ternary Full Adder [17]
Fig. 10: 3-input MUX with ternary control
the same information for Cin to outputs delay while Fig. 15
present the evolution of power according to CL. Considering
the different curves between CL= 0.25fF and CL= 4fF, we
may observe that the delay evolution are close to a linear
one, with different slopes. Power increases more than linearly
according to CL.
Cin to Cout path is through a multiplexer and an inverter
while Cin to Sum is just through a multiplexer. The inverter
restores the signal and has more driving capability than the
multiplexer. It explains why the sum delay is more sensitive
to capacitive load. Input to Cout and Sum paths include the
whole circuit. The final inverter delay for Cout has a limited
impact on the overall delay compared to Sum delay, which
explain why these large delays don’t increase much when CL
is multiplied by 16. Power increases from x2 to x3.
Fig. 16 presents the ratio delays(CL= 4fF)/delays(0.25fF)
when CLis multiplied by 16. It is a figure of the slope of the
quasi-linear evolution of delays(CL). For Cin to output delays,
the sum output is more sensitive to CLthan Cout. It comes
from the Sum MUX output that has less driving capability than
the Cout inverter. Due to the large delays from input to outputs,
the impact of CLis limited for these delays. Fig.17 presents
the power evolution when CLis multiplied by 16. The impact
is slightly more important for Cin to Outputs than for Input
to Outputs as it concerns only MUXes and the final inverter.
The Vdd inverter consumes more than the Vdd/2 inverter.
IV. QUATERNA RY FUL L ADDERS
The common scheme is presented in Fig. 18. The two QFAs
only differ by the carry swing. Carry input values are Vdd/3
(QFA1) and Vdd (QFA2). The control of the two MUX2 is
shown is Fig. 18. The carry output is obtained by inverters
with Vdd/3supply (QFA1) or Vdd (QFA2)
A. Multiplexer Implementation
The common functional scheme is shown in Fig. 18. The
threshold detectors (Fig. 19), the circuits A¹, A², A³ (Fig. 20)
5
Fig. 11: Input to Cout/Sum performance of ternary adders
Fig. 12: Cin to Cout/Sum performance of ternary adders
Fig. 13: TFA-Input to Cout/Sum delays according to CL
and the MUX4 (Fig. 21) are similar to those of [3]. The
two final multiplexers are typical binary multiplexers. Cout
Fig. 14: TFA-Cin to Cout/Sum delays according to CL
is computed from Cout0and Cout1. A final inverter delivers
Cout. 4-input multiplexers with quaternary control are used
6
Fig. 15: TFA-Power dissipation according to CL
Fig. 16: TFA-Delay ratio when CLis multiplied by 16
(Fig. 21). The three inverters with outputs Bnbb,Bibb and
Bpbb operate as buffers because inverters Bnand Bphave
poor driving capability. Paper [3] first uses a quaternary half
adder (sum and carry circuits). A second stage computes the
final result by adding +1 mod(4) to sum when Cin=1 and
computing Cout according to Cin. We directly computes Sum
and Cout within a single stage. Cin to Cout propagation delay
is reduced to a MUX2 and final inverter path.
Fig. 17: TFA-Power ratio when CLis multiplied by 16
Fig. 18: Quaternary Full Adder (MUX approach)
Fig. 19: Threshold detectors
Fig. 20: A¹, A² and A³ circuits
Fig. 21: 4-input MUX with quaternary control
7
Fig. 22: Input to Cout/Sum performance for QFA1 and QFA2
Fig. 23: Cin to Cout/Sum performance for QFA1 and QFA2
B. Performance with a 2 fF capacitive load
For all simulations, the same input waveforms are used. Ex-
tensive simulations have determined that 0→1→2→3→2→1
→0 for input A with Cin=0 lead to the input to Cout /Sum
worst case delays. Similarly, 0→1 (QFA1)/3 (QFA2) →0 with
A=2 and B=1 lead to the Cin to Cout/Sum worst case delays.
These configurations are used to evaluate the performance of
QFA1 and QFA2. The only difference is the amplitude of the
carry swing. The performance results are presented in Fig. 22
and 23. These figures provide the data and allows a direct
comparison for for each feature. The significant information
is Input to Cout (first adder of a CPA), Cin to Cout (following
adders) and Cin to Sum (last adder of a CPA).
QFA1 and QFA2 have simular ΣDi. QFA1 has a small
advantage in term of power. However, it is outperformed by
QFA2 for Cin to Cout delay, which is the critical delay for a
CPA. The situation is the same for PDP. This big advantage
comes from the last carry inverter that performs better with a
Vdd supply than with a Vdd/3supply.
C. Delays and power according to capacitive load
We now present the performance of QFA1 (0.3V carry
swing) and QFA2 (0.9V carry swing).
With a log-log scale, Fig. 24 presents the input to outputs
delays according to CL. Fig. 25 presents the same information
for Cin to outputs delays while Fig. 26 presents the evolution
of power according to CL. Fig. 27 presents the ratio delays(CL
= 4fF)/delays(0.25fF) when CLis multiplied by 16. It is a
figure of the slope of the quasi-linear evolution of delays(CL).
Fig.17 presents the power evolution when CLis multiplied
by 16. The ternary adder and the quaternary adder having
the same basic circuit structure, it is not surprising that the
conclusions detailed in III-D are also valid for the quaternary
adder.
8
Fig. 24: QFA-Input to Cout/Sum delays according to CL
Fig. 25: QFA-Cin to Cout/Sum delays according to CL
V. TH E BINARY FULL ADDERS
A. Presentation
For the comparison with ternary and quaternary full adders,
we consider three different binary adders:
•The first one is a 14T Full Adder (Fig. 31)
•The second one is the typical 28T full adder (Fig.29)
•The third one is a MUX-based full adder (Fig. 30)
that uses the same circuit style than the ternary and
Fig. 26: QFA-Power dissipation according to CL
Fig. 27: QFA-Delay ratio when CLis multiplied by 16
Fig. 28: QFA-Power ratio when CLis multiplied by 16
quaternary adders. Using the same circuit style allows
a fair comparison.
The three binary full adders operate with the same Vdd =
0.9V as the quaternary adder. They can also operate with a
0.45V supply, which roughly divide by 4 the dynamic power
dissipation. Vdd = 0.45V is a too small power supply value to
operate with the three levels of a ternary adder or four levels
of a quaternary adder.
Fig. 29: 28T Binary Full Adder - BFA2
9
Fig. 30: Binary Full Adder-MUX approach (BFA3)
Fig. 31: 14T Binary Full Adder - BFA1
Fig. 32: Binary Adders - Input to Cout/Sum - CL= 2 fF
Fig. 33: Binary Adders - Input to Cout/Sum - CL= 2 fF
B. Performance with a 2 fF capacitive load
Fig. 32 presents the Input to Cout/Sum performance with
CL= 2 fF. Fig. 33 presents the Cin to Cout/Sum performance
with the same capacitive load. While the MUX-approach
(BFA3) is the best approach for ternary and quaternary adders,
it is the worst one for binary adder in terms of delays, power
and ΣDi. All powers for 0.45 Vdd are roughly 1/4 of the
powers of 0.9 Vdd versions, leading to PDD slightly smaller
or equivalent for both Vdd. For input to Cout performance
corresponding to the worst case of the first BFA in a CPA,
BFA2 is better than BFA1. However Cin to Cout BFA1 delay,
which is the critical delay in a CPA, is about 3x smaller than
BFA2 delay both for 0.9V and 0.45V Vdd. The Cin to Sum
delays, which is critical for the last stage of a CPA, are close
for BFA1 and BFA2 (Vdd = 9V) and x2.25 greater for BFA1
(Vdd = 0.45V), but this is conterbalanced by the x3 smaller
Cin to Cout delay for TFA1.
BFA1 is globally the most efficient binary adder in terms
of delays, PDP and ΣDi for the two different power supplies.
C. Delays and power according to capacitive load
We now present the performance of BFA1 according to
capacitive loads and temperature. With a log-log scale, Fig.
34 presents the input to outputs delays according to CL. Fig.
35 presents the same information for Cin to outputs delays
while Fig. 36 presents the evolution of power according to
CL. Fig. 37 presents the ratio delays(CL= 4fF)/delays(0.25fF)
when CLis multiplied by 16. It is a figure of the slope
of the quasi-linear evolution of delays(CL). Fig.38 presents
the power evolution when CLis multiplied by 16. We still
have a quasi linear evolution of delay and power according
to CL. However, the binary adder structure is different of the
m-valued adder structures: there is one MUX for Cout, but
not a series of MUXes as in the Sum output of ternary and
quaternary adders. Globally, the binary adder is more sensitive
to capacitive loads than the ternary and quaternary ones.
Fig. 34: BFA1-Input to Cout/Sum delays according to CL
VI. COMPARING 6-BIT,4-TRIT AND 3-QDIGIT CPAS
Results provided in III-D, IV-C and V-C allow a detailed
comparison of the performance of the different adders to
10
Fig. 35: BFA1-Cin to Cout/Sum delays according to CL
Fig. 36: BFA1-Power dissipation according to CL
be used in a Carry-Propagate Adder. The most significant
information is to compare CPAs computing the same amount
of information. It is strictly the case for 6-bit and 3 quit CPAs.
4-trit input corresponds to 6.34 bits, which corresponds about
to 6% more information than 6-bit or 3 quit.
Several 4-trit CPAs have been presented in the literature [3],
[10], [?] and [17].
Fig. 39 compares the performance of these three CPAs with
two variants: the ternary one uses 0-Vdd/2 and 0-Vdd carry
swing, the quaternary one uses 0-Vdd/3 and 0-Vdd carry swing
Fig. 37: BFA-Delay ratio when CLis multiplied by 16
Fig. 38: BFA-Power ratio when CLis multiplied by 16
and the binary one uses Vdd and Vdd/2 power supplies. The
simulation have been done with a CL= 2 fF capacitive load
and T = 25°C temperature. Other loads or temperatures would
not change the results of the comparisons. From Fig. 39, the
following conclusions can be deduced:
•While the binary CPA uses more full adders, its estimated
chip area is half the chip area of the ternary and quater-
nary CPAs.
•The ternary and quaternary CPAs have less propagation
delays when using full carry swing than when using Vdd/2
or Vdd/3 carry swing
•The 0.45 Vdd binary CPAs has the smallest power dis-
sipation, from 1/2 to 1/4 power dissipation of the other
CPAs. While its input to sum delay is the worst one, this
CPA has the lowest PDP both for sum and carry outputs.
•The quaternary CPA has a small advantage for delays
with full carry swing, but the values are closed.
While ternary and quaternary CPAs have less full adders, they
suffer from the large chip area and don’t provide significant
advantages in term of delays. The best CPA is the binary
one with Vdd = 0.45V supply. Reducing power supply is
possible with binary circuits, but is not possible with ternary
and quaternary circuits as they need a larger Vdd to handle the
different voltage levels.
VII. CONCLUDING REMARKS
We have detailed the performance of binary, ternary and
quaternary full adders that are probably close to the most
performant ones. We have shown that two options are possible
for these adders.
•For ternary and quaternary adders, we used two carry
swings. The first one correspond to 0-1 logical values,
i.e. 0-Vdd/2 for the ternary adder and 0-Vdd /3 for the
quaternary adder. The second one uses the 0-Vdd carry
swing for both adders, as carry values are always 0-
1 logical values for any radix used for addition. It
turns out that full carry swing reduces significantly carry
propagation delays with a small power increase.
•For binary adders, we use both 0.9V and 0.45V power
supplies. The smallest Vdd value reduces significantly
11
Fig. 39: Comparing 6-bit, 4-trit and 3-qdigit CPAs with CL= 2 fF
power (/4 factor), which leads to reduced PDP with a
small increase in delays
The different adders are used in CPAs computing the same
amount of information. In CPAs, carry propagation is the
critical delay. The critical delay paths are similar for the
ternary and quaternary adders. For the binary adder, it consists
in a NOR gate and a MUX. With 6, 4 and 3 adders in
the binary, ternary and quaternary cases, the ternary and
quaternary adders should benefit from the reduced number of
carry paths. It turns out that this is not the case as input to
carry delays are close (they are not in the ratio 6/4/3). Input
to Sum delays are also close with 0.9V Vdd (binary) and 0.9V
swing (ternary and quaternary).
The only figure for which 3-quit CPAs shows a small
advantage is input to output delays with 0.9V carry swing.
The ternary and quaternary CPAs are outperformed by the
binary CPA with 0.45V supply in terms of power and PDP.
CPAs are circuits for which moving from binary to ternary
or quaternary N-digit CPAs is simple: just replace the binary
full adders by ternary or quaternary full adders. Moving from
a N*N digit binary multiplier to a N*N digit ternary or qua-
ternary multiplier is not so simple. Combinational multipliers
using Wallace tree reduction circuits (or equivalent ones) uses
both 1-digit multipliers and adders. Ternary and quaternary
multiplications generates both product and carry values while
binary multiplication (AND gate) only generates 1 bit product.
REFERENCES
[1] S. Lin, Y.-B. Kim, and F. Lombardi, ‘CNTFET-Based Design of Ternary
Logic Gates and Arithmetic Circuits’, IEEE Trans. Nanotechnology, vol.
10, no. 2, pp. 217–225, Mar. 2011, doi: 10.1109/TNANO.2009.2036845.
[2] B. Srinivasu and K. Sridharan, ‘A Synthesis Methodology for Ternary
Logic Circuits in Emerging Device Technologies’, IEEE Trans. Cir-
cuits Syst. I, vol. 64, no. 8, pp. 2146–2159, Aug. 2017, doi:
10.1109/TCSI.2017.2686446.
[3] S. Tabrizchi, A. Panahi, F. Sharifi, K. Navi, and N. Bagherzadeh, “Method
for designing ternary adder cells based on CNFETs”, IET Circuits,
Devices & Systems, vol. 11, no. 5, pp. 465–470, Sep. 2017, doi:
10.1049/iet-cds.2016.0443.
[4] E. Shahrom and S. A. Hosseini, ‘A new low power multiplexer based
ternary multiplier using CNTFETs’, AEU - International Journal of
Electronics and Communications, vol. 93, pp. 191–207, Sep. 2018, doi:
10.1016/j.aeue.2018.06.011.
[5] C. Vudadha, A. Surya, S. Agrawal, and M. B. Srinivas, ‘Synthe-
sis of Ternary Logic Circuits Using 2:1 Multiplexers’, IEEE Trans.
Circuits Syst. I, vol. 65, no. 12, pp. 4313–4325, Dec. 2018, doi:
10.1109/TCSI.2018.2838258.
[6] T. Sharma and L. Kumre, ‘CNTFET-Based Design of Ternary Arithmetic
Modules’, Circuits Syst Signal Process, vol. 38, no. 10, pp. 4640–4666,
Oct. 2019, doi: 10.1007/s00034-019-01070-9
[7] I. Mahmoudi Salehabad, K. Navi, and M. Hosseinzadeh, ‘Two novel
inverter-based ternary full adder cells using CNFETs for energy-efficient
applications’, International Journal of Electronics, vol. 107, no. 1, pp.
82–98, Jan. 2020, doi: 10.1080/00207217.2019.1636306.
[8] S. Kim, S. -Y. Lee, S. Park, K. R. Kim and S. Kang, ”A Logic
Synthesis Methodology for Low-Power Ternary Logic Circuits,” in IEEE
Transactions on Circuits and Systems I: Regular Papers, vol. 67, no. 9,
pp. 3138-3151, Sept. 2020, doi: 10.1109/TCSI.2020.2990748.
[9] S. A. Hosseini and S. Etezadi, ‘A Novel Low-Complexity and Energy-
Efficient Ternary Full Adder in Nanoelectronics’, Circuits Syst Signal
Process, vol. 40, no. 3, pp. 1314–1332, Mar. 2021, doi: 10.1007/s00034-
020-01519-2.
[10] R. A. Jaber, B. Owaidat, A. Kassem and A. M. Haidar, ”A Novel
Low-Energy CNTFET-Based Ternary Half-Adder Design using Unary
Operators,” 2020 International Conference on Innovation and Intelligence
for Informatics, Computing and Technologies (3ICT), 2020, pp. 1-6, doi:
10.1109/3ICT51146.2020.9311953.
[11] S.A. Ebrahimi,M.R. Reshadinezhad, A. Bohlooli, M. Shahsavari, “Ef-
ficient CNTFET-based design of quaternary logic gates and arithmetic
circuits”, Microelectronics Journal, pp 156-166, January 2016
[12] M.H. Moaiyeri, K. Navi, O. Hashemipour, “Design and Evaluation of
12
CNFET-Based Quaternary Circuits”, Circuits Syst Signal Process (2012)
31:1631–1652, DOI 10.1007/s00034-012-9413-2
[13] E. Roosta and S. A. Hosseiny, “A Novel Multiplexer-Based Quaternary
Full Adder in Nanoelectronics”, Circuits, Systems and Signal Processing,
https://doi.org/10.1007/s00034-019-01039-8
[14] D. Etiemble, ”Best implementations of Quaternary Adders”, https://
arXiv:2005.02206
[15] J. Deng and H. -. P. Wong, ”A Compact SPICE Model for Carbon-
Nanotube Field-Effect Transistors Including Nonidealities and Its Ap-
plication—Part II: Full Device Model and Circuit Performance Bench-
marking,” in IEEE Transactions on Electron Devices, vol. 54, no. 12, pp.
3195-3205, Dec. 2007. https://doi.org/10.1109/TED.2007.909043.
[16] S. Wairya, R. K. Nagaria, S. Tiwari, ”Performance Analysis of High
Speed Hybrid CMOS Full Adder Circuits for Low Voltage VLSI De-
sign”, VLSI Design, vol. 2012, Article ID 173079, 18 pages, 2012.
https://doi.org/10.1155/2012/173079
[17] R.A. Jaber, “Two Improved Designs for Ternary Full Adders Using
Unary Operators and Ternary Multiplexers”, Personal communication
13