a
x
b
, respectively.
Denition 2.5. The y-cyclic clock-wise, (y-CWC),
and its counter, (y-CCWC), operations are dened as
y-CWC: x
y
= x + y (mod r) (7a)
y-CCWC: x
y
= x y (mod r) (7b)
From (7a), it can be shown that
x
y
= x
ry
(7c)
Denition 2.6. A k-valued product term is given by
P
k
:
_
i {1,2,...,m}
g(x
i
) = k (8)
where refers to min operator over unary operations,
g(x
i
), such as L, CL, CW and CCW [7], i.e.,
P
k
:
_
k
1
a1
x
b1
1
_
_
k
2
a2
x
b2
2
_
x
a3
3
. . . = k (9)
In this study, product terms are assumed to include only
literals and/or complementary literals as unary opera-
tions. However, cyclic and its derivatives are generally
used to realize symmetric functions.
Denition 2.7. A multi-valued function can be ex-
pressed in terms of the product terms as, [17].
f (x
1
, x
2
, . . . , x
m
) =
_
j R
P
j
(10)
where implies the max operator.
Implementation of Multi-Valued Logic Gates 193
3. Basic Circuit Elements
3.1. Primitives
(1) Sum: Basically, a sum is a node in the circuit,
where some currents enter and/or some currents
leave.
(2) Constant: Logic levels represent constant cur-
rents in CM-MVL. They can be generated
with a reference circuit using either enhance-
ment mode p-type or n-type transistors, depend-
ing on the current sourcing or sinking action
desired.
(3) Current-Mirror: They are used for generating sin-
gle or multiple-replicas of a current. An n-type cur-
rent mirror and its symbol are shown in Fig. 1
where N
d
stands for diode-connected transistor
while N
m,1,..n
for mirror transistor.
A current can be redirected by cascading n and
p type mirrors, as shown in Fig. 2 where z
1
, . . . , z
n
can be any multiple of x.
(4) Input/Output Circuit: Adiode-connected transistor
with aspect ratio of W/L = 1.5/1 is used as the
input and output circuit. This aspect ratio refers
to the minimum feature size to t the technology
used.
Fig. 1. The n-type current mirror and its symbol.
Fig. 2. Multiplying and redirecting a current.
3.2. Secondary Blocks
Inthis study, gate circuits are formedbyusingtruncated
difference operation [10], dened by
x ky =
_
x ky iff x ky
0 otherwise
(11)
It can be shown that
min(x, y) = x(xy) = y(yx) (12a)
max(x, y) = min( x, y) = x +(yx) (12b)
The truncated difference can be realized as shown in
Fig. 3.
Based on the truncated difference, current-mode
threshold operations can also be realized. The upper
and lower threshold operations for a, b, c R are de-
ned as
upper-threshold, th
u
: a
c
b
=
_
c if a b
0 otherwise
(13a)
lower-threshold, th
l
:
c
b
a =
_
c if a b
0 otherwise
(13b)
The th
u
operation can be realized as shown in Fig. 4
[12]. A similar topology is proposed in [9]. However,
the topologyproposedinthis studyprovides full control
over output current. The lower threshold operation, th
l
,
can be realized with a and b interchanged in the circuit
of th
u
.
The transition region, I = 2.5 A,can be adjusted
by changing the W/L ratio of the transistor N4. The
transistors N5, N6 and N7 should be also adjusted ac-
cordingly to satisfy s3 (1.5)
= a 0.5, b
= b +0.5
and r
= (x + y)(x + y)
r
r
(14)
Using the denition of r
above, y-CWCoperation
can be realized as shown in Fig. 9.
Fig. 8. Generating auxiliary parameters for literal gates.
The y-cyclic counter clock-wise, y-CCWC, operation
can easily be realized by replacing y with r y. The re-
peated quantities can be reproduced with proper mirror
structures.
Of above gates, the layouts of min and max gates are
given in Appendix B, while the layout of cyclic gate is
given in Appendix C associated with an application.
5. Simulation of Basic Gates
Designed gates with indicated transistor feature sizes,
Figs. 59, are simulated with input waveforms shown
in Fig. 10. Gate simulation results are given in
Figs. 1114.
6. Comparison with Previous Designs
Due to lack of a proper current-mode switch topol-
ogy, most Postian MVL, [17] studies prefer exploiting
196 Temel and Morgul
Fig. 9. (a) y-CWC operation DC characteristics, (b) block diagram, (c) circuit diagram and (d) minimum feature size transistor aspect ratios.
Fig. 10. Input waveforms for simulations.
0 125 250 375 500 625 750 875 1000
0
20
40
60
80
time, nsec.
o
u
t
p
u
t
,
u
A
min(x,y)
Fig. 11. Simulation results of min(x, y) gate.
the voltage mode-binary logic circuits [1, 7], shown in
Fig. 15, which can be named as hybrid or switched-
current mode approach. Although hybrid-mode MVL
design is studied in [18] with regard to DC behavior,
no condition is put forward on optimumoperating size.
In Fig. 15, N
i
is the input current mirror, x
is
the reected input current, I
o
is the desired output
0 125 250 375 500 625 750 875 1000
0
20
40
60
80
time, nsec.
o
u
t
p
u
t
,
u
A
max(x,y)
Fig. 12. Simulation results of max(x, y) gate.
Fig. 13. Simulation results of: (a) 5
3
x
5
and (b) 3
3
x
5
.
Implementation of Multi-Valued Logic Gates 197
0 125 250 375 500 625 750 875 1000
0
20
40
60
80
time, nsec.
o
u
t
p
u
t
,
u
A
x
y
Fig. 14. Simulation results of y-CWC gate.
Fig. 15. General structure of switchedcurrent-mode, MVLdesigns.
current, M
o
is the input transistor of the following
gate, M
sw
is the NMOS or PMOS switching transis-
tor, M
k
is the current source supplying the current I
o
.
The binary logic level signal at node D is a function of
voltage at node C and it is obtained with conventional
voltage mode logic gates. For proper design, neglect-
ing the required scaling within the voltage-mode block,
the output block, which may be the input block of the
following gate, has to satisfy certain sizing conditions.
M
sw
conducts if VDD V
gs,o
V
th,sw
, where sub-
scripts o and sw denote the transistors M
o
and
M
sw
, respectively. Assuming long-channel transistors
are in use, for a given output transistor size, the upper
bound for output current is given by
I
o
<
[VDD (VTO
o
+ V
th,sw
)]
2
KP
o
2
_
L
W
_
o
(15)
where VTO
o
, is the zero-bias threshold voltage of M
o
and V
th,sw
is the threshold voltage of M
sw
and KP is the
process transconductance of the transistor considered.
The upper bound of output current, or the achievable
radix is determined by M
sw
, which has to operate in
triode-mode. Therefore, it can be shown that
_
W
L
_
sw
2I
o
_
VDD
_
2I
o
(W/L)
o
KP
o
(VTO
o
+ V
th,sw
)
_
2
KP
sw
(16)
As can be seen from Eq. (16), dimensions of M
sw
is lower bounded and increases severely as the out-
put current approaches the desired output current given
by Eq. (15). Moreover, the situation becomes worse
with body-bias effect. The fact, that M
k
has to operate
in saturation as a current source, puts forward larger
(W/L)
sw
values. Hence, the output block area, which
is directly proportional to
o
and
sw
, becomes very
large, where = KP(W/L) is the transistor conduc-
tance.
Aminimumarea of the output block is achieved with
a resized M
o
by a constant factor (1 + ) for a given
radix, with being a scale parameter. Therefore, input
block of the succeeding gate has to be scaled up with
the same factor, which degrades the performance of
its input current comparator. It can be shown that =
3.1 if M
sw
is not body-biased and independent of the
parameters used. If M
sw
is body-biased, = 1.8 is
found, depending on the parameters used. With scaled-
up output transistor, (W/L)
sw
can be calculated from
_
W
L
_
sw
KP
o
(W/L)
o
(1 +)
_
1
_
1
1+
_
2
KP
sw
(17)
Therefore, there exists an asymmetry between input
and output blocks. Due to asymmetric input and output
stages, each gate, and particularly voltage-mode block,
would have to operate with larger capacitive loads. The
output capacitance driven by the voltage-mode block
is further increases due to cascode structure formed
by M
sw
and M
o
. If operation is performed with body-
biased M
sw
, it can be shown that there is a two bounded
operation range, which may not allow for higher-radix
operation.
In order to illustrate radix-limitation, the literal
(5
3
x
5
) has been designed using the circuit given in
[7] for both body-biased and no body-biased M
sw
, in
which input quantities sink while the output sources.
Using the MOS model in [19], transistor dimen-
sions can be estimated for short-channel, small devices.
198 Temel and Morgul
Assuming similar parameter descriptions and linear re-
lationship between V
GS
and I
D
of a short-channel MOS
transistor, it can be shown that
I
o
z
o
[VDD (V
th,o
+ V
th,sw
)] (18)
and switch transistor dimension is given by
_
W
L
_
sw,eff
I
o
K
sw
v
sat,sw
C
ox,sw
L
sw,eff
[VDD (V
th,o
+ V
th,sw
)] z
o
I
o
(19)
In Eqs. (18) and (19), z = Kv
sat
C
ox
W
eff
, (A/V)
with K <1 being a constant accounting for the short-
channel effects for the transistor considered, [19]. For a
saturated transistor, it is simply taken as 0.75. The pa-
rameter v
sat
is the saturation velocity, which is 8
10
4
m/sec and 6.5 10
4
m/sec for electrons and
holes, respectively. W
eff
=W 2WD, L
eff
=L 2LD
where WD and LD are the spice parameters indi-
cating the overlap region width. Solving Eqs. (15)
and (16) with both body-biased and no body-
biased M
sw
shows that, for I
o
= 50 A it is
necessary to set (W/L)
sw,nobody-bias
= 20.5 and
(W/L)
sw, body-bias
= 454 if (W/L)
o
= 1.5. If
(W/L)
o
= 5, (W/L)
sw,body-bias
= 13 satises the equa-
tions.
In Fig. 16, the simulation results show that it is not
possible to obtain the required 50 A current if the
output load transistors W/L ratio, (W/L)
O
, is less
than 5 for body biased transistor and 2.5 for no body
biased transistor. In the gure, > implies no varia-
tion is observed despite further increases in considered
transistors dimensions. Table 1 summarizes HSPice
simulation results of designs given in this study and
[7], respectively.
The DC performance of proposed gates can be stud-
ied with a typical circuit in Fig. 17, where M
m
repre-
sents the output transistor of the gate andM
O
represents
the input circuit of the next gate.
It can be shown that the maximum current range for
which the circuit with long-channel transistors func-
tions linearly may be given by
I
o
= x iff x
_
o
2
__
VDD VTO
o
1 +
o
/
m
_
2
(20)
Table 1. Comparison of simulation results of proposed circuits in
this study (New), and switched current-mode designs with body-bias
(no body-bias) in Ref. [7], for minimum feature size, i.e., (W/L) =
1.5/1.
Trans. Av. del. Av. pow.
Gate count (ns) diss. (mW)
Min New 8 1.6 0.25
Ref. [7] 8 1.6 0.25
Max New 7 1.4 0.23
Ref. [7] NA NA NA
CWC, x
3
_
VDD VTO
o
1/z
d
+1/z
o
_
(21)
Equations (20) and (21) are investigated for (W/L)
o
=
1.5 and (W/L)
d
= (W/L)
m
= 5.5, i.e., = 1,
Implementation of Multi-Valued Logic Gates 199
Fig. 17. The basic circuit to determine DC characteristics of
designed gates.
yielding maximum current values as x
,max
= 73 and
86 A respectively.
7. Analog Design Constraints
on Operating Radix
Linear current range describedinthe previous sectionis
not the only parameter for choosing the operating radix.
The step size between logic levels, which is limited by
the noise margin, must also be considered.
The output current of the mirror circuit in Fig. 17
is constituted with random and nominal components.
The former is attributed to random design parameter
variations, and physical noise current generated in the
devices [20, 21]. Denoting the output current variation
by x
| |3
x
| where
x
is the standard deviation
of x
x + x
7y
3
(23)
=
_
y if 0 x 3
7 y otherwise
Due to disjoint product terms, the max operator is re-
placed with ordinary sum circuit.
As stated previously, most symmetric functions can
be designed using cyclic gates. A useful application
would be a higher radix full-adder. Using Eq. (7)
and its corresponding circuit, a higher-radix full-adder
can be designed as shown in Fig. 20 where (W/L)
200 Temel and Morgul
Fig. 19. Cascaded stage behaviours of min and max gates.
Fig. 20. A higher-radix current-mode full adder circuit.
of M9 is 1.5/1. Taking normalized quantities, the
sum, S, and the carry output, C, can be obtained as
follows:
S = x
y+C
in
=
_
x + y +C
in
iff x + y +C
in
< r
x + y +C
in
r otherwise
(24)
C =
_
0 iff x + y +C
in
< r
1 otherwise
where C
in
is the input carry and r
= r 0.5 as dened
previously. The circuit is similar to that in [10], how-
ever, the operating radix can be controlled externally
with our design. HSPice transient simulation results
of the radix-8 multiplexer and full-adder circuits are
shown in Fig. 21.
Since 3 bits in conventional binary logic can rep-
resent one digit in radix-8, performance comparison
is given between 3-bit binary adder and 1 digit MVL
adder, designed with the same parameters. Two well-
known adder schemes, the ripple-carry, RCA, and
carry-look-ahead, CLA, binary adders, described in
[22], are constructed using the same parameter set and
loaded with a capacitive load of 70 fF. In order to eval-
uate a worst-case performance, a large number of ran-
dom input vectors are applied to binary adders. In both
design styles, delays and power dissipation vary with
Table 3. Performance comparison of new MVL-based and binary
adders (T = time duration of one step).
Av. power diss. (mW)
Adder
Trans.
count
Max.
delay
(ns) T = 40 ns T = 10 ns
Area
(m)
2
MVL, radix-8 21 3.2 0.95 1.05 87 24
3-bit RCA 84 6.5 0.14 0.52 160 85
3-bit CLA 108 7.3 0.23 0.93 175 93
Implementation of Multi-Valued Logic Gates 201
Fig. 21. (a) Input waveforms and simulation results of, (b) Radix-8
multiplexer function and (c) Radix-8 full-adder circuit.
Appendix A: Mietecs ES2 0.7 m HSPice parameters
. MODEL Nchan NMOS LEVEL=6
+ LD=1E07 XL=0.04e6 XW=0.9e6 WD=0.45e6 TOX=1.5E08 BETA=97e6 NSUB=1.965E+16
+ XJ=0.25e6 VTO=0.815 VBO=1.50 GAMMA=0.764 LGAMMA=0.705 SCM=1.733 VSH=0.65
+ NWE=193.7E9 NWM=0.197 UFDS=99.5E3 VFDS=0.2 FDS=84E3 MOB=2 F1=372E+3
+ F2=0.2 UTRA=0.563 LAMBDA=10.63E6 ECRIT=87E+3 CLM=3 MCL=4.63 KCL=1.08
+ MAL=0.295 KA=0.974 MBL=0.555 KU=1.4052 NU=1 NFS=5E+11 NSS=0.0 WIC=2.0
+ WEX=17 TLEV=1 BEX=1.5 TCV=2E3 RSH=65 JS=2E6 ACM=2 CJ=503E6 MJ=0.43
+ CJSW=109E12 MJSW=0.43 PB=0.675 CGDO=200E12 CGSO=200E12