Upload
jing-huang
View
212
Download
0
Embed Size (px)
Citation preview
JOURNAL OF ELECTRONIC TESTING: Theory and Applications 23, 163–174, 2006
* 2007 Springer Science + Business Media, LLC. Manufactured in The United States.
DOI: 10.1007/s10836-006-0548-6
On the Tolerance to Manufacturing Defects in Molecular QCA Tiles
for Processing-by-wire*
JING HUANG, MARIAM MOMENZADEH AND FABRIZIO LOMBARDI
Department of Electrical and Computer Engineering, Northeastern University, Boston, MA 02115, [email protected]
Received February 21, 2006; Revised July 6, 2006
Editor: M. Tehranipoor
Abstract. Among emerging technologies, Quantum-dot Cellular Automata (QCA) relies on innovative computational
paradigms. For nano-scale implementation, the so-called processing-by-wire (PBW) paradigm in QCA is very effective as
processing takes place, while signal communication is accomplished. This paper analyzes the defect tolerance properties
of PBW for manufacturing tiles by molecular QCA cells. Based on a 3�3 grid and various input/output arrangements in
QCA cells, different tiles are analyzed and simulated using a coherence vector engine. The functional characterization
and polarization level of these tiles for undeposited cell defects are reported and detailed profiles are provided. It is shown
that novel features of PBW are possible due to the spatial redundancy of the cells in the tiles that permits to retain at high
probability the fault free function in the presence of defects. Moreover, it is shown that QCA tiles are robust and
inherently tolerant to cell defects (by logic equivalence, also additional cell defects can be accommodated).
Keywords: QCA, defect tolerance, processing by wire, emerging technologies
1. Introduction
As CMOS is approaching its physical limits, new devices
for so-called emerging technologies (such as carbon tubes
and tunneling devices) have been proposed. These
technologies have the potential to achieve an extremely
high density (in the range of 1012 devices/cm2), fast
operating speed at Tera Hertz frequencies, together with
room temperature operation and low power dissipation.
Among these emerging technologies, Quantum dotCellular Automata (QCA) [7] is very promising. QCA
design involves diverse new paradigms, such as process-ing-by-wire (PBW) [10]. PBW refers to the ability of
QCA by which information manipulation can be accom-
plished, while transmission and communication of signals
take place. PBW capabilities can be observed in the
inverter chain as well as in the arrangement of cells in an
MV. An implicit PBW feature has also been analyzed in
[14] for testing of QCA devices and their unique logic
features. PBW is applicable to both combinational and
sequential circuits. Sequential circuits are not considered
in this paper as they have been analyzed in other manu-
scripts [17].
Recent research in QCA manufacturing involves
molecular implementations [3, 8]. It is expected that
homogeneous cell arrangements will be constructed by
either self-assembly, or large scale cell deposition on
insulated substrates [8]. Modular QCA design is well
suited to these manufacturing techniques [5, 6]. However,
very little work has been pursued on modular QCA
design to accomplish PBW. The SQUARES methodology
(based on a 5� 5 grid) has been proposed as initial effort
for modularizing QCA design [1]. However, SQUARES
is very limited, because it does not consider cell
interactions extending over the single grid. A tile-based
approach (using a 3� 3 grid) has been proposed in our
earlier work of [5]. It has been shown that tiles are not
only area efficient, but they also offer versatile logic and
interconnection functions. In previous works [1, 5], logic
and interconnect circuits are assumed to be defect free. In
this paper, the defect tolerance of tiles for QCA design is*This manuscript is an extended version of the paper by the same
authors that appeared in [4].
investigated. Five tiles (originally proposed in [5]) are
analyzed. The main focus of this work is defect tolerance
of tiles as basic devices. Design using the 3� 3 tiles has
been presented in [5, 6]. Various tile-based design issues
have been addressed in [6]. Sample circuits built with the
3� 3 tiles have been demonstrated and compared to the
SQUARES methodology and a gate-based design [6]. In
this paper, defect tolerance of 3� 3 tiles is investigated in
detail. The PBW offered by these tiles is extended by
investigating spatial redundancy in the arrangement of the
cells. The logic behavior of the defective tiles are
compared with the defect-free behavior to relate whether
the PBW capability of a tile is changed by the defects.
The polarization characteristics of the functions for all
analyzed tiles are also found by simulation. It is shown
that due to spatial redundancy, the correct functions of the
tiles are tolerant to defects. This further validates the
premises that tile-based design can be used to efficiently
assemble QCA circuits [5].
The rest of the paper is organized as follows. Preliminar-
ies, including review on QCA, PBW by tiling, the assumed
defect model and the simulation environment, are intro-
duced in Section 2. Section 3 presents detailed results for
defect tolerance of tiles. These results are analyzed in
Section 4. Finally, Section 5 concludes the paper.
2. Preliminaries
2.1. QCA Review
A QCA cell can be viewed as a set of four charge
containers or Bdots,^ positioned at the corners of a square
[15]. The cell contains two extra mobile electrons which
can quantum mechanically tunnel between dots, but not
cells. The electrons are forced to the corner positions by
Coulombic repulsion. The two possible polarization states
represent logic B0^ and logic B1,^ as shown in Fig. 1a.
Logic elements are based on two gate primitives
(inverter, or INV, and majority voter, or MV, as shown
in Fig. 1b and c) to implement combinational circuits.
Two arrangements referred to as the binary wire and the
inverter chain can be utilized as interconnects [15], and
are shown in Fig. 1d and e. Sequential as well as
combinational designs can be realized in QCA [9, 11,
18]. Clocking is also required for QCA [2]; clocking
implements timing over a planar layout in four periodic
zones for quasi-adiabatic switching.
Currently, micro-sized QCA devices have been fabricat-
ed with metal cells which operates at 50 mK [12]. Recent
developments in QCA manufacturing have focused on
molecular implementations [3, 8], in which higher speed
and room-temperature operation are expected. Moreover,
homogeneous cell arrangements can be constructed by
self-assembly or large scale cell deposition.
2.2. PBW by Tiling
As in the early stages of VLSI, QCA requires building
blocks that are versatile for flexible manufacturing and
assembly of different circuits. An early proposal for
modular QCA is the so-called SQUARES methodology
[1]; SQUARES is based on a 5� 5 QCA grid. A novel
arrangement has been proposed recently [17] for con-
structing a tile-based serial memory in QCA. Our previous
work in [5] has demonstrated that tiles based on the 3� 3
grid provides universal and versatile logic functions; it has
been shown that both logic and interconnect circuits can be
accomplished by utilizing these QCA tiles [5]. In this
paper, the defect tolerance of these tiles is investigated; it
will be shown that the proposed tiles are not only versatile
in logic function generation, but they are also inherently
defect tolerant. A tile is constructed with nine QCA cells
arranged in a 3� 3 grid. Tiles can be arranged to generate
the desired logic operations within a Cartesian layout. This
scheme provides the following desirable features at
manufacturing [5, 6]: (1) The 3� 3 grid provides a better
cell utilization than the 5� 5 grid of SQUARES [5]; this
is due to the separation of the logic/interconnect imple-
mentations and the reduced size of the building block, i.e.,
9 versus 25 cells in a grid. (2) Tile-based design allows
compatibility with timing/clocking techniques, thus reduc-
ing the length of the longest line of QCA cells in a zone
(and avoiding kinks).
2.3. Defect Model
Defect tolerance is the main focus of this paper. Currently
during manufacturing using a self-assembly molecular
process, defects can occur in both the chemical synthesisphase (in which the QCA cells are manufactured) and the
deposition phase (when the QCA cells are attached to a
substrate) (Lieberman, personal communication). Defects
are more likely to occur in the deposition phase than in the
chemical synthesis phase, often resulting in perfectly
manufactured but imperfectly placed cells. In this paper,
only the defect resulting from an undeposited cell, is
considered (a defect due to additional and incorrectly
deposited cell will be shown later to be logically
logic "0"
celldot
logic "1"
quantum
C
F=Maj(A,B,C)=AB+AC+BCB
A
F=A'A
(b) Inverter(a) QCA cell
(c) Majority Voter (e) Inverter Chain
(d) Binary Wire
Fig. 1. Basic QCA devices.
164 Huang et al.
equivalent). This represents the case when the defective
cell fails to attach to the substrate. Other defects in the
deposition phase such as due to misplacement are unlikely
to be present in a molecular self-assembly process. In the
assumed model, multiple undeposited cell defects may
occur. It will be shown later that additional cell defects can
also be considered as part of the assumed model.
For defect tolerance, the relationship between a fault-
free tile and a tile with undeposited cells is very
important because it defines the paradigm of PBW for a
tile-based design. It is evident that a tile inherently offers
significant defect tolerance as its nine cells closely
interact in a spatial redundancy arrangement.
2.4. Simulation Setup
For evaluating the effects of PBW in QCA, the software
simulation tool QCADesigner1 v1.4.0 (Unix version) is
used [13]. QCADesigner v1.4.0 features different simu-
lation engines. Throughout this paper, the coherence
vector engine is used due to its accurate and detailed
evaluation of QCA. The coherence vector engine is based
on the density matrix approach [13] which models the
power dissipative effects of QCA. The basic functionality
of QCA is based on the Coulombic interaction among
neighboring cells; this depends on the distance as well as
the angle between the two cells. The radius of effect Rdefines the Coulombic interactions within an area
centered on cell i, i.e., all cells within this area interact
with cell i. In all simulations the cell size, is 10� 10 nm
and the cell-to-cell distance is 2:5 nm. The radius of
effect R is set to 40 nm. All other parameters are set to the
default values of QCADesigner.
Hereafter, the following assumptions are made: (1)
Only undeposited cell defects are considered as likely to
occur in molecular implementations. The set of undepos-
ited tiles is denoted by U. (2) A one-dimensional clocking
scheme is assumed for quasi-adiabatic switching: all cells
in a tile are assumed to be within a timing zone; input and
output cells are assumed to be in distinct timing zones.
So, a total of three timing zones per tiles are used in the
simulations. (3) The no logic state (referred to as the
undefined state and denoted by B-^) may occur for some
defective patterns due to lack of definitive polarization at
the output. This happens when the polarization level is
very low (less than �0:1) for at least some input signals,
or a spike or glitch can be observed at the output.
3. Defect Tolerance of Tiles
Five different tiles based on the 3� 3 grid, as shown in
Fig. 2, are investigated in this paper. The orthogonal tile
is used for implementing MV as well as MV-like (MV
with inversion at selected inputs) logic functions. The
baseline and fan-in tiles can be used for signal switching
in the interconnect. The double and triple fan-out tiles are
also used as part of the interconnect. The defect tolerant
properties of these tiles are detailed in this section.
(b) Double Fan-Out tile (c) Baseline tile
(e) Triple Fan-Out tile(d) Fan-In tile
(a) Orthogonal tile
C
F1 2 3
4 5 6
7 8 9
A
1 2 3
4 5 6
7 8 9
F2
F1B
A
B
F2
F11 2 3
4 5 6
7 8 9
F1
B
F3
F21 2 3
4 5 6
7 8 9
A
B
F11 2 3
4 5 6
7 8 9
B
Fig. 2. Tiles based on a 3� 3 grid.
Tolerance to Manufacturing Defects in Molecular QCA Tiles for PBW 165
For a specific tile, the probability of generating a func-
tion f in the presence of jUj undeposited cells is defined
as follows:
Pf ;U ¼Nð f ;UÞ
CjUj9
ð1Þ
where CjUj9 is the number of all possible pattern combina-
tions when jUj cells are undeposited, Nð f ;UÞ is the number
of occurrences of function f in the presence of jUj un-
deposited cells (determined by simulation). In this paper,
each tile is constructed from a fully populated 3� 3 grid
and therefore each tile consists of nine cells that can be
possibly undeposited. For example, assume two cells are
undeposited, i.e., jUj ¼ 2. Then the number of all possible
pattern combinations C29 ¼ 36. If nine of these patterns
generates the desired function f in a specific tile (i.e.,
Nð f ;UÞ=9), then the probability of generating f is
Pf ;U ¼ 936¼ 25%.
3.1. The Orthogonal Tile
Fig. 2a shows the so-called orthogonal tile. Three input
cells (A, B, C) are connected to the sides of the tile, while
an output cell (F) is provided at the remaining side. In the
defect-free case, the output of this tile is Maj(A,B,C)=
ABþ BCþ AC. Thus, this is the basic logic block in a
tile-based design for QCA and its defect-tolerant proper-
ties are very important to assess.
Table 1 shows the simulation results when at most one
cell is undeposited from the orthogonal tile. The probabil-
ity of generating different majority functions versus the
number of undeposited cells is shown in Fig. 3. An
exhaustive simulation has also been pursued for the
orthogonal tile, i.e., with i undeposited cells, i ¼ 1, ....,8
from the layout (note that for all tiles, the absence of all
cells results in all outputs to take an undefined value). For
the orthogonal tile, the numbers of patterns of every
output function when i cells are undeposited, are shown
in Table 2. Once undeposited cell defects are present, the
three input signals may also interact, such that different
functions can be generated at the output, i.e., the relation
between the inputs and placement of the cells for PBW
may be changed. In particular, variants of the majority
function (with complemented input variables) are
Table 1. Single undeposited defect in orthogonal tile.
Undeposited cell F Undeposited cell F
None Maj(A,B,C) 1 Maj(A,B,C)
2 Maj(A_,B,C) 3 Maj(A,B,C)
4 Maj(A,B,C) 5 Maj(A,B,C)
6 Maj(A_,B,C_) 7 Maj(A,B,C)
8 Maj(A,B,C_) 9 Maj(A,B,C)
Fig. 3. Probability of generating different majority functions
vs. number of undeposited cells in the orthogonal tile.
Table 2. Number of occurrences of output functions in the orthogonal
tile with jUj undeposited cell defects.
F
{jUj}
1 2 3 4 5 6 7 8
A 0 0 2 5 9 9 5 1
A0 0 3 13 28 28 15 3 0
B 0 6 12 13 11 6 1 0
C 0 0 2 5 7 10 5 1
C0 0 3 15 28 30 15 3 0
– 0 4 11 24 24 22 16 6
Maj(A;B;C) 6 10 11 7 5 2 0 0
Maj(A0;B;C) 1 3 3 1 0 0 0 0
Maj(A;B;C0) 1 3 3 2 0 0 0 0
Maj(A0;B;C0) 1 3 12 13 12 5 3 0
NXOR 0 1 0 0 0 0 0 0
Total 9 36 84 126 126 84 36 9
Table 3. Functional characterization of orthogonal tile with multiple
undeposited cell defects.
Observations Results
Number of undeposited cells 1 2 3 4
Number of defective patterns 9 36 84 126
Occurrences of wire func. 0 4 22 32
Wire func. percentage 0% 11.1% 26.1% 25.3%
Occurrences on INV func. 0 4 23 51
INV func. percentage 0% 11.1% 27.3% 40.48%
Occurrences of MV func. 6 13 14 8
MV func. percentage 66.7% 36.1% 16.6% 6.34%
Occurrences of MV-like func. 3 11 14 11
MV-like func. percentage 33.3% 30.5% 16.6% 8.73%
Occurrences of undefined 0 4 11 24
Undefined state percentage 0% 11.1% 13.1% 19.1%
166 Huang et al.
expected due to possible input inversion through the cells
of the tile. The variants of the majority function are
referred to as MV-like functions (the cardinality of the
MV-like function set can be at most 7).
The following observations can be made from the
simulation results: (1) In almost all cases, an orthogonal
tile with undeposited cells (as defects) behaves in the
following two ways: wire/inverting functions or MV/MV-
like functions. (2) Undeposited cell defects occurring on
corner cells (cells 1, 3, 7 and 9) do not change the logic
function of the tile, thus confirming the defect tolerant
design of a majority voter. (3) In the simulations using
the coherence vector engine, whenever cell 6 is undepos-
ited, the polarization level experiences a significant drop.
In all simulated occurrences when cell 6 is present, the
magnitude of the maximum polarization is above �0:9.
However, when cell 6 is undeposited, then the magnitude
of the maximum polarization level drops below �0:77.
When other additional cells are undeposited (besides cell
6), in many cases the polarization level for some input
patterns is so low that no definite logic function can be
observed at the output. The statistical results in the
presence of up to four undeposited cells (corresponding to
a defect level just lower than 50%) are summarized in
Table 3. Note that by definition, the MV-like function set
does not include the MV function.
The average magnitude of the maximum polarization
level of the output when a number of cells are undepos-
ited as defects, is shown in Fig. 4. As for previous tiles,
in some cases the output exhibits the no definite polar-
ization level. In some cases, the polarization level is quite
high; also, when increasing the number of undeposited
cells as defects, the decrease in polarization level is not
significant. The analysis presented in this paper is ap-
plicable to tiles as basic QCA blocks within a single timing
zone (excluding the input and output cells); if a tile is con-
nected to another tile using a long QCA wire an appropriate
mechanism must be employed to maintain the correct
polarization level across multiple clocking zones [2].
3.2. Double Fan-out Tile
The double fan-out tile tile, as shown in Fig. 2b, is
analyzed next. The double fan-out tile has one input
(provided by the horizontal cell B) and two outputs (i.e.,
the horizontal output cell F1 and the vertical output cell
F2). In the defect-free case, both outputs follow the value
of the input cell, i.e., F1 ¼ F2 ¼ B (wire function), hence
the tile behaves as a fan-out point in a CMOS interconnect.
The following observations can be made with respect to
the fan-out tile in the presence of undeposited cell defects:
(1) Inversion can be expected at the outputs if one of the
cells connected to the corresponding output cell (i.e., cell 6
or 8) is undeposited. (2) A straightforward QCA imple-
mentation of the fan-out circuit requires only 4 cells (i.e.,
to deposit cells 4, 5, 6 and 8), so redundancy as basis for
defect tolerance is inherently present in this tile. Table 4
shows the results for a single undeposited cell defect.
When multiple cells are undeposited, it can be
observed that in most cases this tile produces either a
wire, or an inverting function (the output is the
complement of the input). The probabilities of generating
different functions for the two outputs (F1 and F2) are
plotted in Fig. 5. Even with four undeposited cells due to
defects, in almost 90% of the cases the tile can still
function either as a wire, or an inverter. This is caused by
its spatial redundancy as discussed previously, therefore,
an excellent level of resilience in functionality is
accomplished. The following additional observations
can be drawn: (1) The probability of being in an
undefined state for an output signal increases with the
number of undeposited cells. (2) The probability of
having a wire function in the horizontal output F1 is
greater than for the vertical output F2, indicating that
signal propagation in the one-dimensional clocking
scheme is stronger along the direction of signal flow
(perpendicular to the direction of the underlying clock
signal, which is implemented as an E field). (3) The
probability of having an inverting function in the vertical
output F2 is greater than for the horizontal output F1.
This is expected due to the 90- orientation of the output
cell with respect to the input cell and the possible 45-misalignments in the defect-free cells. (4) The polariza-
tion plots for two extreme cases are shown in Fig. 6a
(fault-free case or no defect) and b (all nine cells
undeposited). In the defect-free case, both outputs exhibit
Fig. 4. Average magnitude of the maximum polarization level of an
orthogonal tile.
Table 4. Single undeposited defect in double fan-out tile.
Undeposited cell F1 F2 Undeposited cell F1 F2
None B B 1 B B
2 B B 3 B B
4 B_ B_ 5 B B
6 B_ B 7 B B
8 B B_ 9 B B
Tolerance to Manufacturing Defects in Molecular QCA Tiles for PBW 167
the wire function with high polarization levels. In the
extreme case when all cells in the grid are undeposited,
logically F1 still produces the wire function, while F2
performs the inverting function; however, both outputs
have a very low polarization level (below the 0.1 value),
thus an undefined state function is generated.
These results confirm that PBW takes place in its
simplest form, i.e., it performs inversion. Moreover, the
fan-out tile has excellent processing capabilities (wire and
inverting functions), i.e., 93:6% of the exhaustive number
of combinations for up to four undeposited cells. The
average values of the maximum polarization level
(magnitude) at the outputs (F1 and F2) are shown in
Fig. 7. The diagram shows the average magnitude of the
maximum output polarization level when undeposited
cells are incrementally present. The average magnitude of
the maximum polarization level is reported for the wire
function, the inverting function, the undefined state
function as well as the total. The polarization level of
the wire function is higher than the inverting function for
all cases, thus confirming that this tile provides excellent
defect-tolerant capabilities compared to other functional
defective behaviors.
3.3. Baseline Tile
The baseline tile is shown in Fig. 2c; it has two inputs
(one vertical input cell A and one horizontal input cell B)
Fig. 5. Probability of generating different functions vs. number of undeposited cells in the double fan-out tile.
Fig. 6. Polarization plot of fan-out tile.
168 Huang et al.
and two outputs (the horizontal output cell F1 and the
vertical output cell F2). Table 5 shows the simulation
results when at most one cell is undeposited from the
baseline tile. This tile accomplishes wire crossing by
selectively undepositing cells.
For the baseline tile, altogether 20 output functions
(such as F1F2 ¼ AB, AB0; :::) are observed using an
exhaustive simulation. For undeposited cells, no addi-
tional inversion on the input signals is generated (while
still retaining the crossing property), i.e., F1F2 ¼BA0;B0A;B0A0 are not observed in the simulation. For
the fully populated grid, this tile works as a switch (or
coplanar crossing) with F1F2 ¼ BA, i.e., the two input
signals cross each other. In the presence of undeposited
cells, three types of functions are observed. The first type
is the coplanar crossing, that is also the fault free logic
function. The second type is when the tile works as two
L-shape wires, with F1F2 ¼ AB. Also in this type it is the
L-shape wires with inversion, such as F1F2 ¼ A0B,
F1F2 ¼ A0B0. The third type is the fan-out (when the
two outputs follow the same input, such as F1F2 ¼ BB,
F1F2 ¼ AA) and the fan-out with inversion, such as
F1F2 ¼ B0B. The last type is the undefined function in
which at least one output exhibits the undefined state.
The probabilities of having these function types versus
the number of undeposited cells are plotted in Fig. 8.
In the presence of undeposited cells as defects, no
additional inversion on the input signals is generated
(while still retaining the crossing property).
The average value of the maximum polarization level
(magnitude) at the outputs is shown in Fig. 9. The results
also show a near-symmetry in the operation of the outputs
of this tile. For example, when a number of cells are
undeposited, the probability of generating F1 ¼ A is
nearly the same as F2 ¼ B. Moreover, the average
maximum polarization level of A (B) is stronger at
F2(F1).
3.4. Fan-in Tile
The fan-in tile is shown in Fig. 2d. This tile has two
inputs (one vertical input A and one horizontal input B)
and one horizontal output F. In the fault free case, this tile
acts as a wire, with F following the horizontal input B.
Hence, this tile is referred to as the fan-in tile, i.e., only
the horizontal signal flow is permitted and the vertical
input is isolated. The simulation results when at most one
cell is removed from the tile are shown in Table 6.
The exhaustive simulation shows that altogether five
output functions (A, B, A0, B0 and �) are possible in the
fan-in tile. It can be observed that the output follows
either A (with possible inversion) or B (with possible
inversion) or it exhibits no definite polarization. These
results are summarized in Fig. 10.
Fig. 7. Average magnitude of the maximum polarization level of the double fan-out tile.
Table 5. Generation of output function by undepositing at most one cell
in the baseline tile.
U F1 F2 U F1 F2
None B A 5 A0 B0
1 B A 6 A0 A
2 B B 7 A A
3 B B 8 B B0
4 A A 9 B A Fig. 8. Probability of generating different functions vs. number of
undeposited cells in the baseline tile.
Tolerance to Manufacturing Defects in Molecular QCA Tiles for PBW 169
The average maximum polarization level (magnitude)
of the output F is shown in Fig. 11 for different numbers
of undeposited cells.
3.5. Triple Fan-out Tile
The triple fan-out tile is shown in Fig. 2e; this tile has one
input (the horizontal input cell B) and three outputs (the
three cells F1,F2 and F3). The simulation results when at
most a single cell is undeposited from a triple fan-out tile,
are given in Table 7.
Various output functions can be observed for the triple
fan-out tile, such as F1F2F3 ¼ BBB;B0BB; :::. For this
tile, in most cases even when multiple cells are undepos-
ited, the tile produces either a wire or an inverting function.
The probabilities of generating these functions versus the
number of undeposited cells are plotted in Fig. 12. It can be
concluded that the probability of the output in the unde-
fined state increases with an increased number of unde-
posited cells. Additionally, the probability of having a wire
function at the horizontal input F2 is greater than for a
vertical input (F1 or F2), while the probability of having
an inverting function for a vertical input (F1 or F2) is
greater than for the horizontal input F2.
4. Discussion and Analysis of Results
For the polarization level in the presence of undeposited
cell defects, the simulation results have shown the follow-
ing features:
(1) In all tiles, the total average magnitude of the
maximum polarization level decreases by increasing
the number of undeposited cells, as expected.
(2) In all considered tiles, the probability of no polariza-
tion increases when increasing the number of
undeposited cells. This is expected, because no
polarization will be encountered due to the large
inter-cell spacing. Additionally, in all considered
tiles, the curves for the probability of having the
undefined state versus the number of undeposited
cells have similar shapes (as shown in Figs. 5, 8, 10
and 12). This suggests that the probability of having
an undefined state at the output is independent of the
type of the tile, but it is dependent on the number of
undeposited cells.
As for functional characterization, the tiles show the
following behavior:
(1) The double and triple fan-out tiles have similar defect
tolerant properties (as shown in Figs. 5 and 12). The
fault-free function in these tiles is the wire function.
Table 6. Generation of output function by undepositing at most one cell
in the fan-in tile.
U F U F
None B 5 B
1 B 6 A0
2 B 7 B
3 B 8 B
4 A 9 B
Fig. 9. Average maximum polarization magnitude of baseline tile.
Fig. 10. Probability of generating different functions vs. number of
undeposited cells in the fan-in tile.
170 Huang et al.
It can be observed that with one undeposited cell,
the probability of having the correct wire function
at the outputs is >75% for the double fan-out tile and
> 65% for the triple fan-out tile. In both these tiles,
the probability of obtaining the correct wire function
at the horizontal output is greater than the probability
of obtaining the correct wire function at the vertical
output(s). Even with multiple undeposited cells, in
most cases the tile still produces a stable logic func-
tion: either the wire function, or the inverter function
(as shown in Fig. 13). These functions are very useful
for logic design [6].
(2) The MV-like function in the orthogonal tile com-
bines the logic function of MV and INV and can be
used efficiently in tile-based logic design [6]. Design
using MV-like functions has been investigated in
detail in [6]. Additionally, Fig. 3 illustrates that the
MV and different MV-like functions have unequal
sensitivity to undeposited cell defects. For example,
the probability of having the MV function
(F ¼Maj(A,B,C)) declines sharply by increasing the
number of undeposited cells, while the probability of
having the MV-like function F ¼Maj(A_,B,C_) stays
roughly the same with an increased number of
undeposited cells.
(3) The baseline tile acts as a switch (coplanar crossing)
in fault-free conditions. However, this switching
function is not very fault tolerant. Even with only
one undeposited cell, the probability of having the
switching function is less than 25%. With multiple
undeposited cells, in most cases this tile acts as a fan-
out (with possible inversion) tile, i.e., the two outputs
follow the same input.
(4) The presence of new logic functions (such as the
inverting function in the fan-out tile, the MV-like
functions in the orthogonal tile [6], or the fan-out
function in the baseline tile) shows that defect
tolerance can be utilized in accomplishing PBW even
under a large number of defective cells and high
defect levels.
5. Conclusion
In this paper, the defect tolerance of QCA tiles has been
analyzed under undeposited cell defects, as applicable to
molecular QCA. QCA tiles can be used as basic modular
blocks by which defect-tolerance can be in the future
extended to circuit-level [16]. By logic equivalence, this
analysis also illustrates the tolerance to additional (extra)
cell defects as the results provide the output character-
istics of a 3� 3 tile with less than nine cells (hence, the
presence of additional cell defects can also be accounted).
Specifically, in the defect model each cell can be assumed
to be erroneously deposited or undeposited in a tile for
generating the desired output function. Circuit design
using QCA tiles has been investigated in a previous paper
[6]. If a specific tile with k cells is the desired (fault free)
tile, the simulation results of this paper can also be
utilized for additional cell defects, i.e., for defects in
which unwanted cells are deposited (such that more than
k cells are present in the tile). Hence, a tile with k QCA
cells and jUj undeposited cell defects has the same
characteristics (polarization and functional behavior) as a
tile with k � jUj QCA cells with jUj additional cell
defects (provided all undeposited and additional cell
defects occur at same locations). Therefore, a logic
equivalence exists between a model made of undeposited
cell defects and a model made of additional cell defects
and the results presented in this paper can also be
extended to this last scenario.
The simulation results have shown that PBW by tiling
in the presence of cell defects is still very versatile and
robust. Except for the baseline tile, the capability of
generating the defect-free function is preserved with very
high probability for at most one defective cell per tile.
Moreover in the presence of multiple undeposited cells,
QCA tiles can still be used in most cases to perform
useful logic functions. Throughout an exhaustive simula-
tion, the following logic functions consistently appear at
the output(s) of the tiles: (1) the wire function, (2) the
inverting function, (3) the majority function, and (4) the
majority-like functions (i.e., majority function with one or
two complemented variables at the inputs).
The MV function is only observed in the orthogonal tile,
indicating that this tile plays a significant role in tile-based
QCA design (as the wire and INV functions are observed
Table 7. Generation of output function by undepositing at most one cell
in the triple fan-out tile.
U F1 F2 F3 U F1 F2 F3
None B B B 5 B0 B0 B0
1 B B B 6 B B0 B
2 B0 B B 7 B B B
3 B B B 8 B B B0
4 B0 B0 B0 9 B B B
Fig. 11. Average maximum polarization magnitude of fan-in tile.
Tolerance to Manufacturing Defects in Molecular QCA Tiles for PBW 171
Fig. 13. Probability of generating wire and INV functions in the tiles.
Fig. 12. Probability of generating different functions vs. number of undeposited cells in the triple fan-out tile.
172 Huang et al.
for all simulated tiles). A comparison of the results of this
paper with those of [6] suggests that as tile-based design is
not restrictive as a design utilizing discrete gates (such as
MV and INV) when assembling QCA circuits, defect
tolerance can also be incorporated into PBW. This feature
of QCA tiles exploits design modularity and its ability of
generating the same functions using different tiles with
arrangements in the input/output cells.
Moreover, in the presence of multiple undeposited
cells, tiles have a high probability of performing some
deterministic logic functions, even though it might be a
different logic function. As this is deterministic, a post-
fabrication step can be employed to reprogram the circuit
for performing an useful logic function. A process similar
to healing as for assembly [19] is envisioned.
Current research is been pursued by designing defect-
tolerant QCA circuits using the proposed tiles within a
CAD framework [16].
Note
1. QCADesigner is developed by the ATIPS lab at the University of
Calgary in Canada.
References
1. D. Berzon and T.J. Fountain, BA Memory Design in QCAs Using
the SQUARES Formalism,^ in Proceedings of the 9th Great Lakes
Symposium on VLSI, Ypsilanti, MI, 1999, pp. 166–169.
2. K. Hennessy and C.S. Lent, BClocking of Molecular Quantum-dot
Cellular Automata,^ J. Vac. Sci. Technol., vol. 19, no. 5, pp. 1752–
1755, 2001.
3. W. Hu, K. Sarveswaran, M. Lieberman, and G.H. Bernstein, BHigh-
resolution Electron Beam Lithography and DNA Nano-patterning
for Molecular QCA,^ IEEE Trans. Nanotechnology, vol. 4, no. 3,
pp. 312–316, 2005.
4. J. Huang, M. Momenzadeh, and F. Lombardi, BDefect Tolerance of
QCA Tiles,^ in Proceedings of the IEEE Design, Automation and
Test in Europe Conference, Munich, Germany, 2005, pp. 774–779.
5. J. Huang, M. Momenzadeh, L. Schiano, and F. Lombardi,
BSimulation-based Design of Modular QCA Circuits,^ in Proceed-
ings of the IEEE Conference on Nanotechnology, Paper WE-P7-1,
IEEE CD-ROM 05TH8816C, Nagoya, Japan 2005.
6. J. Huang, M. Momenzadeh, L. Schiano, M. Ottavi, and F.
Lombardi, BTile-based QCA Design Using Majority-like Logic
Primitives,^ ACM Journal of Emerging Technologies in Computing,
vol. 1, no. 3, pp. 163–185, 2005.
7. C.S. Lent, P.D. Tougaw, and W. Porod, BQuantum Cellular Automata:
The Physics of Computing with Arrays of Quantum Dot Molecules,^in Proceedings of the Workshop on Physics and Computing, IEEE
Computer Society Press, Los Alamitos, CA, 1994, pp. 5–13.
8. M. Lieberman, S. Chellamma, B. Varughese, Y. Wang, C.S.
Lent, G.H. Bernstein, G. Snider, and F. Peiris, BQuantum-dot
Cellular Automata at a Molecular Scale,^ Ann. N.Y. Acad. Sci.,
vol. 960, pp. 225–239, 2002.
9. M.T. Niemier and P.M. Kogge, BLogic-in-wire: Using Quantum
Dots to Implement a Microprocessor,^ Int. Conf. Electron., Circuits,
Syst., vol. 3, pp. 1211–1215, 1999.
10. M.T. Niemier and P.M. Kogge, BProblems in Designing With
QCAs: Layout=Timing,^ Int. J. Circuit Theory Appl., vol. 29, no. 1,
pp. 49–62, 2001.
11. M.T. Niemier, A.F. Rodrigues, and P.M. Kogge, BA Potentially
Implementable FPGA for Quantum Dot Cellular Automata,^ 1st
Workshop on Non-silicon Computation, Boston, MA, 2002.
12. A.O. Orlov, I. Amlani, G.H. Bernstein, C.S. Lent, and G.L. Snider,
BRealization of a Functional Cell for Quantum-dot Cellular
Automata,^ Science, vol. 277, pp. 928–930, 1997.
13. QCADesigner Home Page: http://www.qcadesigner.ca/
14. M.B. Tahoori, M. Momenzadeh, J. Huang, and F.Lombardi,
BDefects and Faults in Quantum Cellular Automata,^ VLSI Test
Symposium, Napa Valley, CA, 2004, pp. 291–296.
15. P.D. Tougaw and C.S. Lent, BLogical Devices Implemented
Using Quantum Cellular Automata,^ J. Appl. Phys., vol 75, no. 3,
pp. 1818–1825, 1994.
16. V. Vankamamidi, BNano-based Homogeneous Design of High-
performance Systems by QCA,^ Northeastern University, Ph.D.
Dissertation (in progress).
17. V. Vankamamidi, M. Ottavi, and F. Lombardi, BTile Based Design
of a Serial Memory in QCA,^ in Proceedings of the Great LakesSymposium on VLSI, 2005, Chicago, IL, pp. 201–206.
18. K.Walus, A.Vetteth, G.A.Jullien, and V.S.Dimitrov, BRAM Design
Using Quantum-dot Cellular Automata,^ in Proceedings of the
IEEE Nanotechnology Conference, San Francisco, CA, vol 2, 2003,
pp. 160–163.
19. E. Winfree BSelf-healing Tile Sets,^ Nanoscience Science and
Computation, pp. 55–78, 2006.
Jing Huang recieved her B.S. degree in electronics engineer-
ing from Fudan University, Shanghai, China in 2001 and M.S.
degree in electrical engineering from Northeastern University,
Boston. She is currently a Ph.D. student and research assistant in
ECE department, Northeastern University, Boston. Before that
she has worked in the Computer Aided Test Lab in EE
Department, Fudan University as research assistant from 1999
to 2001. She also worked for Huawei Technologies Inc. China as
hardware design engineer from 2001 to 2002. Her research
interests include testing, design for testability and fault tolerance
of VLSI, reconfigurable systems and nanotechnologies.
Mariam Momenzadeh was born in Tehran, Iran. She
received her B.Sc. degree in electrical engineering from Sharif
University of Technology, Tehran, Iran in 1999 and M.Sc.
degree in computer engineering and Science from University of
Connecticut, Storrs in 2003. In 2006, she recieved her Ph.D.
degree in electrical and computer engineering department at
Northeastern University, Boston. Her research interests lie in
testing, design for testability and fault tolerance issues in
digital systems, ATE systems and nano-technologies.
Fabrizio Lombardi graduated in 1977 from the University
of Essex (UK) with a B.Sc. (Hons.) in electronic engineering.
In 1977 he joined the Microwave Research Unit at University
College London, where he received the master_s degree in
microwaves and modern optics (1978), a diploma in microwave
engineering (1978) and a Ph.D. degree from the University of
London (1982).
Tolerance to Manufacturing Defects in Molecular QCA Tiles for PBW 173
He is currently the holder of the International Test
Conference (ITC) endowed professorship at Northeastern
University, Boston. At the same institution during the period
1998-2004 he served as Chair of the Department of Electrical
and Computer Engineering. Prior to Northeastern University
he was a faculty member at Texas Tech University, the
University of Colorado-Boulder and Texas A&M University.
Dr. Lombardi was an associate editor (1996-2000) of IEEE
Transactions on Computers and a distinguished visitor of the
IEEE-CS (1990-1993). Since 2000, he has been the associate
editor-in-chief of IEEE Transactions on Computers. Currently
he is also an associate editor of the IEEE Design and Test
Magazine and a distinguished visitor of the IEEE-CS; he is
also the chair of the committee on nanotechnology devices
and systems of the Test Technology Technical Council of the
IEEE.
His research interests are testing and design of digital
systems, quantum computing, ATE systems, configurable/
network computing, defect tolerance and CAD VLSI. He
has extensively published in these area and edited six books.
174 Huang et al.