Title: From Quantum Gates to Quantum Learning: recent research and open problems in quantum circuits
1From Quantum Gates to Quantum Learning recent
research and open problems in quantum circuits
- Marek A. Perkowski,
- Portland Quantum Logic Group,
- Department of Electrical Engineering and Computer
Science, - Korea Advanced Institute of Science and
Technology, and - Department of Electrical and Computer
Engineering, - Portland State University, USA.
2The computer as we know it?
1999 Pentium IIIB www.icknowledge.com
1947 First point contact transistor by Bardeen
and Brattain
http//www.pbs.org/transistor/science/events/point
ctrans.html
3Nano-systemHow small is a nanometer?
- 1 meter
- 10 mm
- 1 mm
- 10 nm
- 1nanometer
- 0.1 nm
- 1 picometer
- 1 femtometer
- Size of red blood cell
- a millionth of a meter
- Size of polio virus
- a billionth of a meter
- Size of the hydrogen atom
- a trillionth of a meter
- 10 -15 m, size of a proton
4History
- 1970s and 1980s, introduction of quantum
computers (Richard Feynmann, David Deutsch, and
Paul Benioff) - 1994, Peter Shors factoring algorithm
- 1996, Lov Grover, searching algorithm
- 1998, 1999, 2001 Isaac L. Chuang, developed the
world's first 2-qubit, 3-qubit, 5-qubit and
7-qubit quantum computer
5People
First Ideas(1982)
Turing Machine (1936)
A. Turing
R. Feyemann
Quantum Circuits(1985)
Factorization (1997)
D. Deutsch
P. Shor
6Number of Atoms in a Useful SystemFrom R. Keyes,
IBM J. Res. Develop (1988) atoms to store a
bit dopant atoms/bipolar transistor
7EX Quantum Parallelism
- Quantum
- Put all 7-bits into a superposition state
- superposition allows quantum computer to make
calculations on all 128 possible numbers (27) in
ONE iteration i.e. finishes in 1 second. - Tremendous possibilities imagine doing
computations on even larger sample spaces all at
the same time!!!
8Jiffy Quantum Theory
Info unit 1 bit. Physical system 2 states
1gt
0gt
0gt and 1gt
- Quantum nature a combination of both.
- In preparing the initial state only one of the 2
states - On measurement only one state found.
- Probability the states component in the mix
- Both preparation and measurement in contact with
a macro system
9Qubits as binary Qudits
- In multi-valued (MV) Quantum Computing (QC), the
unit of memory (information) is qudit. - For instance, ternary logic values of 0, 1, and 2
are represented by a set of distinguishable
different basis states of a qutrit. - These states can be a photons polarizations or
an elementary particles spins. - After encoding these distinguishable quantities
into multiple-valued values, qutrit states are
represented by basis states 0gt, 1gt and 2gt ,
respectively. - A qubit, used in binary QC uses only two basis
states, 0gt and 1gt - Qubit and qutrit are then special cases of qudits
10Qudits
- Qudits exist in a linear superposition of
states, and are characterized by a wave function
. - As an example (), it is possible to have light
polarizations other than purely horizontal or
vertical, such as slant 45? corresponding to the
linear superposition of . - In ternary logic, the notation for the
superposition is , where ?, ?, and ? are complex
numbers. - These intermediate states cannot be
distinguished, rather a measurement will yield
that the qutrit is in one of the basis states, ,
, or . - The probability that a measurement of a qutrit
yields state is , state is , and state is . - The sum of these probabilities is one.
- The absolute values are required since, in
general, ?, ? and ? are complex quantities. - Pairs of qutrits are capable of representing nine
distinct states,, , , , , , , , and , as well as
all possible superpositions of these states.
11 12Quantum Logic
Single photon
Specchio
50
1
0
50
Optical sensor
13 strange behavior
0
1
0
1
14Quantum Gate
0
1
0
1
1
0
1
0
NOT
15Qubit
16Qubit in a Ion Trap
17Deterministic Turing Machine
Initial State
Final State
Deterministic Turing Machine transits
deterministically from initial to final state.
18Probabilistic Turing Machine
Probabilistic output states
P4
Probabilities of final output states
P5
P1
P6
P2
P P2P7 P3P8
P7
P3
P8
P9
19Quantum Computation
A A1A2 A3A4
P A1A2 A3A42 A1A2 A3A42 2Re(A1A2A3A
4)
A1
A2
A3
A4
20Decoherence
21A beam-splitter
The simplest explanation is that the
beam-splitter acts as a classical coin-flip,
randomly sending each photon one way or the other.
22Quantum Interference
The simplest explanation must be wrong, since it
would predict a 50-50 distribution.
23More experimental data
24A new theory
The particle can exist in a linear combination or
superposition of the two paths
25Probability Amplitude and Measurement
If the photon is measured when it is in the
state then we get with probability
26Quantum Operations
The operations are induced by the apparatus
linearly, that is, if and then
27Quantum Operations
Any linear operation that takes
states satisfying and maps them to
states satisfying must be UNITARY
28Linear Algebra
corresponds to
corresponds to
corresponds to
29Linear Algebra
corresponds to
corresponds to
30Linear Algebra
corresponds to
31Linear Algebra
is unitary if and only if
32Abstraction
The two position states of a photon in a
Mach-Zehnder apparatus is just one example of a
quantum bit or qubit
Except when addressing a particular physical
implementation, we will simply talk about basis
states and and unitary operations
like and
33 Re
0gt
(c)
Im
1gt
0gt
0gt
(d)
-
1gt
1gt
340gt
1gt
(b)
(a)
(c)
(d)
35An arrangement like
is represented with a network like
36(a)
cos?
-
sin?
cos?
sin?
(b)
370gt
0gt
00gt
0gt
00gt
01gt
1gt
01gt
1gt
1gt
0gt
10gt
10gt
11gt
1gt
11gt
(b)
38More than one qubit
If we concatenate two qubits
we have a 2-qubit system with 4 basis states
and we can also describe the state as or by
the vector
39More than one qubit
In general we can have arbitrary
superpositions
where there is no factorization into the tensor
product of two independent qubits. These states
are called entangled.
40Measuring multi-qubit systems
If we measure both bits of we get with
probability
41Classical Versus Quantum
42Classical vs. Quantum Circuits
- Goal Fast, low-cost implementation of useful
algorithms using standard components (gates) and
design techniques - Classical Logic Circuits
- Circuit behavior is governed implicitly by
classical physics - Signal states are simple bit vectors, e.g. X
01010111 - Operations are defined by Boolean Algebra
- No restrictions exist on copying or measuring
signals - Small well-defined sets of universal gate types,
e.g. NAND,AND,OR,NOT, AND,NOT, etc. - Well developed CAD methodologies exist
- Circuits are easily implemented in fast,
scalable and macroscopic technologies such as CMOS
43Classical vs. Quantum Circuits
- Quantum Logic Circuits
- Circuit behavior is governed explicitly by
quantum mechanics - Signal states are vectors interpreted as a
superposition of binary qubit vectors with
complex-number coefficients - Operations are defined by linear algebra over
Hilbert Space and can be represented by unitary
matrices with complex elements - Severe restrictions exist on copying and
measuring signals - Many universal gate sets exist but the best types
are not obvious - Circuits must use microscopic technologies that
are slow, fragile, and not yet scalable, e.g., NMR
44Quantum Circuit Characteristics
- Unitary Operations
- Gates and circuits must be reversible
(information-lossless) - Number of output signal lines Number of input
signal lines - The circuit function must be a bijection,
implying that output vectors are a permutation of
the input vectors - Classical logic behavior can be represented by
permutation matrices - Non-classical logic behavior can be represented
including state sign (phase) and entanglement
45Quantum Circuit Characteristics
- Quantum Measurement
- Measurement yields only one state X of the
superposed states - Measurement also makes X the new state and so
interferes with computational processes - X is determined with some probability, implying
uncertainty in the result - States cannot be copied (cloned), implying that
signal fanout is not permitted - Environmental interference can cause a
measurement-like state collapse (decoherence)
46Classical vs. Quantum Circuits
Classical adder
47Classical vs. Quantum Circuits
Quantum adder
48Reversible Circuits
49Reversible Circuits
- Reversibility was studied around 1980 motivated
by power minimization considerations - Bennett, Toffoli et al. showed that any classical
logic circuit C can be made reversible with
modest overhead
i
i
Junk
Reversible Boolean Circuit
f(i)
Junk
50Reversible Circuits
- How to make a given f reversible
- Suppose f i ? f(i) has n inputs m outputs
- Introduce n extra outputs and m extra inputs
- Replace f by frev i, j ? i, f(i) ? j where ?
is XOR - Example 1 f(a,b) AND(a,b)
- This is the well-known Toffoli gate, which
realizes AND when c 0, and NAND when c 1.
51Reversible Circuits
- Reversible gate family Toffoli 1980
- Every Boolean function has a reversible
implementation using Toffoli gates. - There is no universal reversible gate with fewer
than three inputs
52Quantum Gates
53Quantum Gates
- One-Input gate NOT
- Input state c00? c11?
- Output state c10? c01?
- Pure states are mapped thus 0? ? 1? and 1? ?
0? - Gate operator (matrix) is
- As expected
54- One-Input gate NOT
- Input state c00? c11?
- Output state c10? c01?
- Pure states are mapped thus 0? ? 1? and 1? ?
0? - Gate operator (matrix) is
- As expected
55 56Quantum Gates
- One-Input gate Square root of NOT
- Some matrix elements are imaginary
- Gate operator (matrix)
- We find
- so 0? ?
0? with probability i/?22 1/2 - and 0? ? 1? with probability 1/
? 22 1/2 - Similarly, this gate randomizes input 1?
- But concatenation of two gates eliminates the
randomness!
57Quantum Gates
- One-Input gate Hadamard
- Maps 0? ? 1/ ? 2 0? 1/ ? 2 1? and 1? ? 1/ ?
2 0? 1/ ? 2 1?. - Ignoring the normalization factor 1/ ? 2, we can
write - x? ? (-1)x x? 1 x?
- One-Input gate Phase shift
-
?
58Quantum Gates
- Universal One-Input Gate Sets
- Requirement
- Hadamard and phase-shift gates form a universal
gate set - Example The following circuit generates y?
cos ? 0? ei? sin ? 1? up to a global factor
59Quantum Gates
- Two-Input Gate Controlled NOT (CNOT)
- CNOT maps x?0? ? x?x? and x?1? ? x?NOT
x? - x?0? ? x?x? looks like cloning, but its
not. These mappings are valid only for the pure
states 0? and 1? - Serves as a non-demolition measurement gate
60(b)
(a)
(c)
(d)
61Quantum Gates
- 3-Input gate Controlled CNOT (C2NOT or Toffoli
gate)
a?
a?
b?
b?
c?
ab ? c?
62(a)
(b)
000gt
000gt
001gt
001gt
010gt
010gt
011gt
011gt
100gt
100gt
(c)
101gt
101gt
110gt
110gt
111gt
111gt
63Quantum Gates
- General controlled gates that control some
1-qubit unitary operation U are useful
etc.
U
U
U
C(U)
C2(U)
U
64Quantum Gates
- Universal Gate Sets
- To implement any unitary operation on n qubits
exactly requires an infinite number of gate types - The (infinite) set of all 2-input gates is
universal - Any n-qubit unitary operation can be implemented
using ?(n34n) gates Reck et al. 1994 - CNOT and the (infinite) set of all 1-qubit gates
is universal
65Quantum Gates
- Discrete Universal Gate Sets
- The error on implementing U by V is defined as
-
- If U can be implemented by K gates, we can
simulate U with a total error less than ? with a
gate overhead that is polynomial in log(K/?) - A discrete set of gate types G is universal, if
we can approximate any U to within any ? gt 0
using a sequence of gates from G
66Quantum Gates
- Discrete Universal Gate Set
- Example 1 Four-member standard gate set
CNOT Hadamard Phase ?/8
(T) gate
- Example 2 CNOT, Hadamard, Phase, Toffoli
67 Quantum Circuits
68Quantum Circuits
- A quantum (combinational) circuit is a sequence
of quantum gates, linked by wires - The circuit has fixed width corresponding to
the number of qubits being processed - Logic design (classical and quantum) attempts to
find circuit structures for needed operations
that are - Functionally correct
- Independent of physical technology
- Low-cost, e.g., use the minimum number of qubits
or gates - Quantum logic design is not well developed!
69Quantum Circuits
- Ad hoc designs known for many specific functions
and gates - Example 1 illustrating a theorem by Barenco et
al. 1995 Any C2(U) gate can be built from
CNOTs, C(V), and C(V) gates, where V2 U
70Quantum Circuits
0? 1? x?
0? 1?
0? 1? x?
0? 1?
0? 1? Vx?
0? 1? x?
?
U
71Quantum Circuits
Example 1 Simulation (contd.)
1? 1? x?
1? 1? Vx?
1? 0?
1? 0? Vx?
1? 1?
1? 1? Ux?
?
- Exercise Simulate the two remaining cases
72Quantum Circuits
Example 1 Algebraic analysis
- Is U0(x1, x2, x3) U5U4U3U2U1(x1, x2, x3)
- (x1, x2, x1x2 ? U (x3) ) ?
73Quantum Circuits
74Quantum Circuits
75Quantum Circuits
- Example 1 (contd)
- U5 is the same as U1 but has x1and x2 permuted
(tricky!) - It remains to evaluate the product of five 8 x 8
matrices U5U4U3U2U1 using the fact that VV I
and VV U
76Quantum Circuits
- Implementing a Half Adder
- Problem Implement the classical functions sum
x1 ? x0 and carry x1x0 - Generic design
x1?
x1?
x0?
x0?
Uadd
y1?
y1? ? carry
y0?
y0? ? sum
77Quantum Circuits
- Half Adder Generic design (contd.)
78Quantum Circuits
- Half Adder Specific (reduced) design
x1?
x1?
CNOT
C2NOT (Toffoli)
x0?
sum
y?
y? ? carry
79Walsh Transform for two binary-input many-valued
variables
Classical logic
Quantum logic
Variable 1
Variable 1
Butterfly is created automatically by tensor
product corresponding to superposition
80Computation
0 0 0
0 0 1
0 1 0
0 1 1
1 0 0
1 0 1
1 1 0
1 1 1
G(0 0 0)
G(0 0 1)
G(0 1 0)
G(0 1 1)
G(1 0 0)
G(1 0 1)
G(1 1 0)
G(1 1 1)
G(x) QC
81Quantum Gate Arrays
1-bit full adder
cgt
cgt
xgt
xgt
ygt
ygt
0gt
sgt
1gt
1gt
0gt
0gt
cgt
0gt
0gt
1gt
Let cgt 1gt, xgt 0gt, ygt 1gt
sgt 0gt, cgt 1gt
82Quantum Gate Arrays
It is possible to construct reversible quantum
gates for any classical computable function f
with m input and k output bits.
There exists a quantum gate array that implements
the unitary transformation Uf x, ygt ? x, y ?
f(x)gt, where ? indicates bitwise xor.
xgt
xgt
ygt
y ? f(x)gt
83Quantum Gate Arrays
The previous transformation Uf is reversible
Uf Uf Uf Uf Uf Uf I
xgt
xgt
xgt
y ? f(x)gt
y ? f(x) ? f(x)gt
ygt
But y ? f(x) ? f(x)gt ygt
84Superposition of Quantum States
Consider the Tofolli Gate
xgt
xgt
ygt
ygt
0gt
xgt ? ygt
Apply T, the Tofolli transform, to the
superposition of all inputs.
T(H0gt ? H0gt ? 0gt ) ½(000gt 010gt 100gt
111gt)
Quantum parallelism Applying the Tofolli
transform to a superposition of all of the input
states produces a superposition of all of the
states in the truth table
85Superposition of Quantum States
BUT
Only one of the superposed states can be
extracted by measurement
0 1
T(H0gt ? H0gt ? 0gt ) ½(000gt 010gt 100gt
111gt)
000gt 010gt 100gt
0
111gt
1
Measurement of the output projects the
superposition onto the set of states consistent
with the result
86Quantum Parallelism
In order to take advantage of quantum parallelism
one must
1. Transform the state in such a way as to
amplify the values of interest so that they
have a higher probability of being selected
during measurement
Grovers unstructured search algorithm
2. Find common properties of ALL the states of
f(x)
Shors factoring algorithm
87Where to learn more
- Web Page of Marek Perkowski
- class 572 - see description of student projects
- Portland Quantum Logic Group
88Kronecker Product of Matrices
- Superposition property may be mathematically
described using the Kronecker product (tensor
product) operation ? - The Kronecker product of two matrices is defined
as follows
89Tensor Products
- Similarly one can define tensor products for any
size of matrices and in particularly for vectors
representing superposed states. - As an example, consider two qutrits with and .
- When the two qutrits are considered to represent
a state, that state is the superposition of all
possible combinations of the original qutrits,
where
90Superposition
- The superposition property allows the qubit
states to grow much faster in dimension than
classical bits, and the qudits faster than
qubits. - In a classical system, n bits represent distinct
states, whereas n qubits correspond to a
superposition of 2n states and n qutrits
correspond to a superposition of 3n states. - In the above formula some coefficient can be
equal to zero, so there exists a constraint
bounding the possible states in which the system
can exist (entanglement). - Allowing d to be arbitrary enables a tradeoff
between the number of qudits making up the
quantum computer and the number of levels in each
qudit. - Because in contemporary quantum technologies
every qubit is costly, higher radices than 2 give
an advantage of improved processing and storage
power at the same realization cost. - This is a strong argument for realization of
multi-valued logic in quantum circuits. - In addition to standard advantages of mv logic,
quantum mv logic may be superior to binary one
because of different nature of entanglement.
91Quantum Notation
- An output of a gate is obtained by multiplying
the unitary matrix of this gate by a vector of
Hilbert space corresponding to this gates input
state. - (Unitary matrix U is one such that U . U I,
where U is a Hermitian matrix of U. A Hermitian
U is a conjugate transpose matrix of U). - A gate or a sub-circuit of a quantum circuit
corresponds to a unitary matrix. - As shown below, the resultant unitary matrix of
an arbitrary quantum circuit is created by matrix
multiplications or Kronecker multiplications of
matrices of its composing sub-circuits. - Various quantum notations contribute to the
difficulty in understanding the concepts of
quantum computing and creating efficient
analysis, simulation, verification and synthesis
algorithms for QC. - Generally, however, we believe that once the
minimal amount of formalism is understood, logic
researchers can quickly contribute to new
designs, since much can be learned from the
history of Electronic Design Automation as well
as from MV logic theory and design. - The lessons learned there should be used to
design efficient QDA tools for MV quantum
computing. - Here we include the absolute minimum amount of
formalism sufficient to start independent
software development by people who have
sufficient background in EDA tools and algorithms
such as search or evolutionary programming
92Quantum Circuits
- In terms of logic operations, anything that
changes a vector of qudit states to another qudit
vector satisfying measurement probability
properties can be considered as a quantum
operator (unitary matrix). - These phenomena can be modeled using the analogy
of a quantum circuit (called also quantum
array). - In a quantum circuit, a wire does not carry
ternary values but corresponds to a 3-tuple of
complex values, ?, ?, and ?. - Quantum logic gates of the circuit map the
complex values on their inputs to complex values
on their outputs. - As mentioned, operation of all quantum gates and
their assemblies is described by matrix
operations. - Any quantum circuit is a composition of parallel
and serial connections of blocks, from small to
large.
93Analysis of Quantum Circuits
- Small blocks correspond to quantum gates that are
easily directly realizable (like Pauli rotations)
or are very simple and require just few basic
quantum operations such as Feynman gates or
Stroud/Muthukrishnan gates. - Serial connection of blocks corresponds to
multiplication of their (unitary) matrices. - Parallel connection corresponds to Kronecker
multiplication of their matrices. - So, theoretically, the analysis, simulation and
verification are easy and can be based on matrix
methods. - Practically they are tough because the dimensions
of the matrices grow exponentially. - All these become much easier when one deals only
with permutative matrices, which are equivalent
to multi-output truth tables of completely
specified functions. In such matrices there is
exactly one 1 in every row and column. - An active research area is to represent
operations on unitary matrices (in particular,
the permutation matrices) by new efficient data
structures and algorithms.
94Calculating output state of QC
- Typically the symbols 0gt and 1gt are not
present in the matrix formulation of the
equations, only the probability amplitudes (i.e.
? and ?) are included however, there are kept in
Equation (1) for illustrative purposes. -
(1)
95- Because the qubit probabilities must be preserved
at the output of the quantum gate, all matrices
representing them are unitary. - An important unitary matrix property is that of a
full rank. - This property implies that quantum gate matrix
rows and columns are orthonormal. - Therefore, past results from spectral methods for
classic digital logic are directly applicable to
quantum logic synthesis. - Furthermore, since quantum logic gates are
represented using unitary orthonormal matrices,
they represent logically reversible gates. - These observations mean that the single
input/output quantum logic gates as represented
in Equation (1) are rotation matrices
characterized by some particular rotation angle
?, where, for example, a cos?, b sin?, c
-sin? and d cos?. - With this viewpoint, it can be seen that there
are, in fact, an infinite number of single
input/output qubit gates.
96Rotation Gates
- However, three elementary gates can be used to
generate any rotation 7. - These are the R, S, and T gates described in
matrix notation by - (1a)
97Quantum XOR gate
- Called also Feynman gate or Controlled Not gate.
- This gate allows inputs of 00gt and 01gt to
appear unchanged at the outputs, but interchanges
the pairs 10gt and 11gt. - For example, consider the quantum XOR gates
operation for an input 10gt.
98XOR logic synthesis is useful for QC
- In this example, the input is 10gt
((0)0gt(1)1gt) ?((1)0gt(0)1gt), and the input
vector is represented by the coefficients shown
in parentheses. - It is a significant fact that the unitary gates
described by Equations (1) and (2) can realize
any quantum logic function (including standard
binary). - There are several strong similarities of quantum
logic to classic digital circuit design using
AND/XOR logic. - Our research group has been heavily involved in
AND/XOR logic circuit design as well as related
algebraic and spectral methods for several years. -
- We found these experiences very useful in quantum
circuit design.
99Bloch Sphere
- The normalization ?2 ?2 1 admits the
parametrization ? cos(?/2) e j? , ? sin(?/2)
e j?. - ?? e j? (cos (? / 2) 0? e j ? sin (? / 2)
1? ). - Since the global phase of ?? has no observable
effect, we may write ?? cos(?/2) 0? e j?
sin(?/2) 1?. - The angles ? and ? define a point on the surface
of a unit sphere the Bloch sphere, see Fig. 1. - The Bloch sphere provides an excellent tool to
visualize the state vector of a qubit. - This is a binary Bloch sphere, but a multi-valued
counterpart of it can be also created.
100Figure 1. Bloch Sphere with 6 values shown
101- The identity matrix and three Pauli matrices
- form a basis for the 2x2 density matrices.
- So every density matrix can be written as p ½
(I ax X ay Y az Z). - We associate with every 1-qubit state p ½ (I
ax X ay Y az Z) the vector (ax, ay, az). If p
?? ?? for a state ?? e j? (cos (? / 2)
0? e j ? sin (? / 2) 1? ). - Then the corresponding vector is (ax, ay, az)
(sin ? cos ?, sin ? sin ? , cos ?). - It can be easily derived that the vectors (ax,
ay, az) satisfy ax2 ay2 az2 1, which
means that all pure states are located on the
surface of the Bloch Sphere. - When there many identical quantum circuits
working together they are described by density
matrices and the (mixed) states may lay inside
the sphere, not on the surface
102One way to realize multi-valued logic using
binary quantum computing.
- Figure shows the location of 6 points, that may
correspond to values of some multi-valued
algebras. - For binary logic we use 0? and 1?.
- For quaternary logic we use 0?, 1?, 0?1?,
and 0?-1?. - For 6-valued logic we may use additionally 0?
j 1? and 0? - j1?. - A rotation or a combination of rotations leads
from one value to any other value.
103Important quantum gates
- Because global phase does not count, the T gate
can be also written as follows - T (?/8) .
- H denotes the important Hadamard gate
104- The Hadamard and the ?/8 gate can be used to
approximate any given single-qubit unitary
operation with arbitrary accuracy. - On the Bloch sphere, T and HTH are rotations by
an angle ?/4 radians around the z- and x-axes,
respectively. - The composition of these two operations gives a
rotation by an angle ?, which is defined by
cos?/2 cos2?/8, around an axis n, which is
defined by n (cos ?/8, sin ?/8, cos?/ 8). - Since ? is irrational, any rotation around the
?-axis can be build, to arbitrary precision, from
T and HTH. - Furthermore, since for ? arbitrary H R n (?) H
R m (?) with m (cos?/8, - sin?/8, cos?/8 ) not
collinear n, there are angles ?, ? , ? such that
any given U can be written U Rn(?) Rm (?) R
n(?). - It can be also shown that any given 2-qubit gate
can be composed from CNOT and a single qubit
gate.
105- Similarly other universal sets of 1-qubit gates
can be found and illustrated using Bloch Sphere. - This sphere is also useful to find operator
identities (quantum generalizations of rules like
Not (Not B) B ) which play fundamental role
in quantum circuit optimization. - Study of universality and power as well as
quantum realization costs of these gates are
still active research areas. - More study should be devoted to multi-valued
Bloch Sphere, operators in it and their
transformations and realization.
106- Above we showed how multiple-valued logic can be
encoded in binary quantum computing. - Quaternary logic requires two binary measurements
(readings). - The first reading distinguishes states 0? and
1?, and the second reading uses additional
rotation gates to distinguish between states
0?1?, and 0?-1?. - It can be shown that the logic with 2n values
requires n readings. - Another approach to multi-valued quantum circuits
requires measurements with more than two basis
states. - Also, new gates should be defined as well as the
synthesis methods for these gates.
107- While several books and numerous papers have
been published on binary quantum circuits
45,72,103 not much information on their
multi-valued counterparts is available. In their
pioneering paper, - Muthukrishnan and Stroud 68 developed in 2000
multi-valued logic for multi-level quantum
computing systems and showed their realizability
in linear ion trap devices. - However, no experimental data are known so far.
In addition, this approach generates circuits
that are too large and no procedure was proposed
to minimize them. - In 2002, Brylinski and Brylinski 13 discussed
the universality of n-qudit gates without giving
any design algorithms. - Since 2001, PQLG group 2-5,51-55 proposed
Galois Field approach to multi-valued quantum
logic synthesis in several regular structures. - They used gates were ternary counterparts of
classical binary Feynman and Toffoli gates. De
Vos 23 proposed two ternary 11 gates and two
ternary 22 gates, but again no synthesis method
was proposed. - In 2002, Perkowski, Al-Rabadi, and Kerntopf 75
proposed a 22 Generalized Ternary Gate (GTG
gate) based on the ternary conditional gate 68
and ternary shift gates 52-54 and showed the
realization of ternary Toffoli gate using GTG
gates. This work introduced for the first time
the practical realizability of Galois Field
circuits in realizable multi-valued quantum
technology.
108Research Challenges on MV quantum
- There are very few papers on
- realization of multiple-valued quantum circuits,
- design of practical MV quantum circuits,
- algorithms using MV quantum circuits,
- Quantum Computational Learning based on MV logic
- No known work on
- testing,
- simulation and
- algorithms for multiple-valued quantum circuit
exist and - Develop respective theories and QDA tools.
- Develop Binary-encoded model of MV quantum
computing. - Develop truly multi-valued quantum model of
multi-valued computing.
109Quantum Circuit Simulation
- Simulation of quantum circuits plays absolutely
fundamental role in many areas of quantum physics
and engineering. - Similarly as in classical circuits design,
simulation is used to verify correctness of the
design, analyze its properties and find some
interesting aspects that cannot be found by hand
and pencil methods. - It is amazing that the first quantum algorithms
were invented without quantum simulators, but now
the researchers routinely use quantum simulators
to help them with inventions and verify their
design guesses. - Quantum simulators are used to simulate a good
circuit and a circuit with inserted faults, for
test generation and fault localization.
110- Moreover, because the search-based synthesis
methods for quantum circuits such as exhaustive
search, genetic algorithms, genetic programming,
simulated annealing or heuristic search do not
use deeper knowledge of circuit structure and
properties, simulation is the only way (to be
used as a part of the fitness function) to direct
the search towards a circuit that satisfies the
given requirements. - The results of the simulation are compared with
the circuit specification many times in the loop
of the search program. - The same is true for quantum fault simulation.
- As we see, in all these applications the
simulation of quantum circuits must be very fast
and the computer memory should be large. - On the other hand, matrix operations on unitary
matrices are slow, thus new methods and
representations should be found to allow for very
fast and low in memory usage simulation. - This is attempted to be achieved by two
fundamental methods - (1) acceleration of standard operations by using
special hardware emulators, parallel computers or
processor networks 71,73, - (2) creating new advanced data structures to
represent quantum data more efficiently using
standard computers.
111Quantum Decision Diagrams
- New data structures, such as QUIDDs Viamontes,
Markov, Hayes allow for implicit parallelism
when executing Kronecker multiplications on them.
- QUIDDs are based on ADDs and MTBDDs,
- so hopefully in future other decision diagrams
may be used to represent quantum circuits. - It is also expected that basic software engines
used successfully in classical CAD (such as for
instance SAT or ATPG methods) may be used to deal
with quantum circuits. - Also, the fast simulators based on new types of
decision diagrams should be in future
parallelized and possibly accelerated in
FPGA-based boards. - Even before quantum computers will be available,
their emulations on standard computers and
ASIC/FPGA may prove useful to solve some
practical problems.
112Multi-valued Quantum Circuit Synthesis
- Let us first briefly summarize current results in
binary quantum circuit synthesis. - This is the most advanced research area and there
are two gate models for synthesis (especially for
permutative circuits) - (1) The first gate model assumes that only
gates with limited number of inputs can be used
(for instance universal Toffoli3 gate that
operates on three qubits Pa, QB, Rab?c). - We will call it the limited qubit gate model.
- Observe that while in binary reversible logic all
2-bit gates are linear and thus cannot be
universal, in quantum logic there are very many
universal 2-qubit gates (theoretically infinite).
- They can be all used in the limited qubit gate
model, but there are no constructive methods yet
to make use of this fact even for binary case.
113Multi-valued Quantum Circuit Synthesis
- (2) The second gate model assumes that for any
given number of qubits N for which a function is
realized, there exist a Toffoli gate ToffoliN (or
a similar universal gate in which one data qubit
is controlled by more than 2 control qubits) that
operates on N qubits. - We will call it the unlimited qubit gate model.
- In the first model it was proved by Shende et al
that every N-qubit reversible function which is
represented by an even number of cycles, is
realizable without constant wires (ancilla bits)
and every N-qubit function that is represented by
an odd number of cycles is realizable with N1
wires (one ancilla bit).
114- (Observe that every permutation matrix specifies
the permutation of input/output minterms, so it
is a permutation and can be described as a set of
cycles of minterm numbers. - Ancilla bits are also called constant inputs,
dummy variables or input garbages). - In general, synthesis using this model is more
difficult, but the results are closer to the
minimum. - In the second model every function is realizable,
regardless its cycles number. - But it is at the cost of expensive and not
necessarily quantum realizable gates (such gates
may require many ancilla bits internally, so they
tend to hide the high cost of realizations
obtained by the methods 27,28,65.) - Otherwise, there are methods to realize these
complex gates with small ancilla, but for large N
the realization of each complex gate necessitates
an exhaustive number of limited-qubit realizable
gates. - The model (2) should be thus combined with
post-processing methods based on local peephole
optimization. - So far, not much comparisons between these
various synthesis models, especially for real
quantum realizable gates, have been done.
115Two ways to synthesize permutative circuits
- The permutative quantum circuit synthesis
problems are formulated in two ways - (a) A complete reversible function is specified
(as a one-to-one mapping, set of permutation
cycles, or as a unitary matrix) - (b) A irreversible single or multi-output
function is specified. - Some subset of input signals should be returned
unmodified as the output signals. - The final circuit, together with its constant
inputs and garbage outputs should be reversible. - A special case of this model is a controlled gate
where all inputs except one have to be
reconstructed on the output and there is no
ancilla bits. - Usually however this model requires M ancilla
bits, as many as the original outputs of the
specification function, one for every output. - In some cases the number of ancilla bits can be
smaller than M.
116- The first method is more elegant and does not
create garbage. - It is restricted in that it assumes that a
Boolean function has been already converted to a
reversible one (by appropriate adding of ancilla
bits). - For some formulations (like evolutionary
programming and search) this method allows to be
easily extended to non-permutative unitary
matrices. - So far, however, only small circuits can be
synthesized using this method, even using very
advanced algebraic and group-theoretic methods to
decompose a larger matrix to a composition of
smaller matrices. - Because of its formulation, the second way is
more similar to traditional logic synthesis. - Methods developed previously for ESOPs, GFSOPs
and similar forms in the AND/XOR logic synthesis
are used for larger circuits, rather than methods
specific to reversible design.
117 What can we do?
118Quantum Computers
- Our community should should develop a systematic
methodology and CAD tools for synthesizing,
verifying, testing and simulating of quantum
computers. - These methods and tools will be counterparts of
what exists now in binary CMOS. - Development of these tools will require
understanding of real quantum circuit technology.
119New Frontiers
120Open Problems in Quantum Circuits
- Synthesis of binary quantum cascades with no
garbage or small garbage - (Maslov, Dueck, Miller, Perkowski, Khlopotine,
Mishchenko, Curtis, Khan, Jha and Agrawal, Hayes,
Markov) - Synthesis of multiple-valued quantum cascades
- (Muthukrishnan and Stroud, Miller et al, Khan,
Perkowski, Curtis, Lee, Denler) - Universal gates, what are the counterparts of
Toffoli and Fredkin gates?
Fredkin
Toffoli
121Open Problems in Quantum Circuits
- What is the Fault Model for quantum circuits?
- Technology dependent?
- Formal Verification of quantum circuits
- Test Generation for quantum circuits
- Fault Localization of quantum circuits
- Synthesis of testable quantum circuits
- Synthesis of fault-tolerant, error correcting
quantum circuits.
122Open Problems in Quantum Circuits
- What are universal gates?
- How to calculate costs of elementary gates for
each quantum technology such as NMR or ion trap? - What are the gates that can be truly realized in
a quantum technology? - What are the synthesis, analysis and test methods
for sequential quantum circuits?
123Open Problems in Quantum Circuits
- Invent new quantum algorithms.
- What are the principles to create quantum
algorithms - The nature of entanglement.
- Quantum computer architectures.
- Quantum formalisms. (Clifford algebras).
- Quantum Logic.
124Research Challenges
- This adapted approach allows now to realize
larger functions than the approach from (a), but
when applied to multi-output functions usually
leads to high garbage (one ancilla bit for each
output). - In the long run, perhaps this kind of methods
will be better scalable since they use the
structure of the function rather than relying on
heuristic search methods, especially that there
are no strong heuristics known so far. - Finding structure in problems and finding good
heuristics are the interrelated problems for
future research, which will perhaps combine both
ways (a) and (b). - The problem of optimal conversion from
irreversible to reversible function has been not
solved yet.
125Four Synthesis Models
- There exist the following synthesis models, both
for binary and multiple-valued logic - limited qubit gate model and full reversible
function (way a). Usually zero or one ancilla
bits are expected. - unlimited qubit gates and full reversible
function (way a). Usually zero or one ancilla
bits are expected. - limited qubit gates and single output function
(way b). Usually at most M ancilla bits are
expected. - unlimited qubit gates and irreversible input
function (way b). Usually at most M ancilla bits
are expected.
126- Comparing to binary quantum circuit synthesis,
multiple-valued quantum circuit synthesis is a
relatively immature area of research. - One can expect that it will repeat the history of
development of algorithms in binary reversible
logic. - In binary, model (1) has been developed in 84.
- As related to multiple-valued quantum circuits,
the model (1) of reversible quantum circuits
synthesis above has been investigated by 20 and
by a Genetic Algorithm approach from 54. - Model (2), investigated for binary case in
27,28,63,65,66, has been not yet investigated
for multiple-valued logic (although 78 explains
how it can be done). - Model (3) is researched in paper 55 and some
other preliminary results appear also in 78. - Model (4) has been investigated in
4,50-55,59,60. - It is important to distinguish among these four
models, to avoid unrealistic claims of
superiority of one method over another, since
obtaining solutions in some of these models is
much easier than in the other ones.
127Research Challenges
- Objective comparison of the methods on many large
examples and using standardized benchmarks
should be a topic of further research. - Much work is left to be done in defining new
universal multi-valued quantum gates and the
(partially regular) structures to be build from
them. - Approaches that use known universal gates have
the benefit of prior research (such as logic
synthesis using Galois Field operations), but can
be very costly and inefficient.
128- Below we give a complete characteristics of
papers in multi-valued quantum logic synthesis.
Khan and Perkowski adapted the GFSOP (Galois
Field Sum of Products) method to permutative
(ternary) quantum circuits 52,53. - The algorithm is based on finding a ternary
decision diagram, and flattening it to quantum
cascade-realizable GFSOP. - In another work 54 these authors use Genetic
Algorithm to synthesize multi-output, no-garbage
cascades of arbitrary ternary quantum gates. - The approach presented by Miller et al 65 is an
extension of their greedy algorithm for binary
circuits 27,28,63. A non-published extension to
their work presents also a method to encode
ternary logic using standard binary qubits 66.
Observe that while binary quantum logic uses 1800
rotation, and the quaternary logic from 49 uses
900 rotations, they use 1200 rotations for one
vertical plane of Bloch Sphere in ternary logic.
While both ternary and quaternary model use two
measurements to distinguish encoded signals, the
quaternary method is more efficient. A paper 49
based on SAT and reachability analysis uses
quaternary quantum logic to synthesize exact
minimum binary circuits from Feynman, Inverter,
Controlled-V and Controlled-V gates. (V is
called a square-root-of-NOT since its repeated
application negates the input signal, V V NOT).
A simple adaptation of this method allows to
realize also quaternary quantum circuits with
arbitrary input and output signals 78.
129Research Challenges
- Recent works suggest that many uniform general
methods can be created to realize various
multiple-valued logics that will use generalized
rotations with respect to 3 orthogonal basis
axes, rotations by angles 2?/k, where kgt1 is a
natural number. - In general, rotations with respect to any axis n
can be used, but using some of Z, X, and Y
simplifies gates. - Every existing algorithm for binary quantum
circuit design can be extended to its various
multiple-valued quantum counterparts, but these
generalizations are not trivial and algorithms
that use these gates are numerically very
challenging. - These problems form then a good base for new
research by people who understand search-based
EDA algorithms and multiple-valued logic.
130Figure 2. 33 Toffoli gate
- Figure 2 presents a standard binary reversible
Toffoli gate. - Its ternary counterpart has Galois Field 2
operations of multiplication and addition
replaced with Galois Field(3) operations.
131- Observe that the internal structure of this gate
is complex when using quantum realizable gates
(Figure 3). The Controlled-V gate works like
this when the control (top) signal is 0gt, the
data input is forwarded to output with no change.
When the control signal is 1gt the operation of
the lower box (so-called V) is executed. In our
case this is a square-root-of-NOT operation. Thus
if two Controlled-V gates in series are
controlled by the same signal A, if A1 then
their qubit data line is a negation. Two such
gates in series serve then as a controlled-NOT or
Feynman gate. Also, the operation of V and V
annihilate ( V V I ) . The reader can simulate
by hand the circuit from Figure 3a to see that
it truly realizes the Toffoli3 gate. Let us
observe that the circuit from Figure 3a can be
redrawn to one from Figure 3b. This circuit
emphasizes that both CNOT, CV and C V are
Controlled-Gates that leave data signal unchanged
when the control is 0gt and apply its internal
transformation (the symbol of this transformation
is in the input to multiplexer) when the control
is 1gt.
132- Observe that the internal structure of this gate
is complex when using quantum realizable gates
(Figure 3). - The Controlled-V gate works like this when the
control (top) signal is 0gt, the data input is
forwarded to output with no change. - When the control signal is 1gt the operation of
the lower box (so-called V) is executed. - In our case this is a square-root-of-NOT
operation. - Thus if two Controlled-V gates in series are
controlled by the same signal A, if A1 then
their qubit data line is a negation. - Two such gates in series serve then as a
controlled-NOT or Feynman gate. Also, the
operation of V and V annihilate ( V V I ) . - The reader can simulate by hand the circuit
from Figure 3a to see that it truly realizes the
Toffoli3 gate. - Let us observe that the circuit from Figure 3a
can be redrawn to one from Figure 3b. - This circuit emphasizes that both CNOT, CV and C
V are Controlled-Gates that leave data signal
unchanged when the control is 0gt and apply its
internal transformation (the symbol of this
transformation is in the input to multiplexer)
when the control is 1gt.
133- Observe that any single-qubit operation can be
written in the box, and also that any single
qubit operation can be inserted to the control
and data lines. - The control can be from top (as in the Figure 3b)
or from the bottom. - The composition of this kind of multiplexed
operations allows to create arbitrary permutative
gate of reversible logic 55,59. - Also, an arbitrary two-qubit quantum gate
(described by a unitary matrix) can be
constructed from such gates. - These methods can be used to hierarchically
synthesize larger circuits 55,59 and can be
generalized to ternary (or in general
multi-valued) logic (see Figure 4) for the
realization of universal ternary permutative
controlled gate. - Universal quantum gate is created when operations
are single-qubit ternary rotations.
134- Figure 3. Smolin/DiVincenzo realization of
Toffoli gate as a prototype of a regular
controlled quantum structure (a) standard
notation, (b) notation used in this paper to
emphasize the similarity
- Observe that any single-qubit operation can be
written in the box, and also that any single
qubit operation can be inserted to the control
and data lines. - The control can be from top (as in the Figure 3b)
or from the bottom. - The composition of this kind of multiplexed
operations allows to create arbitrary permutative
gate of reversible logic 55,59. - Also, an arbitrary two-qubit quantum gate
(described by a unitary matrix) can be
constructed from such gates.
135- Also, an arbitrary two-qubit quantum gate
(described by a unitary matrix) can be
constructed from such gates. - These methods can be used to hierarchically
synthesize larger circuits 55,59 and can be
generalized to ternary (or in general
multi-valued) logic (see Figure 4) for the
realization of universal ternary permutative
controlled gate. - Universal quantum gate is created when operations
are single-qubit ternary rotations
- Figure 4 Conceptual ternary multiplexer
- op Logical Operations
- 0, 1, 2 represent Galois Addition of constants
0, 1, and 2, respectively - 01, 02, 12 represent logical replacement i.e. a
01 operation will replace 0-gt1, 1-gt0, and 2-gt2
136Other problems in MV QC
- New models of gates, such as above, that will be
close to realization and at the same time would
allow creation of efficient synthesis algorithms,
also for large circuits. - Development of methods based on unitary matrix
decomposition, group theory, Lie groups and
Clifford algebras, - Methods for incompletely specified functions, to
be used in machine learning and data mining, - Geometrical and topological visualization methods
to help intuition of designers to design
multi-qubit circuits (for instance
generalizations of Bloch sphere, QUIDDs and
Karnaugh Maps), - Efficient methods for local optimization of
quantum circuits on many levels of description, - High-level quantum hardware description languages
that will play in QDA a role similar to VHDL and
Verilog in EDA, - Development of formalisms and synthesis methods
for sequential circuits.
137Testing and diagnosis of quantum circuits
- Patel, Markov, and Hayes showed that reversible
circuits are much better testable than
irreversible circuits. - This is because every test covers half faults and
every fault is covered by half tests. - The reversible circuits are then transparent to
faults, making them well observable and
controllable. - We showed that fault localization in reversible
circuits is easier. - We presented preliminary results on testing
binary quantum circuits and on fault localization
of quantum circuits.
138Testing Quantum Circuits (1)
- The good circuit is simulated.
- Next every possible quantum fault is inserted
(our fault model is inserting arbitrary matrix in
place of fault, this allows to simulate many
different types of faults) and the circuit with
fault is simulated in Hilbert space (no
measurement). - All possible measurement values are calculated
with their probabilities. - The comparison of a measurement from the unitary
matrix of a correct circuit and a circuit with
fault determines which input combinations (tests)
give different measurements. - In some cases the circuit is modified for
multi-valued realization in order to distinguish
the values.
139Testing Quantum Circuits (2)
- Observe that in contrast to standard testing and
reversible circuits testing, there are three
types of faults in quantum domain - (1) faults that can be detected
deterministically, - (2) faults that cannot be detected (like global
phase faults), and - (3) faults that can be detected by repeated
application of tests, possibly with special
measuring gates. - These faults can be detected only with certain
probability. - Thus, quantum testing is probabilistic testing.
140Research Challenges in Quantum Test
- Open problems include basically everything
- fault models,
- fault simulation,
- test generation,
- test minimization,
- fault coverage,
- fault localization using probabilistic adaptive
trees.
141Quantum Computational Intelligence (QCI)
- The two most famous quantum algorithms to date
were created by Peter Shor and Lov Grover. - Shors algorithm is for factoring integers
- It produces an exponential computational speedup
over classical algorithms - It can break the RSA encryption techniques.
- Grovers algorithm searches an unordered list of
data, to find a particular item. - It has a provable quadratic speedup over the best
classical algorithm. - It is like looking for name of a person in yellow
pages knowing only his telephone number.
142Research Challenges in Quantum Algorithms for
Computational Intelligence
- How these algorithms can be used in the field of
Computational Intelligence?. - Quantum computing is in every particular instance
at least as powerful