Quantum Quantum Architectures Architectures June 1, 2005 June 1, - - PowerPoint PPT Presentation
Quantum Quantum Architectures Architectures June 1, 2005 June 1, - - PowerPoint PPT Presentation
Quantum Quantum Architectures Architectures June 1, 2005 June 1, 2005 Computing? Computing? e of computing devices that exhibit quantu e of computing devices that exhibit quantu chanical behavior chanical behavior ehavior of isolated
Computing? Computing?
e of computing devices that exhibit quantu e of computing devices that exhibit quantu chanical behavior chanical behavior
ehavior of isolated ions ehavior of isolated ions Bose Bose-
- Einstein condensate in a magnetic well
Einstein condensate in a magnetic well hoton interactions hoton interactions ire and transistor interactions in the near future ire and transistor interactions in the near future
e computing model is largely unexplored computing model is largely unexplored
⊆ ⊆? QP ? QP ⊆ ⊆? NP ? NP
e usage model is still being debated usage model is still being debated
hy are Architects Involved hy are Architects Involved
e know what the “killer app” will be e know what the “killer app” will be
Error correction will be >99% of the work Error correction will be >99% of the work
e physicists don’t know computation e physicists don’t know computation
“Don’t worry, it’s polynomial...” “Don’t worry, it’s polynomial...”
e theorists don’t know physics e theorists don’t know physics
“Simplify the problem by removing communication...” “Simplify the problem by removing communication...”
short, architects can be the reality check short, architects can be the reality check
Identify physical bounds that supersede theoretical on Identify physical bounds that supersede theoretical on Determine what aspects of computation will be the m Determine what aspects of computation will be the mo challenging challenging
Outline Outline
hat makes quantum different? hat makes quantum different?
Quantum bits Quantum bits Operations and measurement Operations and measurement Decoherence Decoherence
hat will a quantum computer look like hat will a quantum computer look like uantum architecture research at UW uantum architecture research at UW
Current research: Simulation Current research: Simulation Building quantum wires Building quantum wires
Classical vs. Quantum Classical vs. Quantum
sic element: 0 or 1 sic element: 0 or 1 ts are independent ts are independent ata may be copied ata may be copied destroyed at will destroyed at will ata is static ata is static
n n Bits are continuous va
Bits are continuous va
n n Bits may be entangled
Bits may be entangled interfere interfere
n n Data may not be copie
Data may not be copie
n n Operations must be
Operations must be reversible reversible
n n Data
Data decoheres decoheres with t with t
Qubits Qubits: Quantum Bits : Quantum Bits
and and qubits qubits both have two states: 0 an both have two states: 0 an perposition: perposition:
ubit ubit may be in both states simultaneously may be in both states simultaneously
ase: ase:
ubit ubit may have a negative quantity of a st may have a negative quantity of a st
uantum States and Measuremen uantum States and Measuremen
e e qubit qubit probabilistically represents two states probabilistically represents two states a |0> + b |1> a |0> + b |1> ch additional ch additional quibit quibit doubles the number of sta doubles the number of sta a |00> + b |01> +c |10> + d |11> a |00> + b |01> +c |10> + d |11> easurement sends a easurement sends a qubit qubit into a classical stat into a classical stat is may alter the states of other is may alter the states of other qubits qubits
a |00> + b |11> (EPR state) a |00> + b |11> (EPR state)
easurement and Copy Protectio easurement and Copy Protectio
uantum data cannot be copied uantum data cannot be copied
Copying involves a read and a write Copying involves a read and a write “Reading” destroys the state “Reading” destroys the state
uantum data can be transferred uantum data can be transferred
One One qubit qubit can swap its state with another can swap its state with another Quantum state can be “teleported” over infinite Quantum state can be “teleported” over infinite distance (but the sender loses the data) distance (but the sender loses the data)
any quantum algorithms are probabilistic any quantum algorithms are probabilistic d involve iteration d involve iteration
Other Operations Other Operations
easurement: the only irreversible operati easurement: the only irreversible operati l other operations are reversible l other operations are reversible
2nd Law: Reversible operations conserve ener 2nd Law: Reversible operations conserve ener “Not” is reversible “Not” is reversible “And” and “Or” are not reversible “And” and “Or” are not reversible
- st “traditional” operations must produce
- st “traditional” operations must produce
ditional output ditional output
How do you make “+” reversible? How do you make “+” reversible? Scratch bits must be protected Scratch bits must be protected
Noise: Noise: Decoherence Decoherence
eoretical systems are “closed” eoretical systems are “closed”
No energy may enter the system unless explicitly No energy may enter the system unless explicitly introduced introduced
al systems can’t be completely isolated al systems can’t be completely isolated
Performing an op adds “noise” to a Performing an op adds “noise” to a qubit qubit Over time, Over time, qubits qubits will simply “ will simply “decohere decohere” ” At higher temperatures (higher energy), At higher temperatures (higher energy), decoherence decoherence
- ccurs more quickly
- ccurs more quickly
e will need massive cooling systems e will need massive cooling systems y usable quantum system will require error y usable quantum system will require error
rror Correction is Crucial rror Correction is Crucial
ror correcting codes are available ror correcting codes are available
Operations exist for computing on encoded dat Operations exist for computing on encoded data Since Since qubits qubits cannot be read, an error correction cannot be read, an error correction routine manipulates the coded bits to fix errors routine manipulates the coded bits to fix errors
ror correction must often be applied ror correction must often be applied cursively to error correction routines cursively to error correction routines e Threshold Theorem e Threshold Theorem
An error rate of 10 An error rate of 10-
- 4
4 per op can be tolerated
per op can be tolerated This error rate requires nearly continuous error This error rate requires nearly continuous error correction correction
To Summarize... To Summarize...
uantum bits are awesome uantum bits are awesome
...except that no one has really done anything ...except that no one has really done anything amazing with them (yet) amazing with them (yet) …because measurement really hurts …because measurement really hurts
- ring and moving data will be difficult
- ring and moving data will be difficult
Decoherence Decoherence limits storage and transit time limits storage and transit time
e know what quantum computers will do e know what quantum computers will do
Error correcting...all the time Error correcting...all the time
rforming computation in the presence of rforming computation in the presence of
Outline Outline
hat makes quantum different? hat makes quantum different?
Quantum bits Quantum bits Operations and measurement Operations and measurement Decoherence Decoherence
hat will a quantum computer look like hat will a quantum computer look like uantum architecture research at UW uantum architecture research at UW
Current research: Simulation Current research: Simulation Building quantum wires Building quantum wires
Pick a Technology Pick a Technology
hat device technology will be used? hat device technology will be used?
Who knows... Who knows...
evelop first order assumptions evelop first order assumptions
Classical control of quantum gates Classical control of quantum gates Silicon to interface and control Silicon to interface and control
n n Provides rough size constraints
Provides rough size constraints
Individual control of quantum bits Individual control of quantum bits
ck a likely technology that fits ck a likely technology that fits
For example, ion traps For example, ion traps
Consider Building Blocks
- nsider Building Blocks
Processor: Computation Processor: Computation
Several sets of universal gates exist Several sets of universal gates exist Different device technologies can more easily implem Different device technologies can more easily implem some gates than others some gates than others Measurement and “zero” creation is also important Measurement and “zero” creation is also important
emory: Storage emory: Storage
Storage is difficult because of Storage is difficult because of decoherence decoherence Constant error correction may be performed Constant error correction may be performed
n n Decoherence
Decoherence-
- free subspaces are being researched
free subspaces are being researched
Hence, memory looks a lot like the processor Hence, memory looks a lot like the processor
Interconnect: Communication Interconnect: Communication
reger reger-
- Stickles and
Stickles and Balensiefer Balensiefer
e computer is a grid of traps and wires e computer is a grid of traps and wires ch trap has a flexible gate that can perform ch trap has a flexible gate that can perform easurement or a quantum op easurement or a quantum op mmunication is performed by moving ions mmunication is performed by moving ions
Outline Outline
hat makes quantum different? hat makes quantum different?
Quantum bits Quantum bits Operations and measurement Operations and measurement Decoherence Decoherence
hat will a quantum computer look like hat will a quantum computer look like uantum architecture research at UW uantum architecture research at UW
Current research: Simulation Current research: Simulation Building quantum wires Building quantum wires Representing quantum algorithms Representing quantum algorithms
mulating Quantum Compute ulating Quantum Compute
nt work by nt work by Kreger Kreger-
- Stickles,
Stickles, Balensiefer Balensiefer, and O , and O compare architectures, performance data is n compare architectures, performance data is n ulation is the only choice ulation is the only choice
Fully” simulating a quantum system takes exponential Fully” simulating a quantum system takes exponential y modeling only reliability, can be done in linear time y modeling only reliability, can be done in linear time
e Simulation Infrastructu e Simulation Infrastructur
n error correction compiler error correction compiler
Takes a quantum algorithm for one specific inpu Takes a quantum algorithm for one specific inpu Produces a fault tolerant version of the algorithm Produces a fault tolerant version of the algorithm
device scheduler device scheduler
Takes a program and an architecture Takes a program and an architecture Produces a static operation and communication Produces a static operation and communication schedule schedule
reliability simulator reliability simulator
Takes a scheduled application and architecture Takes a scheduled application and architecture Produces performance and reliability metrics Produces performance and reliability metrics
Preliminary Results Preliminary Results
ror correction is less effective than we assum ror correction is less effective than we assum
Correcting errors requires that bits be moved Correcting errors requires that bits be moved
e Threshold Theorem is optimistic e Threshold Theorem is optimistic
Realistic execution constraints (communication) limit Realistic execution constraints (communication) limit acceptable error rate acceptable error rate The realistic threshold is four orders of magnitude wo The realistic threshold is four orders of magnitude wo
antum computer size will be limited by the er antum computer size will be limited by the er te of the underlying technology te of the underlying technology
The further bits have to move, the lower the rate has The further bits have to move, the lower the rate has t be be
ilding a Quantum Wire (Oski ilding a Quantum Wire (Oski
get technology is Kane’s silicon ion get technology is Kane’s silicon ion-
- tra
tra
- ns are embedded into silicon traps
- ns are embedded into silicon traps
- Spin of 31P holds quantum state
Spin of 31P holds quantum state
- Ions spaced
Ions spaced ≤ ≤ 20nm apart for quantum effect to o 20nm apart for quantum effect to o
- ≤
≤ 1.5Kelvin for reasonable coherence time 1.5Kelvin for reasonable coherence time
- cal magnetic field arbitrates gates
- cal magnetic field arbitrates gates
- Controlled by “classical” pins
Controlled by “classical” pins
- Driven by high frequency (10
Driven by high frequency (10-
- 100Mhz) clock
100Mhz) clock
- Gated by “lower” frequency (0.01
Gated by “lower” frequency (0.01 – – 10Mhz) clock 10Mhz) clock
A A Short Short Quantum Wire Quantum Wire
nstructed from swap gates nstructed from swap gates
Unless the particle that holds the quantum state physically moves, the information “flows” in discrete steps from particle to particle. Each step requires 3 quantum controlled-not operations, effectively performing a “swap” of the quantum states.
rchitecture of a Long Wir rchitecture of a Long Wire
EPR Generator
Teleporation Unit Teleporation Unit Entropy Exchange Purification Coded Tele- Portation
Classical control channel Quantum EPR channel R channel
Two Possible Wires Two Possible Wires
- rt range: swapping
- rt range: swapping-
- channel
channel
“Pitch matching” causes structural concerns “Pitch matching” causes structural concerns
n n Qubits
Qubits are 20nm apart, so control is limited to 5nm are 20nm apart, so control is limited to 5nm
n n @ 1.5 Kelvin, cannot drive an AC current
@ 1.5 Kelvin, cannot drive an AC current
n n Other dimensions must be increased to > 100nm
Other dimensions must be increased to > 100nm
Length is limited due to Length is limited due to decoherence decoherence
ng range: teleportation ng range: teleportation-
- channel
channel
Length is arbitrary Length is arbitrary Many additional structures are required Many additional structures are required Bandwidth is constrained by EPR pair productio Bandwidth is constrained by EPR pair productio
presenting Quantum Computati resenting Quantum Computati
lop an alternative notation for quantum computing lop an alternative notation for quantum computing
presentation: dealing with groups of bits is ha presentation: dealing with groups of bits is ha
nsure operations are insensitive to state space size nsure operations are insensitive to state space size troduce shorthand for common entangled states troduce shorthand for common entangled states acilitate computation on large, highly entangled states acilitate computation on large, highly entangled states
asoning: interesting states are difficult to iden asoning: interesting states are difficult to iden
entify quantum properties explicitly entify quantum properties explicitly efine operations by the quantum properties they induc efine operations by the quantum properties they induc avor local transformations over global ones avor local transformations over global ones
An Example: EPR Pairs An Example: EPR Pairs
atrix notation:
H(x) → x 0 : x 0 + x1 || x1 : x 0 CNot(x,y) → x 0 : x 0y || x1 : x qubit p ⇒ p0 qubit q ⇒ q0 H(p) on p0 ⇒ p0 + p1 CNot(p,q) on (p0 + p1)q0
= = ⊗ = − 1 2 1 1 1 2 1 1 1 1 2 1 1 1 , 1 1 2 1 1 1 1 1
In the algebra:
The High Order Bits The High Order Bits
uantum architecture is a wide uantum architecture is a wide-
- open field
- pen field
We can’t even agree how to build wires! We can’t even agree how to build wires!
chitects have an important role chitects have an important role
We act as intermediaries between the physicist We act as intermediaries between the physicist and algorithm designers and algorithm designers