Towards practical classical processing for the surface code arXiv:1110.5133 Austin G. Fowler, Adam C. Whiteside, Lloyd C. L. Hollenberg Centre for Quantum.

Slides:



Advertisements
Similar presentations
Artificial Intelligence 12. Two Layer ANNs
Advertisements

Q U RE: T HE Q UANTUM R ESOURCE E STIMATOR T OOLBOX Martin Suchara (IBM Research) October 9, 2013 In collaboration with: Arvin Faruque, Ching-Yi Lai, Gerardo.
University of Queensland
Sergey Bravyi, IBM Watson Center Robert Raussendorf, Perimeter Institute Perugia July 16, 2007 Exactly solvable models of statistical physics: applications.
Lindsey Bleimes Charlie Garrod Adam Meyerson
Slides from: Doug Gray, David Poole
Resource Management §A resource can be a logical, such as a shared file, or physical, such as a CPU (a node of the distributed system). One of the functions.
Efficient access to TIN Regular square grid TIN Efficient access to TIN Let q := (x, y) be a point. We want to estimate an elevation at a point q: 1. should.
Lecture 24 Coping with NPC and Unsolvable problems. When a problem is unsolvable, that's generally very bad news: it means there is no general algorithm.
Limits on ILP. Achieving Parallelism Techniques – Scoreboarding / Tomasulo’s Algorithm – Pipelining – Speculation – Branch Prediction But how much more.
1 Complexity of Network Synchronization Raeda Naamnieh.
Filling Arbitrary Holes in Finite Element Models 17 th International Meshing Roundtable 2008 Schilling, Bidmon, Sommer, and Ertl.
University of Queensland
Local Fault-tolerant Quantum Computation Krysta Svore Columbia University FTQC 29 August 2005 Collaborators: Barbara Terhal and David DiVincenzo, IBM quant-ph/
Beyond the DiVincenzo Criteria: Requirements and Desiderata for Fault-Tolerance Daniel Gottesman.
CSE 421 Algorithms Richard Anderson Lecture 4. What does it mean for an algorithm to be efficient?
Lecture 11. Matching A set of edges which do not share a vertex is a matching. Application: Wireless Networks may consist of nodes with single radios,
Memory Hierarchies for Quantum Data Dean Copsey, Mark Oskin, Frederic T. Chong, Isaac Chaung and Khaled Abdel-Ghaffar Presented by Greg Gerou.
Randomness in Computation and Communication Part 1: Randomized algorithms Lap Chi Lau CSE CUHK.
Route Planning Vehicle navigation systems, Dijkstra’s algorithm, bidirectional search, transit-node routing.
Fault-Tolerant Quantum Computation in Multi-Qubit Block Codes Todd A. Brun University of Southern California QEC 2014, Zurich, Switzerland With Ching-Yi.
Lecture 11. Matching A set of edges which do not share a vertex is a matching. Application: Wireless Networks may consist of nodes with single radios,
From Essentials of Computer Architecture by Douglas E. Comer. ISBN © 2005 Pearson Education, Inc. All rights reserved. 7.2 A Central Processor.
A Fault-tolerant Architecture for Quantum Hamiltonian Simulation Guoming Wang Oleg Khainovski.
Requirements and Desiderata for Fault-Tolerant Quantum Computing Daniel Gottesman Perimeter Institute for Theoretical Physics Beyond the DiVincenzo Criteria.
Quantum Error Correction Jian-Wei Pan Lecture Note 9.
Lecture 23. Greedy Algorithms
Paraty - II Quantum Information Workshop 11/09/2009 Fault-Tolerant Computing with Biased-Noise Superconducting Qubits Frederico Brito Collaborators: P.
Quantum Error Correction and Fault-Tolerance Todd A. Brun, Daniel A. Lidar, Ben Reichardt, Paolo Zanardi University of Southern California.
Surface Simplification Using Quadric Error Metrics Michael Garland Paul S. Heckbert.
COMPUTER AIDED DESIGN -(CAD)-3
Network Aware Resource Allocation in Distributed Clouds.
Quantum Information Jan Guzowski. Universal Quantum Computers are Only Years Away From David’s Deutsch weblog: „For a long time my standard answer to.
Graph Algorithms. Definitions and Representation An undirected graph G is a pair (V,E), where V is a finite set of points called vertices and E is a finite.
IT253: Computer Organization
05/04/06 1 Integrating Logic Synthesis, Tech mapping and Retiming Presented by Atchuthan Perinkulam Based on the above paper by A. Mishchenko et al, UCAL.
Digital Image Processing CCS331 Relationships of Pixel 1.
Michael A. Nielsen Fault-tolerant quantum computation with cluster states School of Physical Sciences Chris Dawson (UQ) Henry Haselgrove (UQ) The University.
Global control, perpetual coupling and the like Easing the experimental burden Simon Benjamin, Oxford. EPSRC. DTI, Royal Soc.
CSC 211 Data Structures Lecture 13
1 Compacting Test Vector Sets via Strategic Use of Implications Kundan Nepal Electrical Engineering Bucknell University Lewisburg, PA Nuno Alves, Jennifer.
Quantum Computing Paola Cappellaro
A new Ad Hoc Positioning System 컴퓨터 공학과 오영준.
Graph Colouring L09: Oct 10. This Lecture Graph coloring is another important problem in graph theory. It also has many applications, including the famous.
An Efficient Linear Time Triple Patterning Solver Haitong Tian Hongbo Zhang Zigang Xiao Martin D.F. Wong ASP-DAC’15.
Mesh Coarsening zhenyu shu Mesh Coarsening Large meshes are commonly used in numerous application area Modern range scanning devices are used.
September 12, 2014 Martin Suchara Andrew Cross Jay Gambetta Supported by ARO W911NF S IMULATING AND C ORRECTING Q UBIT L EAKAGE.
Copyright © 2006 Keio University Distributed Quantum Error Correction First International Conference on Quantum Error Correction December 19, 2007 Rodney.
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual.
CS 484 Load Balancing. Goal: All processors working all the time Efficiency of 1 Distribute the load (work) to meet the goal Two types of load balancing.
CS 147 Virtual Memory Prof. Sin Min Lee Anthony Palladino.
Theory and Practice of Software Testing
Granularity in the Data Warehouse Chapter 4. Raw Estimates The single most important design issue facing the data warehouse developer is determining the.
Capabilities and limitations of quantum computers Michele Mosca 1 November 1999 ECC ’99.
Efficient Placement and Dispatch of Sensors in a Wireless Sensor Network You-Chiun Wang, Chun-Chi Hu, and Yu-Chee Tseng IEEE Transactions on Mobile Computing.
Computational Physics (Lecture 10) PHY4370. Simulation Details To simulate Ising models First step is to choose a lattice. For example, we can us SC,
QUANTUM COMPUTING: Quantum computing is an attempt to unite Quantum mechanics and information science together to achieve next generation computation.
Hamiltonian quantum computer in one dimension
CS 326A: Motion Planning Probabilistic Roadmaps for Path Planning in High-Dimensional Configuration Spaces (1996) L. Kavraki, P. Švestka, J.-C. Latombe,
Visual Inspection Planning with Sensor Constraint Graph
Mean Shift Segmentation
Randomized Algorithms CS648
Analysis of Algorithms
Using surface code experimental output correctly and effectively
The performance requirements for DSP applications continue to grow and the traditional solutions do not adequately address this new challenge Paradigm.
Haitao Wang Utah State University SoCG 2017, Brisbane, Australia
Fast Min-Register Retiming Through Binary Max-Flow
Communication Driven Remapping of Processing Element (PE) in Fault-tolerant NoC-based MPSoCs Chia-Ling Chen, Yen-Hao Chen and TingTing Hwang Department.
Determining the capacity of any quantum computer to perform a useful computation Joel Wallman Quantum Resource Estimation June 22, 2019.
Magic-State Functional Units
Presentation transcript:

Towards practical classical processing for the surface code arXiv: Austin G. Fowler, Adam C. Whiteside, Lloyd C. L. Hollenberg Centre for Quantum Computation and Communication Technology School of Physics, The University of Melbourne, Australia ?

Overview why study the surface code? what is the surface code? classical processing challenges –correcting errors fast enough –interpreting logical measurements complexity optimal error processing –performance in detail –3 second fault-tolerant distance 1000 summary and further work

Why study the surface code? no known parallel arbitrary interaction quantum computer (QC) architecture many 2-D nearest neighbor (NN) architectures –ion traps, superconducting circuits, quantum dots, NV centers, neutral atoms, optical lattices, electrons on helium, etc concatenated codes perform poorly on 2-D NN architectures –Steane, Bacon-Shor p th ~2x10 -5 –distance d = 9 Steane code currently requires 2304 qubits –d = 27 requires 100k+ –nq = d 3.52 surface code performs optimally on a 2-D NN QC surface code has –p th ~10 -2, so lower d required –low overhead implementations of the entire Clifford group –flexible, long-range logical gates –distance 9 surface code currently requires 880 qubits –distance 10 requires 1036 –distance 27 requires 4356 –nq = 6d 2 –lower overhead and 10x higher p th than any other known 2-D topological code The surface code is by far the most promising and practical code currently known.

Surface code: memory a minimum of 13 qubits (13 cells) are required to create a single fault- tolerant surface code memory the nine circles store the logical qubit, the four dots are repeatedly used to detect errors on neighboring qubits an arbitrary single error can be corrected, multiple errors can be corrected provided they are separated in space or time in a full-scale surface code quantum computer, a minimum of 72 qubits are required per logical qubit the error correction circuitry is simple and transversely invariant detect X error detect Z error minimum-size logical qubit scalable logical qubit order of interactions

Single logical qubit gates when using error correction, only a small set of single logical qubit gates are possible for the surface code, these are, in order of difficulty, X, Z, H, S and T X and Z are applied within the classical control software H can be applied transversely after cutting the logical qubit out of the lattice S and T involve the preparation of ancilla states and gate teleportation the ancilla states must be distilled to achieve sufficiently low probability p of error an efficient, non- destructive S circuit exists, removing the need to reinject and redistill  Y  states Hadamard after cutting (stop interacting with qubits outside the green square) state distillation p → 7p 3 gate teleportation S, T, R Z (  ) ancilla state required for S ancilla state required for T non-destructive S circuit

Surface code CNOT there are two types of surface code logical qubit, rough and smooth the simplest CNOT (C X ) has smooth control and rough target C X can be defined by its effect on certain operators given a state XI  = , C X  = C X XI  = C X XIC X C X  = XXC X  XI→XX, IX→IX, ZI→ZI, IZ→ZZ defines C X moving (braiding) the holes (defects) that define logical qubits deforms their associated logical operators, performing computation movement is achieved by dynamically changing which data qubits are error corrected more complex braid patterns are required for rough-rough C X smooth qubit rough qubit smooth-rough CNOT CNOT braid patterns

Long-range CNOT stretching the braid pattern enables arbitrarily long-range CNOT the need to propagate error correction information near each braid prevents violations of relativity Clifford group computations can proceed without waiting for this information this enables CNOTs to be effectively bent backwards in time note that single control, multiple target CNOT is a simple braid pattern arbitrary Clifford group computations can be performed in constant time

Classical processing challenges correcting errors fast enough –a single round of surface code QEC can involve as few as five sequential quantum gates –depending on the underlying technology, this may take less than a microsecond –need to be able to process a complex, infinite size graph problem in a very small constant time interpreting logical measurements –surface code quantum computing is essentially a measurement based scheme –logical gates, including the identity gate, introduce byproduct operators –determining what these byproduct operators are is exceedingly difficult, especially after optimizing a braid pattern

Correcting errors fast enough consider the life cycle of a single error in the bulk of the lattice, a single gate error is always detected at two space-time points given a detailed error model for each gate, the probability that any given pair of space-time points will be connected by a single error can be calculated these probabilities can be represented by two lattices of cylinders recently completed open source tool to perform such error analysis and visualisation detect X error detect Z error

the primal (Z errors) and dual (X errors) lattices are independent deterministically constructed objects terminology: dots and lines weight of line –ln(p line ) stochastically detected errors are represented by vertices associated with specific dots edges between vertices have weight equal to the minimum weight path through the lattice by choosing a minimum weight matching of vertices, corrections can be applied highly likely to preserve the logical state Correcting errors fast enough

minimum weight perfect matching was invented by Jack Edmonds in 1965 very well studied algorithm, however best publicly available implementations have complexity O(n 3 ) for complete graphs don’t support continuous processing don’t support parallel processing actually quite slow Correcting errors fast enough as recently as this year [PRA 83, (R) (2011)], distance ~ 10 was the largest surface code that had been studied fault-tolerantly renormalization techniques have been used to study large non- fault-tolerant surface codes [PRL 104, (2010), arXiv: ] need something much better if surface code quantum computer is to be built

to describe our fast matching algorithm, replace complex 3-D lattice with simple uniform weight 2-D lattice (grey lines) vertices (error chain endpoints) are represented by black dots matched edges are thick black lines shaded regions are space-time locations the algorithm has explored Correcting errors fast enough

Complexity optimal error correction

choose a vertex

Complexity optimal error correction explore local space-time region until other objects encountered

Complexity optimal error correction if unmatched vertices encountered, match with one

Complexity optimal error correction choose another vertex

Complexity optimal error correction expand until other objects are encountered, build alternating tree 1 2

Complexity optimal error correction it alternating tree outer space-time regions can’t be expanded, form blossom 1

Complexity optimal error correction uniformly expand space-time region around blossom until other objects encountered 1

Complexity optimal error correction two options in this case, unmatched vertex or boundary, choose vertex 1

Complexity optimal error correction choose another vertex

Complexity optimal error correction form alternating tree 1 2

Complexity optimal error correction expand outer nodes, contract inner nodes 1 2

Complexity optimal error correction form blossom 1

Complexity optimal error correction grow alternating tree 1 2

Complexity optimal error correction expand outer nodes, contract inner nodes, forbidden region entered 1 2

Complexity optimal error correction undo expand outer, contract inner 1 2

Complexity optimal error correction undo grow alternating tree

Complexity optimal error correction undo form blossom

Complexity optimal error correction undo expand outer, contract inner... done until have additional data

Complexity optimal error correction noteworthy features: –many rules, but each rule is simple and cheap –on average, each vertex only needs local information –fewer vertices = faster, simpler and more local processing –algorithm remains efficient at and above threshold average runtime is O(n) per round of error correction the runtime is independent of the amount of history, which can be discarded after a delay, finite memory required algorithm can be parallelized to O(1) using constant computing resources per unit area (unit width in the example) current implementation is single processor only

Performance in detail we simulate surface codes of the form shown on the right depolarizing noise of strength p is applied after every unitary gate, including the identity measurement reports the wrong eigenstate with probability p but leaves the state in the reported eigenstate for each d and p, our simulations continuously apply error correction until 10,000 logical errors have been observed the probability of logical error per round of error correction is then calculated and plotted

Performance in detail % 0.5%

Performance in detail matching, while still efficient, is much slower around and above the threshold error rate very large blossoms appear at d = 55, p = 10 -2, approximately 6 GB of memory required at an error rate p = 10 -3, it is straightforward to simulate d = 1000 over 4 million qubits 3 seconds per round of error correction approximately 100 µs per vertex pair, at least 90% of that is memory access time parallelized, this would be sufficiently fast for many quantum computer architectures but not all... achieving sufficiently fast, high error rate, fault-tolerant, parallel surface code quantum error correction is extremely challenging without further work, classical processing speed will limit the maximum tolerable error rate

Summary and further work complexity optimal algorithm –O(n 2 ) serial –O(1) parallel accurately handle general error models sufficiently fast at p = for many architectures next 12 months: –parallelize the algorithm –benchmark on commercially available parallel computing hardware –reduce memory usage, improve core speed, raise practical p –take error correlations into account –develop algorithms capable of simulating braided logical gates –get building!