Internet Explorer word sinds april 2017 niet meer ondersteund. Gebruik a.u.b. een moderne browser om deze site te bezoeken.

Workshop on Real-Time Decoding for Fault-Tolerant Quantum Computing

Qblox (spin-out of QuTech, Netherlands) and Muhammad Usman (University of Melbourne, Data61/CSIRO) invite you to the first-of-its-kind workshop focused on real-time decoding to survey current challenges and share ideas for future directions.

By bringing together researchers from quantum error correction and classical computing backgrounds, the workshop aims to cover all aspects, including the choice of the decoding algorithm, the software stack, the classical computational resources to implement it and the supporting architecture.

This workshop is a great opportunity to foster collaborations between researchers, and to evaluate the technical maturity of this technology for industry and research labs.

IEEE Quantum Week 2022 - September 18-23

The workshop will be held as part of IEEE Quantum Week 2022 on Thursday September 22nd, both with in-person and virtual attendance. The workshop will cover 3 sessions of 1.5 hours each (10:00-11:30, 13:00-14:30, 15:15-16:45 MDT).

We can offer workshop attendees a 15% discount on registration using the code P01-QBLOX-15. Please visit and register via IEEE Quantum Week website to join the workshop.

Francesco Battistel, (main contact), Qblox.

Yemliha Bilal Kalyoncu, Qblox.

Muhammad Usman, University of Melbourne.


Workshop Abstract:

Fault-tolerant quantum computation stands as a turning point for reaching quantum advantage. The quantum error correcting code forms the core to realize fault tolerance. However, the efficacy of a code, such as the surface code, is underpinned by the decoder, which can detect errors and suggest appropriate corrections. While previous research has primarily focused on the accuracy and threshold of a decoder, its real-time implementation is still understudied. Since the speed and scalability of the decoder are as critical as its accuracy, real-time decoding manifests itself as a multi-layer challenge: an efficient decoding algorithm must be implemented with the appropriate software layer, which must be executed on fast classical hardware. It is of great importance to find the right combination of these key elements to build a practically-useful decoding architecture.

This workshop aims to create a comprehensive discussion on the subject of real-time decoding and pool ideas for directions in the near and long term. Challenges and advantages of different approaches will be discussed within both a quantum track and a classical-architecture track. The topics of the workshop cover a variety of decoding algorithms, decoding architectures and hardware, as well as co-design strategies for software and hardware. The market-readiness of real-time decoding will also be discussed to shed light on a possible future roadmap for the broader quantum-technology industry.

Keywords: Real-Time Decoding, Decoding, Quantum Error Correction, Computing Architectures, Computer Science, Fault-Tolerant Quantum Computing

Target audience: Researchers and experts from the quantum error correction, quantum computation, computer science, classical architectures and digital design communities, both from academia and industry.

Workshop program:

Walk-in, Ponderosa Room

SESSION I (10:00-11:30 MDT)

Speaker: Francesco Battistel, Qblox
Title: Introduction to the workshop and overview of real-time decoding.

Speaker: Christopher Chamberland, Amazon (virtual)
Title: Techniques for combining fast local decoders with global decoders under circuit-level noise.

Speaker: Natalie Brown, Quantinuum
Title: Wasm + QASM: Assembling real-time decoding.

Speaker: Muhammad Usman
, Univ. of Melbourne and Data61/CSIRO (virtual)
Title: Machine-learning-based fast and autonomous decoder for scalable surface codes


SESSION II (13:00-14:30 MDT)

Speaker: Luka Skoric
, Riverlane
Title: No maximum latency requirement for decoding quantum error correction syndromes.

Speaker: Swamit Tannu,
Univ. of Wisconsin-Madison
Title: Understanding System-level Complexity of Running Quantum Error Correction Codes

Speaker: Ramon Overwater
, TU Delft (virtual)
Title: Hardware Considerations of Neural Network Decoders.

Speaker: Yosuke Ueno
, Univ. of Tokyo & TU Munich
Title: Online decoding of surface code with a superconducting circuit


SESSION III (15:15-16:45 MDT)

Speaker: Poulami Das
, Georgia Tech (virtual)
Title: An architect’s role in error decoding for fault-tolerant quantum computers.


Kenneth Brown, (Duke University)

Christopher Chamberland, (Amazon, virtual)

Poulami Das, (Georgia Tech, virtual)

Ciaran Ryan-Anderson, (Quantinuum)

Niels Bultink, (Qblox)

Topic: Scaling up real-time decoding in the near and long term

About Real-Time Decoding

Quantum error correction will stand at the core of future fault-tolerant quantum computers on the course to quantum advantage. These machines will be universal quantum computers where each logical operation will be error-corrected to ensure fault-tolerance. After strong theoretical foundations for quantum error correction have been laid down during the past two decades, the academic institutes and industry are now focusing on developing quantum error correction in practice. In particular, the focus has been on surface-code(-like) schemes given the nearest-neighbors connectivity and the relatively high error threshold. The first experimental prototypes of small quantum error correcting codes have been demonstrated by ETH Zurich [1], TU Delft [2], Google [3], IBM [4], Duke Univ. [5], Quantinuum [6] et al. Realizing such systems on a large scale is of interest for a broader community that includes theoretical and experimental physicists in quantum computing, as well as researchers and engineers from classical computing background and software engineers.

In quantum error correction, the error syndrome is repeatedly measured to extract coarse-grained information about errors, and this information is then fed into a decoder that tries to infer which error actually occurred and provides a correction. Decoders have generally been optimized for accuracy, rather than speed or scalability. Only recently research efforts have focused on the scalability and latency of decoder architectures [7-23], with a particular focus on the widely-adopted surface code. However, in the long term, in the case of a universal quantum computer, the decoder must run in real time, thus providing a correction within one, or at most a few, quantum-error-correction cycles.

The short quantum-error-correction cycle (less than 1 microsecond) poses extremely tight bounds for real-time decoding in the case of platforms such as superconducting qubits, whereas these bounds are looser for platforms such as trapped ions (see Fig.1). In either case, implementing a fast decoder and the whole supporting architecture is still a great challenge. In the trapped-ion experiment in Ref. [6], a lookup-table decoder was applied in real time, however, lookup tables are not scalable and thus more sophisticated decoding strategies are mandatory towards large-scale fault tolerance. Examples of decoding strategies that have been recently shown to offer promising performance in decoder implementation include machine learning [7, 8, 15, 20, 21] and union-find decoding [9, 10, 14]. In Fig.1 we summarize the performance of these and other decoder implementations that have been proposed in the literature. One can see that, for the stricter case of superconducting qubits, the target is met only by a few proposals that use either union-find decoding [9], neural networks [8] or single-flux-quantum logic [11, 18, 19]. Note, however, that all of these consider ASICs that have not been built yet. Lookup-table decoders meet the target in terms of speed but memory constraints limit them to distance-5 [12, 13]. Minimum-weight perfect matching, while representing the standard in terms of accuracy, is at the moment very far from reaching the target, unless coupled with a low-level decoder in a hierarchical strategy [10]. All in all, more work is needed for the practical realization of each of these strategies, and they will require innovative, suitable software implementation and underlying classical-computing hardware for execution alongside a quantum processor. Real-time decoding is thus a multi-layer challenge that requires novel, multi-layer strategies.

In this workshop, we want to explore all layers of real-time decoding and foster the development of a community around this topic, in order to accelerate the path towards fault-tolerant quantum computing. While many conferences and workshops in the past have been dedicated to quantum error correction in general, to our knowledge this would be the first workshop with a sharp focus on real-time decoding. The workshop wants to survey challenges and advantages of the pioneering works on the topic, as well as to pool ideas for future directions and to spur new scientific collaborations among the attendees in the shorter term. The workshop concerns all layers, from the choice of decoding algorithm to the software stack implementing it, from the classical decoding hardware to the architecture that surrounds it.

Furthermore, the theory-oriented participants could strengthen or start new collaborations with experimental partners to streamline the deployment of real-time-decoding prototypes into actual quantum-computing platforms for intermediate-scale codes. We aim at involving researchers both from a quantum-error-correction and from a classical-computing-architecture background since we believe that the cross-pollination between these two fields could be particularly fruitful to bring innovative ideas into the industry for the longer term. The workshop will also represent an occasion for industry and research labs to evaluate the market readiness of this technology and shape their investments in R&D projects related to real-time decoding.

Fig.1 Comparison of Decoders: Summary of decoders for a distance-11 surface code (unless otherwise specified). Distance-11 is chosen since it represents a target that is challenging to implement but that at the same time might be attainable within a few years. We include the papers that report the decoding time. If no data point was available for distance-11, we extrapolate it from a fit of the available data if possible (if not, we specify the distance d). Decoder types are marked by similar color shades: look-up table (LUT, red), union-find (UF, green), neural network (NN, purple), minimum-weight perfect-matching (MWPM, blue), low-level (LL, cyan), modified MWPM (brown). The labels on the left report the decoder type and the classical computing hardware used in the given paper. SFQ refers to single-flux-quantum logic. By choosing a uniform distance we enable a comparison between decoders, even though the physical error rates vary.

Indeed, the speed of especially UF and MWPM depends on the physical error rate as more detection events require longer to be matched. However, the physical error rates are fairly uniform, being 5-10% in the case without measurement errors, and 0.1-0.2% if measurement errors are considered (phenomenological or circuit-level noise); exceptions are [15] with 1% and [17] with 5%. Vertical dotted or dashed lines represent estimates of the available time to compute a correction and feed it back to the qubits before the following quantum-error-correction (QEC) cycle. For superconducting transmon qubits the estimate (400 ns) is based on [1, 2, 8-10, 12] and for trapped ions we take 1000 times that (i.e. 400 µs) since the clock speed of ions is much lower than transmons, even though the QEC-cycle time of 200 ms reported in [6] suggests that the bounds for ions might be even much looser.


[1] Krinner S. et al., “Realizing Repeated Quantum Error Correction in a Distance-Three Surface Code”, [2112.03708] (

[2] Marques J.F. et al., “Logical-qubit operations in an error-detecting surface code”, Nature Physics

[3] Chen Z. et al., “Exponential suppression of bit or phase flip errors with cyclic error correction”, Nature

[4] Sundaresan N. et al., “Matching and maximum likelihood decoding of a multi-round subsystem quantum error correction experiment”, [2203.07205] (

[5] Egan et al., Fault-Tolerant Operation of a Quantum Error-Correction Code, [2009.11482] (

[6] Ryan-Anderson C. et al., “Realization of real-time fault-tolerant quantum error correction”, [2107.07505] (

[7] Gicev S. et al., “A fast and scalable artificial neural syndrome decoder for surface codes”, [2110.05854] (

[8] Overwater R. et al., “Neural-Network Decoders for Quantum Error Correction using Surface Codes: A Space Exploration of the Hardware Cost-Performance Trade-Offs”, [2202.05741] (

[9] Das P. et al., “A Scalable Decoder Micro-architecture for Fault-Tolerant Quantum Computing”, [2001.06598] (

[10] Delfosse N., “Hierarchical decoding to reduce hardware requirements for quantum computing”, [2001.11427] (

[11] Ueno Y., “QECOOL: On-Line Quantum Error Correction with a Superconducting Decoder for Surface Code”, [2103.14209] (

[12] Das P. et al., LILLIPUT: A Lightweight Low-Latency Lookup-Table Based Decoder for Near-term Quantum Error Correction, [2108.06569] (

[13] Riste D. et al., Real-time decoding of stabilizer measurements in a bit-flip code, [1911.12280] (

[14] Huang S. et al., Fault-Tolerant Weighted Union-Find Decoding on the Toric Code, Phys. Rev. A 102, 012419 (2020)

[15] Meinerz K. et al., Scalable Neural Decoder for Topological Surface Codes, Phys. Rev. Lett. 128, 080505 (2022)

[16] Fowler A. et al., Towards practical classical processing for the surface code: timing analysis, Phys. Rev. A 86, 042313 (2012)

[17] Higgott O. et al., PyMatching: A Python package for decoding quantum codes with minimum-weight perfect matching, [2105.13082] (

[18] Holmes A. et al., NISQ+: Boosting quantum computing power by approximating quantum error correction, [2004.04794] (

[19] Ueno Y. et al., QULATIS: A Quantum Error Correction Methodology toward Lattice Surgery, IEEE Conference Publication | IEEE Xplore

[20] Ueno Y. et al., NEO-QEC: Neural Network Enhanced Online Superconducting Decoder for Surface Codes, [2208.05758] (

[21] Chamberland C. et al., Techniques for combining fast local decoders with global decoders under circuit-level noise, [2208.01178] (

[22] Smith S. et al, A local pre-decoder to reduce the bandwidth and latency of quantum error correction, [2208.04660] (

[23] Ravi G.S. et al., Have your QEC and Bandwidth too!: A lightweight cryogenic decoder for common / trivial errors, and efficient bandwidth + execution management otherwise, [2208.08547] (