BAQIS Quantum Science Forum 21: Shallow neural network for real-time fault-tolerant decoding of surface codes


Date: Nov 13 2020 Friday
Time: 10:00-11:00

Venue: Meeting room 526

Webinar: Tencent Meeting 腾讯会议(ID:999577580)


Topic: Shallow neural network for real-time fault-tolerant decoding of surface codes

Speaker:Dr. Zheng, Senior research fellow, Tencent Quantum Lab (TQL) 

Abstract: To experimentally implement fault-tolerant quantum computation~(FTQC) on surface codes, fast decoding becomes essential due to the short lifetime of qubits (~100us for superconducting qubits at most) in the lab. A decoder based on neural networks (NN) is promising since once trained, it can be easily parallelized on specific hardware (like FPGA or ASIC), and execute in constant time. However, fault-tolerant NN decoders (FTNND) are hard to design and train when the distance of the code $L$ is large due to the quick growth of the size of the networks. In this paper, we propose a fast and scalable FTNND decoder based on shallow depth convolutional neural networks (CNNs) inspired by the Renormalization Group (RG) decoder. The network model of such CNN decoders are simple and can be constructed systematically for the arbitrary size of the code. The computation resource required to train the network also scales up reasonably with $L$, which typically only requires several Nvidia V100 GPUs for L < 15. The depth of such neural networks grows slowly as $O(\log L)$, putting a small lower bound for the decoding latency for near term experiments. The performance for the shallow CNN decoder is near-optimal in the low error rate region with a threshold of around 0.5% in the fault-tolerant scenario for the circuit depolarization noise model. We also quantized the CNN models for $L = 5$ and distribute them on single Intel Stratix V FGPA chips, showing the real-time feedback FTQEC can be done in around 700ns, which is 3 orders of magnitude faster than the best benchmark results before.

About the speaker:

Dr. Zheng is a senior research fellow working on quantum computing at Tencent Quantum Lab (TQL) in Shenzhen since 2018. Before joining Tencent, he has been a post-doc working at Centre for Quantum Technologies and Yale-NUS College, Singapore since 2015. He got his M.S in Computer Science and Ph.D. in Electrical Engineering from the University of Southern California in 2013 and 2015, respectively. His research interests focus on the architecture of fault-tolerant quantum computation, quantum error correction/mitigation, quantum simulation, and physical platforms like superconducting qubits, quantum dots, and neutral atoms.