A Classical Architecture For Digital Quantum Computers

Fang Zhang, Xing Zhu, Rui Chao, Cupjin Huang, Linghang Kong, Guoyang Chen, Dawei Ding, Haishan Feng, Yihuai Gao, Xiaotong Ni, Liwei Qiu, Zhe Wei, Yueming Yang, Yang Zhao, Yaoyun Shi, Weifeng Zhang, Peng Zhou, Jianxin Chen

Abstract

As quantum computing scales toward fault-tolerance, the classical control infrastructure faces significant challenges in instruction bandwidth, synchronization, and real-time decoding latency. Researchers from Alibaba’s DAMO Academy have developed a scalable classical architecture that integrates multi-core RISC-V CPUs with custom control electronics. By utilizing a "broadcast" instruction pipeline and a star-topology hierarchy, the system demonstrates the capability to perform real-time syndrome decoding for large surface codes ($d=67$) in under one microsecond. This flexible, software-defined approach overcomes the limitations of traditional rigid hardware, providing a path toward large-scale digital quantum computing.

Introduction

The transition from small-scale quantum prototypes to large-scale, fault-tolerant quantum computers (FTQC) is hindered by a "classical bottleneck." This bottleneck is characterized by three primary stressors:

  1. Instruction Bandwidth: Controlling thousands of qubits requires an immense volume of instructions, potentially exceeding standard communication capacities.
  2. Timing Synchronization: Maintaining picosecond-level precision across a distributed array of control electronics is a significant engineering hurdle.
  3. Decoding Latency: FTQC requires real-time error correction. If the classical system cannot decode syndromes faster than the quantum cycle time (typically $\sim 1\mu s$), an "exponential syndrome backlog" occurs, leading to system failure.

Current solutions often rely on specialized ASICs or FPGAs, which lack the flexibility needed to adapt to evolving quantum algorithms and error-correction protocols.

Proposed Scalable Architecture

The proposed architecture balances high-performance execution with general-purpose flexibility by integrating a multi-core RISC-V CPU directly into the control loop.

Detailed block diagram of the integrated classical control unit and quantum electronics interface.

Multi-core RISC-V Integration

Instead of hard-coded decoders, the system utilizes high-performance RISC-V cores. This allows decoding algorithms to be implemented as firmware, facilitating rapid iteration. The architecture employs a "Sandwich Decoder" approach, where the decoding task is parallelized across multiple cores using a windowing technique.

Instruction Broadcasting and Star-Topology

To mitigate instruction bandwidth stress, the architecture leverages the repetitive "brickwork" structure of quantum circuits. A single instruction can be broadcast to multiple qubits simultaneously (SIMD), significantly reducing the required data rate. A star-topology hierarchy is used for physical connections, ensuring that timing jitter remains minimal (~6ps) and independent of system size.

Instruction Set Architecture (ISA) Extensions

The system bridges the gap between classical processing and quantum control through custom RISC-V instruction extensions. These extensions allow the CPU to issue pulse-level and gate-level commands directly to the quantum electronics via memory-mapped IO (MMIO).

Mapping of RISC-V instruction extensions to low-level In-house Quantum Electronics (IQE) instructions.

Experimental Evaluation

The architecture was benchmarked using two primary models: a quantum memory experiment and a Bell-state experiment involving lattice surgery. Performance was evaluated across several RISC-V platforms, including the 1088-core ET-SoC-1.

Key Results

  • Real-time Decoding: The system achieved sub-microsecond decoding for surface code distances up to $d=67$ under optimistic noise models, and $d=47$ under realistic conditions. This represents a significant advancement over previous specialized hardware.
  • Bandwidth Efficiency: The instruction pipeline operated at a manageable 0.96 Gbps, demonstrating that standard differential pair wiring is sufficient for large-scale control.
  • System Stability: Timing jitter was maintained at approximately 6ps, ensuring high-fidelity gate operations across the entire array.
  • End-to-End Validation: The architecture was successfully validated in a live laboratory environment by performing a $T_1$ relaxation experiment on a superconducting quantum processor.

Conclusion and Future Work

This work demonstrates that a general-purpose, software-defined classical architecture can meet the stringent requirements of fault-tolerant quantum computing. By moving the "intelligence" of the control system from rigid hardware to flexible RISC-V firmware, the architecture provides a scalable and adaptable foundation for future quantum processors.

Future research will focus on scaling the physical implementation to the 10,000-qubit regime, addressing potential thermal and physical bulk challenges. Additionally, the integration of cryo-CMOS components and the execution of more complex multi-logical-qubit algorithms remain key areas for development.

Summary updated

BibTeX
@article{zhang2024classical,
  author = {Fang Zhang and Xing Zhu and Rui Chao and Cupjin Huang and Linghang Kong and Guoyang Chen and Dawei Ding and Haishan Feng and Yihuai Gao and Xiaotong Ni and Liwei Qiu and Zhe Wei and Yueming Yang and Yang Zhao and Yaoyun Shi and Weifeng Zhang and Peng Zhou and Jianxin Chen},
  title = {{A Classical Architecture For Digital Quantum Computers}},
  journal = {ACM Trans. Quantum Comput},
  year = {2024},
  doi = {10.1145/3626199},
}
Figure