2026-03-23
Quantum Random Access Memory (QRAM) provides an efficient interface for data exchange between quantum and classical devices, making it a key foundational component for advancing quantum computing toward practical applications. Recently, the research team involving core members of Logical Qubit Technology achieved the first experimental demonstration of QRAM on a superconducting quantum computing platform. The work demonstrated the noise resilience of QRAM under practical operating conditions and provided important support for the scalable development of QRAM systems. The related results were published in Nature Physics as an Article under the title A bucket-brigade quantum random access memory.
Quantum access to classical data is a fundamental requirement for many quantum algorithms, including quantum machine learning, quantum simulation, and Grover's algorithm. In practical applications, this query process often dominates the computational time cost of quantum algorithms, thereby limiting the realization of overall quantum speedup. Classical random access memory retrieves stored data based on input address bits, while QRAM can accept address bits in quantum superposition states, enabling parallel access to multiple memory cells. As a result, QRAM is regarded as a key technology for overcoming the data transfer bottleneck between quantum and classical systems.
Among various architectures, the bucket-brigade QRAM architecture is considered one of the most promising implementations because its query error scales only logarithmically with memory size. This architecture mainly consists of quantum routers arranged in a binary tree structure. Each quantum router contains four qubits: one selector qubit, one input qubit, and two output qubits. During operation, the selector qubit stores address information, while the input and output qubits transmit quantum information. Depending on the state of the selector qubit, the router directs the quantum state of the input qubit to the corresponding output qubit. Each output qubit then serves as the input qubit for the next-level router, propagating information layer by layer through the tree. When the selector qubit is prepared in a quantum superposition state, different routing paths also coherently superpose, enabling the parallel addressing capability of QRAM. However, quantum routing operations require high-precision control of four-body interactions, making experimental implementation extremely challenging and significantly limiting the development of QRAM technologies.
To address this challenge, the research team proposed an efficient quantum router gate decomposition scheme that reduced the quantum circuit depth required for QRAM operation by approximately 30%. Building upon this approach, and leveraging a high-performance quantum chip independently designed and fabricated with the participation of core members of Logical Qubit Technology — featuring single-qubit gate fidelity of 99.96% and two-qubit gate fidelity of 99.7% — the team achieved high-fidelity quantum routing operations with a fidelity of 94.5%. By cascading quantum routers and combining quantum error mitigation techniques, the team successfully constructed 4-address and 8-address QRAM architectures, achieving query fidelities of 80.9% and 60.4%, respectively. These results demonstrated the experimental feasibility of bucket-brigade QRAM on a superconducting quantum platform.
Following the full experimental demonstration of the QRAM process, the research team further investigated the mechanism of noise propagation within the architecture. In the experiments, error mitigation techniques were separately applied to queried and non-queried QRAM branches. The results showed that error mitigation on non-queried branches produced significantly weaker improvements in query fidelity compared with queried branches, preliminarily revealing the non-uniform impact of errors within QRAM systems.
To further validate this phenomenon, the team intentionally introduced different levels of noise into different QRAM branches and systematically analyzed how the location of errors affected the final query results. The experiments showed that when the branch containing the introduced noise became increasingly distant from the queried branch, the influence on the final output port became dominated by progressively higher-order error processes. This finding verified the localized nature of error propagation in QRAM architectures.
In addition, by measuring the entanglement entropy of router qubits at different hierarchical levels, the team observed that the entanglement entropy gradually decreased with increasing circuit depth. This result indicates that quantum correlations between QRAM branches weaken as the hierarchy deepens, revealing the physical origin of the intrinsic noise resilience of bucket-brigade QRAM architectures. These findings provide important experimental evidence for the scalable and noise-resilient development of QRAM systems.
