A Cascaded Random Access Quantum Memory

Abstract

Dynamic random access memory (DRAM) is critical to classical computing but notably absent in current superconducting quantum processors. Integrating high-coherence memory units would enable resource-efficient control of logical qubits and allow the separate optimization of logic and storage subsystems. Here, we realize an 8-bit cascaded random access quantum memory (RAQM). By introducing a buffer layer between the processor and a multimode storage cavity, we leverage the control resources of a single transmon to address eight memory modes while isolating them from processor non-linearities. We demonstrate arbitrary random access with an average infidelity of $\lesssim 1.5\%$ per mode, characterizing the many-body interactions that dominate the error budget. This architecture enables a significant reduction in control lines per logical qubit and supports transversal operations within the memory module, establishing a scalable unit cell for fault-tolerant quantum architectures.

Figures (17)

• Figure 1: (a) Proposed fault-tolerant architecture using Random Access Quantum Memory (RAQM). Each data qubit in the transmon processor (red) also serves as a register qubit for its corresponding RAQM. Each RAQM consists of a buffer mode (green) and a storage multimode (blue). Modes from different cavities are organized into layers to store logical qubits. Logical qubits are transferred from storage layers (blue) to the buffer layer for processing. (b) RAQM device schematic. The readout (black), buffer (green, $2$ modes), and storage (blue, $7$ modes) 3D flute cavities are fabricated on the same high-purity aluminum bulk. The transmon (red) is dispersively coupled to the readout and buffer cavities, enabling universal single-buffer-mode control. The RF-flux-modulated SQUID coupler (purple) enables parametrically driven beam-splitter interactions between coupled cavity modes, transferring information between buffer and storage layers. (c) Frequency spectrum and mode coherence. The storage modes are evenly distributed between $5.3\,\mathrm{GHz}$ and $6.4\,\mathrm{GHz}$, with the highest $T_1$ exceeding $1.2\,\mathrm{ms}$ (see Supplementary Section I). Error bars are smaller than the marker size. Details about dump modes are provided in Supplementary Section II.
• Figure 2: Single storage mode access. (a) Using four-wave mixing interactions $\left|f0\right\rangle\leftrightarrow\left|g1\right\rangle$ between the transmon and buffer mode, we prepare $\left|1\right\rangle$ in the buffer within $0.615\,\mu\mathrm{s}$. (b) By modulating the coupler's RF flux at the frequency difference between the buffer and storage mode $S_2$, a $0.439\,\mathrm{MHz}$ beam-splitter interaction is activated. (c) Randomized Benchmarking (RB) of the $B S_2$ swap. We prepare $\left|1\right\rangle$ in the buffer, execute RB sequences, and measure the joint buffer–storage state $\left|B S_i\right\rangle$. The top panel shows raw and post-selected probabilities of measuring $\left|B S_2\right\rangle$ returning to $\left|10\right\rangle$. Post-selection keeps outcomes with exactly one photon in the buffer–storage pair, isolating non-decoherence errors. The RB pulse sequences are shown in (d). $\left|B S_i\right\rangle$ is extracted using two parity measurements and a swap, with the transmon actively reset to $\left|g\right\rangle$ between parity checks. Swap fidelities for all $B S_i$ are shown in (e). RAQM reset protocols are discussed in Supplementary Section III.
• Figure 3: Random access quantum memory operation. (a) RAM RB circuit on $7$ storage modes. Each storage access gate is highlighted in a different color, containing a storage-buffer swap (read), a buffer gate $G_i^n$, and a buffer-storage swap (write). The storage access gate in the initialization stage does not contain the storage-buffer swap (read), as the storage modes are initialized in the ground state. At the end of the circuit, the target storage mode information is swapped into the buffer for readout. (b) Measured random read fidelity for $7$ storage modes. The reference RB is the buffer mode's RB in the ${\left|0\right\rangle,\left|1\right\rangle}$ subspace. (c) Measured average random read infidelity at different RAQM sizes. The region below the depolarization error correction wootton_2012_high_thresholdbravyi_decoding_surface_code_2014Criger2018multipathsummationKuo2022 threshold for surface code is highlighted.
• Figure 4: RAQM crosstalk benchmarking. (a) Measured cross-Kerr values between storage and buffer modes (log-scale). (b) Three RAQM periods for a target storage mode $S_i$. Buffer gates $G_1$ and $G_2$ are applied during $S_i$'s active and inactive access periods. (c)$BS_i$ active access error due to states stored in $S_j$. The error rate is extracted using the bottom RB sequence. (d) The additional dephasing rate during $S_j$'s inactive access period, where the state stored in spectator modes $S_i$$(i\neq j)$ undergoes random gates. This acts as a dephasing channel to $S_j$, with the rate measured using the bottom sequence. (e) The many-body dephasing rate during the idle period, where the state stored in $S_i$ experiences additional dephasing due to cross-Kerr interactions between storage mode. This rate is measured using the bottom sequence. (f) The random read fidelity error budget. The dominant error channels come from storage decay and $BS_i$ cross-Kerr interactions. Details of the error budget are shown in Supplementary Section X.
• Figure 5: Coupler DC flux sweep. (a) Coupler and buffer (inset) mode frequencies. At each DC flux point, the buffer mode is initially prepared in $\left|1\right\rangle$, followed by an RF-flux modulation frequency sweep. The blue colors in the plot indicate possible transitions. The black-box simulation of the coupler frequency is shown as the red dashed line. The X-axis shows the coupler frequency by subtracting the buffer mode frequency. The scan is performed in two parts due to the RF channel frequency limitations. (b) T$_1$ measurements of the coupler and buffer modes. All RAQM experiments are conducted with the coupler at the dashed line ($0.269\pi$).