Minute-Scale Photonic Quantum Memory

TL;DR

This work tackles the challenge of photonic quantum memories with truly minute-scale storage by implementing a noiseless photon echo (NLPE) protocol in $^{151}$Eu$^{3+}$:Y$_2$SiO$_5$ at a ZEFOZ magnetic field and coupling it with a universally robust dynamical decoupling sequence based on CHS adiabatic pulses (CHS-UR4). The combination enables direct use of the crystal's strong absorption and long spin coherence to achieve a $1/e$ storage lifetime of $T_M=27.6\pm0.5$ s and preserve time-bin qubits with a fidelity of $F_t=88.0\pm2.1\%$ at $5.6$ s, with single-photon-level storage up to $42$ s and an SNR above unity. Key technical advances include replacing rectangular pulses with adiabatic CHS pulses, embedding them in the UR4 sequence, and locking timing to a rubidium standard to maintain phase coherence across long sequences. These results establish minute-scale photonic quantum memory as a viable building block for global quantum networks and deep-space quantum experiments, with prospects for further enhancement via improved magnetic-field homogeneity, heavier Eu isotopes (e.g., $^{153}$Eu), optimized DD schemes, and cavity-enhanced absorption.

Abstract

Long-lived storage of single photons is a fundamental requirement for enabling quantum communication and foundational tests of quantum physics over extended distances. While the implementation of a global-scale quantum network requires quantum storage times on the order of seconds to minutes, existing photonic quantum memories have so far been limited to subsecond lifetimes. Although $^{151}$Eu$^{3+}$:Y$_2$SiO$_5$ crystals exhibit substantially extended spin coherence times at the `magic' magnetic field, the concomitant weak optical absorption has until now prevented single-photon storage. Here, we overcome this challenge by integrating a noiseless photon echo protocol--which makes full use of the crystal's natural absorption for photonic storage--with a universally robust dynamical decoupling sequence incorporating adiabatic pulses to efficiently protect delocalized spin-wave excitation, enabling long-lived quantum storage at the `magic' magnetic field. At a storage time of 5.6 s, we achieve a time-bin qubit storage fidelity of 88.0 $\pm$ 2.1%, surpassing the maximum fidelity attainable via classical strategies. Our device reaches a $1/e$ storage lifetime of 27.6 $\pm$ 0.5 s, enabling single-photon-level storage for 42 s with a signal-to-noise ratio greater than unity. This work establishes photonic quantum memory in the minute-scale regime, laying a solid foundation for global-scale quantum network and deep-space quantum experiments.

Figures (7)

• Figure 1: Schematic of the experimental setup. The polarization of signal and control beams are controlled with half-wave plates (HWPs), to be polarized parallel to the $D_1$ axis of the memory crystal (MC) to maximize interaction strength. Both beams overlap inside the MC an angle of 1.5$^{\circ}$. The signal beam propagates along the MC’s b-axis, guided by three steering mirrors, and is retro-reflected along the same path. A 95(R):5(T) beam splitter (BS) separates the output signal, which is then coupled into a single‑mode fiber. The MC is precisely aligned inside the magnet using two goniometer stages. Dynamical decoupling on the spin transitions is implemented by applying amplified RF pulses via solenoid coils surrounding the MC. Photonic time-bin encoded qubits are stored inside the MC with a $1/e$ lifetime of 27.6 s. The readout signal subsequently passes through a temporal gate based acousto-optic modulator (AOM), followed by spectral filtering with a double-passed filter crystal (FC) and a 0.5-nm band-pass filter (BPF). The FC is prepared by a pump beam to create a 0.8-MHz transparent spectral window centered at the signal frequency. The filtered signal is finally detected with a fiber‑coupled single‑photon detector (SPD).
• Figure 2: Characterization of the long-lived quantum memory.a. Energy level structure of $^{151}$Eu$^{3+}$ in Y$_2$SiO$_5$ crystal under the ZEFOZ magnetic field of 1.2968 $\pm$ 0.0002 T. Colored lines represents optical and spin transitions: red for the signal beam, orange and coral for control beams; the green marker indicates the ZEFOZ spin transition at 12.456 MHz. b. Experimental sequence of the long-lived quantum memory. The gray region indicates initialization. This is followed by the NLPE-DD storage sequence, which employs a DD process based on the CHS-UR4 sequence. The final photonic echo is emitted at $t_e$ = $t_4$ + $t_3$ - $t_2$ - $t_1$ + $t_0$. c. Decay of spin echo amplitude with total evolution time for two-pulse spin echo (triangle) and the CHS-UR4 sequences (diamond). Spin echoes are detected via pulsed Raman heterodyne detection 1hour. The data are fitted to Mims' equation $e^{-(t/T_2)^m}$Mims, yielding $T_2$ = 18.7 s, $m$ = 1.05 for the two-pulse spin echo, and $T_2$ = 33.1 s, $m$ = 1.25 for the CHS-UR4 sequence. d. Photonic storage efficiency of the NLPE-DD memory as a function of storage time. The data are fitted to $e^{-(t/T_M)^m}$ with $T_M$ = 27.6 $\pm$ 0.5 s and $m$ = 1.70. Red circles denote measurements using classical light; purple triangles represent single-photon-level inputs. The slightly higher efficiency in single-photon measurements results from the uncorrected contribution from noise counts, which are negligible under classical inputs. All error bars indicate $\pm$1 standard deviation throughout.
• Figure 3: Storage of single-photon level inputs.a and b. Photon counting histogram of NLPE memory (a) and NLPE-DD memory (b), measured using weak coherent input pulses with a mean photon number $\mu$ = 1.18 and integrated over 35,000 experimental trials. The blue, red, and green curves denote the input photon counts, output signal, and noise level (measured without input), respectively. For clarity, data to the left of the dashed line are scaled down by a factor of 4. The shaded region indicates the 2.1 µ s detection window, yields SNRs of 65.2 $\pm$ 27.2 and 11.3 $\pm$ 2.5 at readout times of 28.1 µ s and 5.600032 s, respectively.
• Figure 4: Storage of time-bin qubits for 5.600032 s.a. Photon counting histogram for the input qubits $\ket{e}$ and $\ket{l}$. Output counts for $\ket{e}$ and $\ket{l}$ are shown as red and green bars, respectively. The gray-shaded region marks the 2.1 µ s detection window. The mean input photon number per qubit is $\mu_q$ = 1.16, integrated over 35,000 experimental trials. The measured fidelities are $F_{\ket{e}}$ = 92.7 $\pm$ 1.4% and $F_{\ket{l}}$ = 92.7 $\pm$ 1.4%. b,c. Photon counting histogram for the input qubit $\ket{e}+\ket{l}$ and $\ket{e}+i\ket{l}$, respectively. red and green bars correspond to constructive and destructive interference measurements for each qubit. All other experimental settings match those in a. The measured fidelities are $F_{\ket{e}+\ket{l}}$ = 85.8 $\pm$ 2.5% and $F_{\ket{e}+i\ket{l}}$ = 85.6 $\pm$ 2.5%. d. Classical fidelity limit for memory as a function of $\mu_q$. The green dashed line indicates the classical bound for a measure-and-prepare strategy under 8.2% storage efficiency. The gray dot marks the measured memory fidelity of 88.0 $\pm$ 2.1% at $\mu_q$ = 1.16. The solid red line represents the expected fidelity calculated from the experimentally determined efficiency and noise.
• Figure 5: Storage of qubit $\ket{l}$ for 42.00032 s. Photon counting histogram of of the input qubit $\ket{l}$ encoded with weak coherent pulses with $\mu_q$ = 4.14, accumulated over 28,000 experimental trials. The blue curve indicates the input photon counts, and the red curve shows the storage process. For visual clarity, photon counts to the left of the vertical dashed line are scaled down by a factor of 20. The dark and light shaded regions mark the detection windows for the orthogonal time bin $\ket{e}$ and target time bin $\ket{l}$, respectively.