Quantum memory with strong and controllable Rydberg-level interactions

Realization of distributed quantum systems requires fast generation and long-term storage of quantum states. Ground atomic states enable memories with storage times in the range of a minute, however their relatively weak interactions do not allow fast creation of non-classical collective states. Rydberg atomic systems feature fast preparation of singly excited collective states and their efficient mapping into light, but storage times in these approaches have not yet exceeded a few microseconds. Here we demonstrate a system that combines fast quantum state generation and long-term storage. An initially prepared coherent state of an atomic memory is transformed into a non-classical collective atomic state by Rydberg-level interactions in less than a microsecond. By sheltering the quantum state in the ground atomic levels, the storage time is increased by almost two orders of magnitude. This advance opens a door to a number of quantum protocols for scalable generation and distribution of entanglement.

Prior to execution of the quantum memory protocol, optical pumping is employed to prepare atoms in the ground state |a⟩ = |5s1/2,F = 1,mF = 0>. To empty the F = 2 hyperfine sub-level, we employ two laser fields: a π-polarized field Ωπ and a σ + -polarized field Ωσ+.
Both fields are resonant with the |5s1/2,F = 2> ↔ |5p1/2,F = 2> transition. The Ωπ propagates along the x axis while the Ωσ+ field is mixed into the beam path of the control field Ωc. To prepare atoms in the |F = 1,mF = 0> state, a π-polarized field Ωop resonant with the |5s1/2,F = 1> ↔ |5p1/2,F = 1> transition is used. After atoms are loaded and cooled in the dipole trap, alternating pulses of the Ωop field and the Ωπ +Ωσ+ fields are applied for 200 µs.
The 20-µs-long quantum memory protocol is repeated 8,000 times for each sample preparation. The overall duration of one experimental cycle is 0.78 s. Supplementary The Rydberg transfer field Ω2 and read-out field Ωr have the same wave vector k as the Rydberg excitation field Ω1 and the control field Ωc, respectively. As a result, the retrieved field is approximately phase-matched into the spatial mode of the probe field Ωp, which is coupled into a single mode fiber and split by a 50/50 fiber beamsplitter for the g (2) (τ) measurement.

Supplementary Note 3. Preparation efficiency for single atomic excitations.
The probability of photoelectric detection P is proportional to the single excitation preparation efficiency ξ: P = ηrηtdξ. Here ηr is the efficiency of converting a single excitation in state |b⟩ into a retrieved-field photon. The photon transmission and detection efficiency ηtd = ηaηfηd = 0.24, where ηa = 0.75, ηf = 0.65 and ηd = 0.5 are AOM diffraction efficiency, fiber coupling efficiency and single photon detection efficiency, respectively. We extract ηr from the overall efficiency of light storage ηL = ηrηs, using the storage efficiency ηs determined from the transmitted fraction of probe field Ωp. From the measured values of ηL = 0.00069(2) and ηs = 0.0111(2), we infer ηr = 0.062(2).
The efficiency of preparing single atomic excitations in state |b⟩ can be determined as ξ = P/(ηrηtd) = P × 67(2). (1) Using the measured value of P = 0.12(1)%, we find ξ ∼ 8.1(6)%. For the data shown in Figure 2(b) of the main text, the measured values of P (PR) are normalized by ηr and ηtd to obtain N (NR). Here P and PR are the probabilities of photoelectric detection with and without coupling to the Rydberg state, respectively.
For the interaction-induced dephasing mechanism, the efficiency of preparing a retrievable single excitation is limited by 1/e. By employing Rydberg levels of higher principal quantum number n and/or smaller ensemble volumes, transition into the regime of Rydberg excitation blockade can be achieved, with a corresponding increase in preparation efficiency ξ. The latter is also affected by the (motional) Rydberg-ground decoherence, which can be mitigated by adopting a state-insensitive trap for ground and Rydberg atoms.