Simple Atomic Quantum Memory Suitable for Semiconductor Quantum Dot Single Photons

Quantum memories matched to single photon sources will form an important cornerstone of future quantum network technology. We demonstrate such a memory in warm Rb vapor with on-demand storage and retrieval, based on electromagnetically induced transparency. With an acceptance bandwidth of $\delta f=0.66\mathrm{GHz}$, the memory is suitable for single photons emitted by semiconductor quantum dots. In this regime, vapor cell memories offer an excellent compromise between storage efficiency, storage time, noise level, and experimental complexity, and atomic collisions have negligible influence on the optical coherences. Operation of the memory is demonstrated using attenuated laser pulses on the single photon level. For a 50 ns storage time, we measure ${\eta }_{\mathrm{e}2\mathrm{e}}^{50\mathrm{ns}}=3.4(3)\%$ end-to-end efficiency of the fiber-coupled memory, with a total intrinsic efficiency ${\eta }_{\mathrm{int}}=17(3)\%$. Straightforward technological improvements can boost the end-to-end-efficiency to ${\eta }_{\mathrm{e}2\mathrm{e}}\approx 35\%$; beyond that, increasing the optical depth and exploiting the Zeeman substructure of the atoms will allow such a memory to approach near unity efficiency. In the present memory, the unconditional read-out noise level of $9\times {10}^{-3}$ photons is dominated by atomic fluorescence, and for input pulses containing on average ${\mu }_1=0.27(4)$ photons, the signal to noise level would be unity.

Figure 1

(a) Energy levels of the Rb $D_1$ line and transitions involved in the memory experiments. Virtually all atoms are initially prepared in the $F=1$ ground state. The vertically polarized signal to be stored is detuned by $\mathrm{\Delta }$ from the $F=1\to F^{\prime }=1$ transition, while the horizontally polarized control laser is detuned by $\mathrm{\Delta }$ from the $F=2\to F^{\prime }=1$ transition. (b) Experimental setup for the memory experiment. EOM: electro-optic modulator; AWG: arbitrary waveform generator; TA: tapered amplifier; PBS: polarizing beam splitter; Rb: vapor cell; $\lambda /2$, $\lambda /4$: wave plates; detector: single avalanche photodiode (APD), a pair of APDs in a Hanbury Brown and Twiss configuration, or a single APD at one output of a heavily unbalanced Mach-Zehnder interferometer. (c) Shape of the signal pulses used in the storage and retrieval experiments measured with 250 ps timing resolution (left panel) and its Fourier transform (right panel). The latter indicates a FWHM bandwidth of 0.66 GHz.

Figure 2

Arrival time histogram for the photons detected in a memory experiment with a storage time of 50 ns, for a coherent input state with the envelope shown in Fig. 1 containing one photon on average [$|\alpha {|}^2=1.0(1)$] and for blocked input signal ($|\alpha {|}^2=0$). Bin size is 1.3 ns. The time shifted input pulse is shown for reference ($0.1\times \mathrm{signal}$ in). The measured, noise corrected end-to-end efficiency of the memory setup including the filtering system is ${\eta }_{\mathrm{e}2\mathrm{e}}^{50\mathrm{ns}}=3.4(3)\%$, while the signal to noise ratio $\mathrm{SNR}=3.7(6)$ for the single photon level input pulse.

Figure 3

(a) Typical interference between two subsequently stored and retrieved pulses, measured via thermal phase fluctuations of an unbalanced Mach-Zehnder interferometer. Integration time is 1 s for each data point. For a single photon level input pulse, a fringe visibility of $V=0.65(5)$ is achieved (curve “retrieval”). The read-out noise shows a fringe visibility below $V=0.15$ originating from random intensity fluctuations. To compensate for an imperfect mode overlap in the interferometer, data are normalized such that the input laser shows unity fringe visibility $V=(\max -\min /\max +\min )$, where max (min) refers to the maximum (minimum) observed photon flux indicated by dashed lines. (b) Second order autocorrelation $g^{(2)}(\tau )$ of photons detected during input pulse and read-out process, respectively. Because of contamination by noise, the read-out signal shows a $g^{(2)}(0)=1.3$, while $g^{(2)}(0)=1.0$ is expected in a perfect memory experiment. As expected for thermal (coherent) radiation, the read-out noise (input signal) exhibits $g^{(2)}(0)=2.0$ [$g^{(2)}(0)=1.0$]. Data are normalized to the peaks at $\pm 600\mathrm{ns}$ and input signal (retrieval) is shifted by $-16$ $(+16)\mathrm{ns}$ for better visibility.

Figure 4

Total efficiency for storage and subsequent retrieval of the experimentally realized signal as a function of the detuning $\mathrm{\Delta }$ for various situations. (i): Measured data. Simulation with same OD and level scheme as in the experiment for (ii): the experimentally realized Gaussian control pulses with 120 mW peak power, (iii): Gaussian control pulses with higher laser power, (iv): optimal control pulses. Simulation with suppressed parasitic single-photon transitions for (v): $\mathrm{OD}=5$ and (vi): $\mathrm{OD}=35$.