Fig. 1: Quantum memory scheme.
From: AI-assisted hyper-dimensional broadband quantum memory with efficiency above 90% in warm atoms
a Experimental setup. VR vortex retarder, HWP half-wave plate, QWP quarter-wave plate, BD beam displacer, FR Faraday rotator, FM Flip mirror, PBS polarization beam splitter, CCD charge coupled device camera, Rb cell 87Rb atomic vapor cell. W write pulse, R read pulse, Sin input signal, SL leaked signal, SR retrieved signal, D1-D3 photo detector, M1 folding mirror. The powers of the W/R beam is 270 mW and 180 mW, respectively. The pump and input signal are 30 μs and 20 ns long, respectively. b Atomic energy level and light frequencies. \(\left\vert g\right\rangle\): \(\vert {5}^{2}{S}_{1/2},F=1\rangle\) and \(\vert m\rangle\): \(\vert {5}^{2}{S}_{1/2},F=2\rangle\) are the two hyperfine ground states of 87Rb D1-line; \(\left\vert e\right\rangle\): \(\left\vert {5}^{2}{P}_{1/2},F=2\right\rangle\) is the excited state; the detuning frequencies of W and R lights are ΔW = 1.8 GHz and ΔR = −1.7 GHz, respectively. Sin and W satisfy two-photon resonant condition. c The flowchart of AI algorithm, that is, Chebyshev sampling differential evolution (CSDE) algorithm, which is the mixing of Chebyshev sampling, residual network and convex optimization. The waveform of W is optimized via CSDE algorithm. d Efficiency as a function of the iteration number of convex optimization and CSDE algorithm. Memory efficiency ηm is the ratio of the average photon number of SR to that of Sin.
