Quantum memory on a nanophotonic silicon chip

Abstract

Integrated photonic circuits offer great promise for quantum technologies. However, due to the rapid propagation of light, many envisioned applications require efficient on-chip quantum memories with a programmable delay, compact footprint, and high fidelity. Implementing this based on standard semiconductor processing technology is an outstanding challenge. Here, we realize such memories using erbium-doped silicon waveguides, fabricated as part of a multi-wafer project by a nanophotonic foundry. We demonstrate light storage with a $44.2(9)\ \text{MHz}$ bandwidth and a programmable delay exceeding $1\ μ\text{s}$ in a device with a footprint of only $1.5\times 10^{-2}\ \text{mm}^2$, outperforming on-chip delay lines by many orders of magnitude. The phase of the read-out light field is preserved with a visibility of $91.3(30)\ \%$. The efficiency of $1.89(28)\times 10^{-8}$ can be improved in future devices through resonator enhancement and higher dopant concentrations. With this, the demonstrated approach will pave the way towards applications in photonic quantum computing based on scalable silicon processing technology.

Figures (4)

• Figure 1: Setup and device characterization. a, Photograph of the nanophotonic chip hosting the quantum memory on top of a 1-Euro-cent coin for a size comparison. b, Optical microscope image of the lensed fiber tip (top) used to edge-couple light into one of the waveguides on the silicon nanophotonic chip (bottom). c, Schematic of the fiber-based measurement setup. The light of a frequency-stabilized laser (top left) is switched and frequency-shifted by optical modulators. Using a fiber-based beam-splitter, part of the light is directed to the sample, which is mounted in a closed-cycle cryostat. The light emitted from the sample is guided to a single-photon detection system via a fiber-based temporal and spectral filtering system. d, Pulsed resonant fluorescence spectrum of the Er ensemble emission at 0.75T. The sample is excited with laser pulses of 15µ s duration and the fluorescence is measured within 265µs after the pulses (grey data). A Lorentzian fit (blue) yields a linewidth of 351 ± 16MHz. The detuning is defined as the frequency difference with respect to the emission wavelength of Erbium in site A at 0T 1537.7629 ± 0.00009nm. Error bars denote one standard deviation after averaging 1e3 measurements.
• Figure 2: Spectral hole burning in Er:Si waveguides. a, Left inset: Level scheme. Burn pulses are applied repeatedly on the $\ket{\downarrow}_\text{g} \rightarrow \ket{\downarrow}_\text{e}$ transition (thick red arrow). Spontaneous decay on the spin-flip transition (thin red arrow) transfers resonant dopants to $\ket{\uparrow}_\text{g}$. Right inset: Pulse sequence. 20.0 excitation pulses (red) of 6µs duration each, separated by $t_\text{sep}=100µs$ are applied at the center of the inhomogeneous line. After spontaneous decay (light red) during a delay of $t_\text{delay}=1.5ms$, a probe pulse (blue) is applied, and the fluorescence (grey) is recorded within a 300µs interval, and averaged over 4.0e3 repetitions (main panel: grey data). A Lorentzian fit (blue curve) yields a FWHM of 1.29 ± 0.10M Hz and an amplitude of -0.48 ± 0.02. b, Inset: The decay of the spectral holes as a function of $t_\text{delay}$ depends on the magnetic field and temperature. At 20mT and 6.5K (light grey data), an exponential fit (solid) gives a spectral hole lifetime of 11.6 ± 3.8ms, while 468.7 ± 104.7ms are obtained at 20mT and 5.15K (dark grey). Main panel: Spectral hole lifetimes as a function of magnetic field (blue data, bottom axis) at 1.3K and at different temperatures (red data, top axis) at 20mT. Fits (solid lines) to phonon-induced relaxation processes allow extracting the coefficients of the Orbach (red) and direct (blue) decay processes. Error bars: 1 S.D. in all panels
• Figure 3: Quantum memory in an Er:Si waveguide. a, Inset: Central part of the prepared AFC. After irradiating 20 burn laser pulses, frequency-modulated to form a regular comb with a period of 3M Hz, the normalized fluorescence after a probe pulse reveals a periodic pattern of spectral holes. Lorentzian fits (blue) reveal a hole width of 0.74 ± 0.07MHz. Main panel: Quantum memory experiment. Retrieved light after storing a faint, 9.9±0.2n s long laser pulse in the erbium ensemble contained in the nanophotonic waveguide, averaged over 7.5e6 measurements. The storage time is pre-programmed by the comb period $\Delta\nu_\text{AFC}$, such that the light is retrieved after $1/\Delta\nu_\text{AFC}$ (orange); a weaker second echo is obtained at $2/\Delta\nu_\text{AFC}$ (orange dashed). b, Memory efficiency. The retrieved light is compared to the input to determine the end-to-end efficiency. The efficiency drops at longer storage times (bottom axis) as the comb period (top axis) approaches the spectral hole width. A fit (red line) to the expected efficiency (eq. \ref{['eq:eff']}) exhibits an optical depth of 2.6 ± 0.1e-3 (see main text). Error bars: 1 S.D. in all panels.
• Figure 4: Coherence measurement. a, Pulse sequence. The memory is probed by sending in faint laser pulses (bottom left, red and dark blue). To demonstrate that the fraction of the pulse (red) that is stored and retrieved from the memory (top) is phase coherent with the input light, we interfere it with a second laser pulse (green) with the same envelope as the storage pulse, sent with a matching delay $t_s$ and a controlled phase $\phi$. b, Phase-dependent readout signal. We program the AFC to a storage time of 0.5µ s and irradiate 9.9±0.2n s long Gaussian pulses; the sequence is averaged over 5e6 repetitions. Yellow: When the interference pulse has the same phase as the input, constructive interference leads to the observation of a strong signal at the programmed delay. Green: When the phase is changed by $\pi$, destructive interference leads to an almost complete disappearance of the signal. c, Continuous phase sweep. Fitting a sinusoidal oscillation (red) to the interference signal integrated over the pulse duration (gray data) gives a visibility of 91.3 ± 3.0%. The green and yellow circles mark the measurements in panel b. Error bars: 1 S.D. in all panels.