Space-like separation of entangled quantum states is a central concept in fundamental investigations of quantum mechanics and in quantum communication applications. Optical approaches are ubiquitous in the distribution of entanglement because entangled photons are easy to generate and transmit. However, extending this direct distribution beyond a range of a few hundred kilometres to a worldwide network is prohibited by losses associated with scattering, diffraction and absorption during transmission. A proposal to overcome this range limitation is the quantum repeater protocol, which involves the distribution of entangled pairs of optical modes among many quantum memories stationed along the transmission channel. To be effective, the memories must store the quantum information encoded on the optical modes for times that are long compared to the direct optical transmission time of the channel. Here we measure a decoherence rate of 8 × 10(-5) per second over 100 milliseconds, which is the time required for light transmission on a global scale. The measurements were performed on a ground-state hyperfine transition of europium ion dopants in yttrium orthosilicate ((151)Eu(3+):Y2SiO5) using optically detected nuclear magnetic resonance techniques. The observed decoherence rate is at least an order of magnitude lower than that of any other system suitable for an optical quantum memory. Furthermore, by employing dynamic decoupling, a coherence time of 370 ± 60 minutes was achieved at 2 kelvin. It has been almost universally assumed that light is the best long-distance carrier for quantum information. However, the coherence time observed here is long enough that nuclear spins travelling at 9 kilometres per hour in a crystal would have a lower decoherence with distance than light in an optical fibre. This enables some very early approaches to entanglement distribution to be revisited, in particular those in which the spins are transported rather than the light.

Storing and retrieving a quantum state of light on demand, without corrupting the information it carries, is an important challenge in the field of quantum information processing. Classical measurement and reconstruction strategies for storing light must necessarily destroy quantum information as a consequence of the Heisenberg uncertainty principle. There has been significant effort directed towards the development of devices-so-called quantum memories-capable of avoiding this penalty. So far, successful demonstrations of non-classical storage and on-demand recall have used atomic vapours and have been limited to low efficiencies, of less than 17 per cent, using weak quantum states with an average photon number of around one. Here we report a low-noise, highly efficient (up to 69 per cent) quantum memory for light that uses a solid-state medium. The device allows the storage and recall of light more faithfully than is possible using a classical memory, for weak coherent states at the single-photon level through to bright states of up to 500 photons. For input coherent states containing on average 30 photons or fewer, the performance exceeded the no-cloning limit. This guaranteed that more information about the inputs was retrieved from the memory than was left behind or destroyed, a feature that will provide security in communications applications.

We report on the demonstration of light storage for times greater than a second in praseodymium doped Y 2 SiO 5 using electromagnetically induced transparency. The long storage times were enabled by the long coherence times possible for the hyperfine transitions in this material. The use of a solid-state system also enabled operation with the probe and coupling beam counterpropagating, allowing easy separation of the two beams. The efficiency of the storage was low because of the low optical thickness of the sample; as is discussed, this deficiency should be easy to rectify.

We present a simple quantum memory scheme that allows for the storage of a light field in an ensemble of two-level atoms. The technique is analogous to the NMR gradient echo for which the imprinting and recalling of the input field are performed by controlling a linearly varying broadening. Our protocol is perfectly efficient in the limit of high optical depths and the output pulse is emitted in the forward direction. We provide a numerical analysis of the protocol together with an experiment performed in a solid state system. In close agreement with our model, the experiment shows a total efficiency of up to 15%, and a recall efficiency of 26%. We suggest simple realizable improvements for the experiment to surpass the no-cloning limit.