Building a Quantum Internet
Author: Joshua Nunn
Affiliation: Clarendon Laboratory, Department of Physics, Oxford University, UK
We all grew up in a post-industrial society; you are reading this thanks to a communications revolution. Now we are on the cusp of the era of quantum information. Quantum information is a new kind of information that is more powerful than the digital bits found inside today's computers . A quantum bit (qubit) can exist as both a zero and a one simultaneously. Therefore a single qubit has twice the computational power of a conventional bit. A pair of qubits has four times the power of its classical version. Twenty qubits are a million times more powerful! If technologies based on quantum information can be scaled up to work reliably with many qubits, the fields of telecommunications and computation, as well as metrology (i.e. sensing) and imaging will be changed forever. There is currently a global research effort to develop these technologies to produce a viable quantum internet (see Fig. 1), with qubits beamed across the world carrying quantum information at high speed .
Qubits can be made using photons -- single particles of light. As made famous in the `double slit experiment' , a single photon can be delocalized, meaning that it exists in two places at once; this is exactly the sort of super-position property that characterizes a qubit. Therefore the future of quantum information technology rests on the ability to manipulate single photons. In particular, a key ingredient of any future technology is a quantum memory . Just as a classical computer needs memory in order to shuffle information around and store it, so any quantum processor will need a way to hold quantum information -- photons -- for controlled periods of time. In fact, quantum information processing doesn't work deterministically like its classical counterpart. Instead, it succeeds only probabilistically, because the behaviour of qubits in superposition states can never be predicted perfectly.
If the success probability of one processor is p, then without a memory, the probability of two processors succeeding is p2; for N processors it is pN. Thus the success probability becomes smaller at an exponential rate in the number of processors, which essentially means a large quantum computer will never work. But if a memory is available, this changes the game. Now a successful processing step can be stored while other processors continue to work. This changes the success probability of each processor to Mp, where M is the number of attempts that can be tried within the lifetime of the memory. As M approaches 1/p, we end up with near 100% efficiency for each processor, meaning that large numbers of them can be combined without encountering the exponential scaling disaster. In this context, quantum memories for storing photons are the vital, enabling technology which will bring quantum information processing out of the laboratory and into our homes and businesses!
Figure 1: The quantum internet. In the future, could qubits (|ψ>) carrying powerful quantum information be transmitted as easily as digital information is today? The key ingredient is a quantum memory for storing photons.
We have been working on a quantum memory for photons based on a process known as Raman scattering, which is (i) fast and efficient (ii) clean, adding very little noise, and (iii) works with many materials at many different frequencies, and at room temperature .
Raman scattering is inelastic scattering, where a beam of light passes through a collection of atoms, leaving some energy behind and emerging with a lower
frequency. Normally, atoms and light only exchange energy when the light hits a resonance in the atoms, in the same way that you can't push a swing too fast or too slow; you have to hit the right pushing frequency. But Raman scattering is different: it occurs for light frequencies that are far from resonance, and this is crucial. The problem with resonances is that they work both ways -- the swing pushes back on you. If you excite an atomic resonance with light, that atom will also emit light. This process is what makes res glow and lasers lase. If you are trying to make a quantum memory which stores light, then using a resonant frequency is problematic, because the stored light will be re-emitted by the atoms and lost. Even worse, when you attempt to retrieve the light you have stored, the atoms will mix in some of their `spontaneously emitted' radiation, and this will show up as noise -- distortions in the quantum information you have stored. Raman scattering provides a way to transfer energy into the atoms without exciting a resonance, avoiding the noise and losses introduced by spontaneous emission.
Figure 2: A photo of our cesium cell, glowing because we are sending too much laser light into it!
To demonstrate the low-noise properties of this idea, we built a Raman quantum memory using a small cell filled with cesium vapour (see Fig. 2). Cesium is the element used to make atomic clocks, and the second is now internationally defined as "9,192,631,770 oscillations of the cesium atom". The simple structure of cesium makes it ideal to implement our memory. It has three energy levels, shown in part (b) of Fig. 3. The two levels at the bottom are the `clock levels', so they are separated by the famous 9.2 GHz. The level at the top is the resonance that interacts with light -- but this is the resonance we want to avoid. To make the memory work, we send in a weak signal, containing approximately one photon, with a frequency tuned away from the resonance (wiggly line in Fig. 3(b)). To stimulate the onset of Raman scattering, we send in a strong write pulse tuned to the lower frequency produced by the Raman process. The signal photon is absorbed and at the same time one of the atoms in the memory flips into the upper clock state, which corresponds physically to the cesium nucleus swapping its direction of spin. This `spin flip' is now a stationary qubit (it's not going anywhere!); we have succeeded in transferring the quantum information in the photon into our memory. After waiting a few hundred nanoseconds (this sounds short but it's a long time for the atoms) we can flip the spin back again and retrieve our stored photon by sending in another strong pulse, similar to the write pulse. This read pulse stimulates the reverse interaction, causing energy to be removed from the atoms and producing our original photon again.
Figure 3: A Raman quantum memory. (a) A weak signal field containing about one photon is sent into a cloud of cesium atoms, where it is stored by a strong write pulse. Some time later, the stored signal can be retrieved by sending in a read pulse. (b) The simple three-level structure of each cesium atom. Both the signal and control fields are tuned away from the noisy resonance, but the energy in the signal is still transferred to the `clock states' through Raman scattering, which enables the storage of the quantum information carried by the signal.
If the interaction is noisy, it is possible to retrieve a photon even when you didn't store one in the first place. Previously, resonant memories have been too noisy to operate with single photons at room temperature . We hoped that our Raman memory would perform better. To test the noise floor of our memory, we counted how often we retrieved a spurious photon when we stored no signal. It turns out that with our simple Raman memory, the signal is perfectly clean 75% of the time. So, there is some noise, but it is small enough to operate our Raman quantum memory at room temperature and effectively store quantum information (single photons). This, combined with the high speed of our memory (we are able to store photons with pulse lengths less than a third of a nanosecond long), the long storage time (we can store pulses for up to a microsecond, three thousand times longer than the pulses!) and the high efficiency (we have achieved 50% storage and 70% retrieval efficiencies), suggests that this type of technology could indeed form the basis for the large scale, ultra-powerful quantum networks of the future.
We are excited by this new technology, and we are trying to develop it to improve its performance. We want to make the storage and retrieval efficiencies higher (we believe we can get above 90% for both, with more laser power), and the storage time longer (we believe we can store for hundreds of microseconds if we carefully shield the memory from magnetic fields). We also believe we can reduce the noise even further; down to less than 5%, by carefully filtering the frequencies and spatial shape of the beams (some of the noise `gives itself away' by looking different to the true signal). The next steps are to demonstrate this memory inside a simple quantum processor, to show how it can really be used to improve the scaling and make quantum information usable on a large scale.
Let's hope it works!
 M. A. Nielsen and I. L. Chuang. "Quantum Computation and Quantum Information". (Cambridge University Press, 2004).
 R. Ursin, T. Jennewein, J. Koer, JM Perdigues, L. Cacciapuoti, CJ de Matos, M. Aspelmeyer, A. Valencia, T. Scheidl, A. Acin, et al. "Spacequest, experiments with quantum entanglement in space." Europhysics News, 40, 26-29 (2009). Abstract.
 G.I. Taylor. "Interference fringes with feeble light". In Proceedings of the Cambridge Philosophical Society, volume 15, pages 114-115 (1909).
 A.I. Lvovsky, B.C. Sanders, and W. Tittel. "Optical quantum memory". Nature Photonics, 3, 706-714 (2009). Abstract.
 K. F. Reim, P. Michelberger, K. C. Lee, J. Nunn, N. K. Langford, and I.A. Walmsley. "Single-photon-level quantum memory at room temperature". Phys. Rev. Lett., 107, 053603, (Jul 2011). Abstract.
 S. Manz, T. Fernholz, J. Schmiedmayer, and J.W. Pan. "Collisional decoherence during writing and reading quantum states". Physical Review A, 75(4):40101 (2007). Abstract.