Computing techniques have been around since antiquity. One of the earliest effective tools—the abacus—appeared around 2400 B.C.E. The first calculator didn’t follow until several thousand years later, during the 17^{th} century. By the dawn of the 20^{th} century, machines that could perform a single task, such as punch clocks and desk calculators, had become commonplace. The term “computer” was coined in the middle of the 20^{th} century, as machines became bigger and more able to perform complex functions.

Today, computers have penetrated nearly every aspect of our lives. As powerful as they are, however, classical computers cannot perform some important tasks. For example, they can’t factor large integers for public key cryptography, which is important for computer security, or simulate quantum mechanics, which helps to deepen scientists’ understanding of quantum physics and leads to the development of new materials. Such tasks could easily take a classical supercomputer hundreds of years. By contrast, a quantum computer might someday be able to do them in a matter of seconds because such computers would exploit the superposition principle and non-classical correlations of quantum mechanics.

Single particles, such as atoms, photons, nuclear spins and charges in superconductors, are fundamental resources for quantum computing, and progress toward quantum computation has been made for all of these particles. Single photons, compared with other approaches, have relatively weak interactions with the environment—which reduces decoherence of the quantum state and makes it more tolerant of various conditions. Especially with linear optics, there are great advantages associated with making high-precision manipulations and measurements of single photons.

## Linear optical quantum computing

When scientists first explored the field of quantum computing, the prevailing belief was that, because of the inefficiency of photodetectors, nonlinear coupling between optical modes was a basic requirement for scalable optical quantum computing. In 2001, Knill, Laflamme and Milburn tried to provide a theoretical proof for such an intuition. Surprisingly, however, they reached an opposite conclusion—that linear optics is sufficient for scalable quantum computing with single photons, with the help of feedback from photodetectors. This was quite a breakthrough, considering that linear optics are much easier to use than nonlinear coupling between optical modes. In the proposal of Knill, Laflamme and Milburn, the basic requirements are single-photon sources, beam splitters, phase shifters, photodetectors and feedback from photodetector outputs.

Like a traditional electronic computer, the Knill, Laflamme and Milburn scheme works by assembling various logic gates to address data flows. The main challenge is to perform a two-qubit entangling gate in the near-deterministic fashion required for scalable quantum computation. To accomplish this, more than 10,000 optical elements are needed to implement a simple two-qubit gate with a 95 percent probability of success. In short, the scheme is not economical in its use of optical components and is thus difficult to implement.

Various alternate approaches have been proposed to reduce the complexity of the Knill, Laflamme and Milburn scheme while improving its theoretical efficiency. Nielsen made a remarkable advance in 2004, when he proposed combining the scheme of Knill and colleagues with cluster-state quantum computation, which had been proposed by Raussendorf and Briegel in 2001. This approach avoids the elaborate teleportation and Z-measurement error correction required in the original scheme, and therefore dramatically decreases the number of optical elements needed to construct nontrivial linear optical gates with a certain probability of success.

Based on these ideas, some experiments have elucidated the feasibility of quantum computing. For example, implementations of nondestructive CNOT gates were reported by Gasparoni et al. in 2004, Zhao et al. in 2005 and Bao et al. in 2007. Walther et al. in 2005 and Prevedel et al. in 2007 reported experimental one-way quantum computers, which were accomplished with four-photon cluster states. Most recently, six-photon cluster states have been realized by Lu and colleagues. More quantum computing experiments will undoubtedly be performed in the near future.

Nevertheless, we would like to point out that single photons—the most fundamental resource in each of these experiments—were all generated probabilistically with spontaneous parametric down-conversion (SPDC). This means that quantum computing based on such a resource cannot be scaled up and thus the complexity of the problems solved is absolutely limited.

An efficient solution is to use quantum memories, in which single photons can either be stored or generated deterministically. Here, we will focus on the atomic ensemble-based quantum memory to elucidate the deterministic generation of single photons. Such an idea originates from a scheme proposed by Duan et al. in 2001 for long-distance quantum communication with atomic ensembles and linear optics. Several groups are currently active in this field, including the Kimble group at Caltech, the Kuzmich team at Georgia Tech, Lukin and colleagues at Harvard, Pan et al. at Heidelberg and the Polzik group at Copenhagen.

## A nondestructive CNOT gate

With the help of an auxiliary entangled photon pair or two auxiliary photons, we can demonstrate a nondestructive CNOT gate for two independent photonic qubits. The CNOT gate is a fundamental logic operation for quantum computing as well as classical computing. It has two qubits: a controlled one *c* and a target one *t*. The logic table of the CNOT operation is given by |0›* _{c}* |0›

*→|0›*

_{t}*|0›*

_{c'}*, |0›*

_{t'}*|1›*

_{c}*→|0›*

_{t}*|1›*

_{c'}*, |1›*

_{t'}*|0›*

_{c}*→|1›*

_{t}*|1›*

_{c'}*, and |1›*

_{t'}*|1›*

_{c}*→|1›*

_{t}*|0›*

_{c'}*.*

_{t'}Optical CNOT gates have now been implemented in several experiments, and can be divided into two groups: destructive and nondestructive. For the former, the output qubits are destroyed when the gate proceeds. Thus, its further application is limited.

For the latter, the qubits are still alive after the gate proceeds and can be used for further operations to achieve efficient LOQC. With a nondestructive CNOT gate, an arbitrary entangled state can be constructed in an efficient way, especially favoring the cluster state used in one-way quantum computation.

Ancilla photons are unavoidably required in building a nondestructive gate. Following the suggestion by Pittman et al. in 2001, Zhao and colleagues demonstrated the full realization of a nondestructive CNOT gate without any noise reduction using a five-photon entanglement. The scheme is shown in part (a) of the figure above, with qubits implemented by the polarization states of photons—i.e., horizontal polarization state |*H*› as logic |1› and the vertical one |*V*› as logic |0›.

With the help of an entangled photon pair, which is in the state (|*H*›|*V*›±|*V*›|*H*›)/√2, the nondestructive CNOT gate between the control and target photons *c* and *t* can be accomplished on the condition of detecting a (|*H*›–|*V*›)/√2 photon in mode 1 and a |*H*› photon in mode 4.

Such a CNOT gate requires an additional entangled state for assistance, and thus the imperfections of the entangled photon pair cause degradation to the fidelity of the gate, making it even more difficult to achieve high-precision gate operation. Recently, Bao et al. used two independent single photons (not entangled) as assistance to demonstrate the nondestructive CNOT gate.

As shown in part (b) of the figure, the two auxiliary photons with polarization state of (|*H*›+|*V*›)/√2 and |*H*› are required. According to the classical information of the jointly measured result of photons 2 and 3, appropriate single-qubit operations are applied on the state of photon 1 and 4 to accomplish the nondestructive CNOT gate between the control and target photons.

Single photons (or photon pairs) are normally generated by SPDC sources. Due to the probabilistic character of the source, the constructed CNOT gate is unsuitable for large-scale quantum computing. Corresponding improvements can be undertaken to achieve deterministic sources.

## A one-way quantum computer

Unlike the standard quantum computation based on sequences of unitary quantum logic gates, the one-way (cluster-state) quantum computer proposed by Raussendorf and Briegel proceeds by a sequence of measurements. This new model requires qubits to be initialized in a highly entangled cluster state. Using CNOT gates, one can exploit the cluster state step-by-step. A sequence of single-qubit measurements (whose order and choices determine the algorithm computed) is then applied, where the choice of later measurement may depend on earlier measurement outcomes. The final result of computation is determined from the classical data of all the measurement outcomes.

However, directly using the CNOT gates to produce cluster states requires too many resources. By exploiting exactly the same setup of entanglement purification by Pan and colleagues in 2003, Walther et al. demonstrated the four-photon cluster state for one-way computing in 2005. A two-qubit gate operation is shown here as an example for one-way quantum computing.

The four photon cluster state can be described as the following:

|Φ

_{cluster}›=(|H›_{1}|H›_{2}|H›_{3}|H›_{4}+|H›_{1}|H›_{2}|V›_{3}|V›_{4}+|V›_{1}|V›_{2}|H›_{3}|H›_{4}–|V›_{1}|V›_{2}|V›_{3}|V›_{4})/2 .

Such a state can be prepared with two pairs of photons generated from an SPDC source, as shown in the top figure on the right of the experimental set-up for producing four-photon cluster states. With the local unitary operation H_{1}⊗*I*_{2}⊗*I*_{3}⊗*H*_{4} on the physical qubits, the four-photon state can be transferred to a horseshoe-like cluster state. (Here, H* _{i}* is a Hadamard operation on qubit

*i*(H

*|*

_{i}*H*›

*=|*

_{i}*H*›

*+|*

_{i}*V*›

*and H*

_{i}*|*

_{i}*V*›

*=|*

_{i}*H*›

*–|*

_{i}*V*›

*),*

_{i}*I*denotes the identity operation, and “⊗” is the symbol for a tensor product.)

_{i}The horseshoe-like diagram denotes the measurement on qubits 2 and 3 and readout on qubits 1 and 4 based on the cluster state |0›_{1}|+›_{2}|0›_{3}|+›_{4} + |0›_{1}|–›_{2}|1›_{3}|+›_{4} + |1›_{1}|–›_{2}|0›_{3}|+›_{4} + |1›_{1}|+›_{2}|1›_{3}|–›_{4}. Such a state performs the following quantum circuit:

|Ψ›

_{out}=(H_{1}⊗H_{2})[R(-α)⊗_{z}R(-β)]CPhase|Ψ›_{z}_{in},

where *R _{z}*(α) is a single-qubit rotation defined as

*R*(α)=exp(-

_{z}*i*ασ

*/2) and the CPhase operation is defined as |*

_{z}*j*›|

*k*›→(-1)

*|*

^{jk}*j*›|

*k*› with (

*j*,

*k*∈0,1). Single-bit measurements on qubits 2 and 3 of the cluster state are equivalent to apply the circuit in the horseshoe-like four-photon cluster state. For an input state |Ψ›

_{in}=|+›

_{1E}|+›

_{2E}and the case where photons 2 and 3 are both measured in the state |+›, the computation is equivalent to the transformation |+›

_{1E}|+›

_{2E}→ (|0›

_{1E}|+›

_{2E}+|1›

_{1E}|–›

_{2E})/√2, where the output is maximally entangled. (Here the numerical subscript labels the logic input qubits, and the subscript E is used to distinguish the logic qubits from the physical qubits.)

The experimentally measured output density matrix of photons 1 and 4 is shown in the figure on the right.

This experiment is not scalable in principle. Fortunately, however, in 2005, Browne and Rudolph proposed a so-called parallel fusion scheme for building up large cluster states. Following such a scalable scheme, most recently, Lu and colleagues exploited a six-photon entanglement source, and generated six-photon cluster states; this will enable more quantum computation protocols to be demonstrated in the near future.

## Deterministic single-photon source

Many kinds of light emitters—for example, single atoms (ions) in an optical cavity, quantum dots in condensed matter and color centers in diamonds—can generate deterministic single photons. Compared with these single emitters, a cloud of atoms can also serve as a single-photon source and have a very striking quality.

The principle of an atomic ensemble-based single-photon source is shown in the figure on the right. The atoms (^{87}Rb) are trapped and cooled by a magneto-optical trap and initially prepared in one of the ground states |g›. A weak write pulse Ω_{W} illuminates the atoms to induce the spontaneous Raman transition. The scattered Raman field *â*_{AS} (anti-Stokes photon) is collected at 3° relative to the propagating direction of the write beam.

Ideally, a single spin excitation can be generated in the atomic ensemble with certainty—as long as one and only one anti-Stokes photon is detected in the single-photon detector. After a controllable time delay *t _{s}*, another classical read pulse with the Rabi frequency Ω

_{R}is applied to retrieve the spin excitation and generate a single photon of Stokes field

*â*

_{S}.

If the excitation probability of the write process is much smaller than 1, the spin excitation generated in the atomic ensemble and retrieved Stokes photon is probabilistic, similar to the entangled photon pairs from the SPDC source. In order to generate the Stokes photon deterministically, the single spin excitation in the atomic ensemble should be generated deterministically at each circle. By taking advantage of the memory, with the help of feedback circuit N(>>1), one can apply independent write sequences on the atomic ensemble until an anti-Stokes photon is detected. Then, in each period of Δ*T*, single photons of the Stokes field can be generated deterministically.

Currently, the overall detection rate of single-photon production is roughly 100 s^{-1}, with the lifetime of the collective state about 15 µs. The production rate is mainly limited by the lifetime of the quantum memory and by the retrieve efficiency during the read process. The lifetime of the collective state can be significantly increased by using a better compensation of the residual magnetic field and field-insensitive clock states. Moreover, by further improving the control circuit, one can apply more write pulses within the lifetime. In the case with a write period of 300 ns, the single photon can be generated with a probability of 95 percent within a lifetime of 300 µs.

## Efficient entanglement generation

Multipartite entangled states are the resources for one-way quantum computing. Thus, we shall describe how to generate a two-photon entangled state in an efficient way with deterministic single-photon source.

As shown in the figure on the right, Alice and Bob prepare their own deterministic single-photon source in parallel. When they agree to keep a single photon for each, they synchronize their read pulses and read out the two photons at the same time. The two deterministic photons overlap at a beam splitter for the two-photon interference.

The indistinguishability of the two independent single photons is verified by means of the measurement of Houg-Ou-Mandel (HOM) interference. The HOM dips are observed both in the time and frequency domains, which are consistent with each other. With the narrow-band property of the retrieved single photons, it is technically easy to synchronize their timing at nanosecond precision and overlap them.

The polarizations of the two retrieved photons are set orthogonally to efficiently generate the entangled state |Ψ^{–}› =(|*H*›_{1}|*V*›_{2}–|*V*›_{1}|*H*›_{2})/√2 when the two photons are emitted at ports 3 and 4, respectively. The entanglement is verified by observing the violation of a Clauser-Horne-Shimony-Holt type Bell’s inequality with *S*=2.37±0.07, where the absence of entanglement would require *S*≤2. Theoretically, the probability *P*_{entangle} for successfully generating the desired state |Ψ^{–}› is 0.5 per run. Given two probabilistic single-photon sources in the same experiment, *P*_{entangle} is 0.5*p*^{2}, where p<<1 is the probability for single-photon production.

In the experiment, *P*_{entangle} with feedback circuits is two orders higher than that without feedback circuits. This is very striking and shows the bright future of the deterministic sources with atomic ensemble based quantum memories.

## A new era with quantum computing

Significant progress has been achieved in the field of quantum computing, both theoretically and experimentally. However, tremendous improvements are still needed before quantum computers can become a practical reality.

One-way quantum computing is, in principle, a promising model. However, the scalability under realistic noise conditions of such a scheme is still an open question.

For the efficient generation of the computing resources, the current deterministic single photon source is still limited by the quality of quantum memories. In order to improve the quality of quantum memories based on atomic ensembles, we should enhance the crucial parameters of lifetime and retrieve efficiency of spin excitations. There are two solutions for the former. The first is to use the so-called “clock” states as the memory levels, which extends the lifetime to about 1 s—five orders of magnitude longer than the current lifetime. The second is to confine the atoms in a dipole trap to avoid escaping the interaction region because of the thermal motion of atoms. The retrieve efficiency can be improved by increasing the optical depth of atomic ensembles; this can also be accomplished by using a dipole trap resulting in an optical depth to 100, which is one order of magnitude higher than the current value and is expected to promote the retrieve efficiency to unity.

Although practical quantum computing is not yet within our grasp, current theories support scalable approaches, and we believe that a new era of quantum computing is closer than ever.

*The authors acknowledge the support from the Deutsche Forschungsgemeinschaft, the Alexander von Humboldt Foundation, the Chinese Academy of Sciences and the National Fundamental Research Program.*

[Zhen-Sheng Yuan and Jian-Wei Pan are working at the Institute of Physics, University of Heidelberg, Germany and Hefei National Laboratory for Physical Sciences at Microscale, China. Yu-Ao Chen and Shuai Chen are with the Institute of Physics, University of Heidelberg, Germany.]

### References and Resources

>> E. Knill, R. Laflamme and G.J. Milburn. Nature **409**, 46 (2001).

>> Z. Zhao et al. Phys. Rev. Lett. **94**, 030501 (2005).

>> Z. Zhao et al. Nature **430**, 54 (2004).

>> X.-H. Bao et al. Phys. Rev. Lett. **98**, 170502 (2007).

>> R. Rossendorf and H.J. Briegel Phys. Rev. Lett. **86**, 5188 (2001).

>> M. A. Nielsen, Phys. Rev. Lett. **93**, 040503 (2003).

>> J.-W. Pan et al. Nature **423**, 417 (2003).

>> P. Walther et al. Nature **434**, 169 (2005).

>> C.-Y. Lu et al. Nature Physics **3**, 91 (2007).

>> L.-M. Duan et al. Nature **414**, 413 (2001).

>> C. W. Chou et al. Science **316**, 1316 (2007).

>> S. Chen et al. Phys. Rev. Lett. **97**, 173004 (2006).

>> T. Chanelière et al. Nature **438**, 833 (2005).

>> M. D. Eisaman et al. Nature **438**, 837 (2005).

>> Z.-S. Yuan et al. Phys. Rev. Lett. **98**, 180503 (2007).