System Overview

Artemis is a discrete-variable system based on the sampling of multi-photon states from a N-mode universal interferometer (where N will depend on the particular Artemis deployment in use). An overview of the system is shown in the figure below.

There are 3 key stages to any computation on Artemis:

1. Input generation - First the required multi-photon input state is generated using the single photon source (SPS) and demultiplexer. The SPS emits temporally separated photons which the demultiplexer then spatially separates and time delays such that they exist in the same time bin. The exact state generated depends on the provided input of the job, and the demultiplexer is automatically reconfigured to ensure the target input state is realised.

2. Computation - Once prepared, the multi-photon input state then enters the photonic processor. This processor is a universal interferometer, meaning it can implement any NxN target unitary matrix. The chosen unitary matrix will be defined by the particular problem being solved on the system and is calculated within Lightworks. As the photons propagate through the processor they will interfere which each other, creating output states which diverge from those expected using classical light.

3. Measurement - The photon output from the processor is then measured using SNSPDs, which are highly efficient single photon detectors. On Artemis, the detectors are referred to as threshold detectors as they can only resolve up to 1 photon per output mode.

For computation, Artemis supports two paradigms of quantum computing, photonic-native and gate-based. These are discussed more in the following sections.

Photonic native

In photonic-native operation, multi-photon states are input into the system, and the outputs from a target unitary. This is sometimes referred to as boson sampling, which was was first proposed in [AA10]. The interaction of photons as they propagate through the system turns out to create an output probability distribution which is equivalent to the calculation of the permanent for the unitary implemented by the interferometer. The permanent of a matrix is computationally hard, and so this potentially offers a route towards quantum advantage for large enough systems.

Gate-based

In gate-based operation, a spatial dual-rail encoding is used to encode a qubit as a photon existing across a pair of photonic modes. This is further explained in Gate-based computation. Using this dual-rail encoding, it is possible to realise any single qubit gates with the existing linear optic components in a circuit.

Multi-qubit gates are trickier, due to the limited nature of photon-photon interactions, however it turns out to be possible to probabilistically implement some gates with linear optics. Two of the key proposals for this originate from [RLBW02] and [KLM01]. The former describes a 6 mode system which is able to implement a CNOT gate with probability of 1/9 through the use of a set of post-selection rules on the output. The latter is an 8 mode system, which requires the introduction of two additional heralding photons, and these are used to herald the gate success with a probability of 1/16. These additional heralding photons provide the advantage that the qubits themselves do not need to be measured for gate success, and can therefore progress to be used in future operations.