Optical State Encoding

Optical State Encoding

In the Q-Memory photonic platform, information is encoded as the phase and routing of photons through the programmable optical network. This page explains how that encoding works, how quantum information and classical computation both use the same hardware, and how non-volatile optical memory preserves state between computations.

Information as Light

In a conventional computer, information is stored as voltage levels — high or low — representing 0 and 1. In a photonic quantum system, information is encoded differently:

• Quantum information is carried by individual photons; a single photon can be in a superposition of multiple paths simultaneously — this is the quantum bit (qubit)
• Classical computation (including AI matrix operations) is performed by encoding numerical values as the amplitude and phase of optical signals and exploiting interference in the beam splitter network

Both modes of operation use the same programmable optical mesh. The difference lies in the input: quantum operations use individual photons, while AI acceleration uses coherent optical signals.

Phase Encoding

The programmable elements in the optical network — phase shifters and beam splitters — implement transformations by controlling the relative phase of light in each waveguide arm.

Each phase element can be set to any value between 0 and 2π, with the precision determining the resolution of the computation:

• Quantum gate accuracy: Phase precision of better than a few milliradians is required for high-fidelity quantum operations
• AI weight precision: Phase precision maps to numerical precision; the platform supports multiple bits of effective precision per element

Phase values can be set by:

1. Thermal phase shifting: A resistive heater changes the local temperature of the waveguide, which shifts its refractive index and therefore the optical phase. Well-characterised and widely used; requires continuous power to hold a value.

2. Electro-optic phase shifting: An applied voltage directly shifts the refractive index through the electro-optic effect. Nanosecond response time; essential for real-time feed-forward operations during quantum computation.

3. Non-volatile optical memory: A material is switched between two or more stable structural states by brief optical or electrical pulses. The state persists indefinitely without any applied power. The element holds its programmed phase value with zero static energy consumption.

Non-Volatile Optical Memory: Multi-Level State

The optical memory elements used in Phase 1+ can hold multiple distinct states, not just two. This multi-level capability is important for two reasons:

1. AI weight storage: Neural network weights require fine-grained numerical precision. Multi-level optical memory elements store more bits of effective precision per element, reducing the number of elements needed to represent a given weight.

2. Reduced re-programming frequency: Fine-grained states mean the element rarely needs to be updated — it can hold the programmed value through many thousands of computation cycles.

The number of distinguishable states per element is determined by the optical contrast between the material’s structural configurations and the noise floor of the readout system. Demonstrated values in research literature range from tens to over a hundred levels per element.

Quantum State Encoding

For quantum computation, the encoding follows the rules of linear optical quantum computing:

• Each photon occupies one of N waveguide modes simultaneously (in superposition)
• The optical mesh implements a unitary transformation on the N-mode state
• Entanglement between photons is created by joint measurement (fusion) at beam splitters
• The measurement outcome is intrinsically quantum random — this randomness is used for quantum key distribution and certified random number generation

Path Encoding

A photon in mode $i$ is a quantum bit encoded as: which of N waveguide paths the photon travels. The mesh transforms this by routing and interfering photons across paths.

Dual-Rail Encoding

Two waveguide modes represent one logical qubit: the photon being in the first mode represents $|0\rangle$, the second represents $|1\rangle$, and a superposition represents a general qubit state. This encoding is robust to photon loss detection and is used in fault-tolerant photonic quantum computing protocols.

AI Matrix Encoding

For neural network acceleration, input vectors are encoded as the amplitudes of optical signals injected into the N input ports of the mesh. The mesh — programmed to represent a weight matrix — transforms the input signals through optical interference. The output amplitudes, measured at the N output ports, give the result of the matrix-vector multiplication.

This encoding is exact: the physics of optical interference implements the matrix multiplication directly, without approximation.

Encoding property	Value
Matrix size	N × N (N = mode count)
Computation time	Constant — set by photon transit time across chip
Precision	Determined by phase element precision and detector resolution
Reprogramming speed	Microseconds (thermal) to nanoseconds (electro-optic)

Error Correction and Calibration

Thermal Drift Compensation

Phase elements change slightly with temperature fluctuations. The CMOS control electronics include a calibration loop that:

1. Periodically measures the optical transmission through reference paths on the chip
2. Computes the drift in each element’s phase
3. Applies corrections to maintain the programmed values within tolerance

This is particularly important for thermal phase shifters, which are sensitive to ambient temperature changes. The Phase 0 target is less than 5 milliradians per minute of drift — characterised during the validation tests.

Photon Number Resolving

The detection system can distinguish how many photons arrived at a given output port in a given time window — not just whether any photon arrived. This photon number resolution (PNR) capability is essential for:

• Distinguishing successful two-photon fusion events from single-photon noise
• Implementing more sophisticated quantum error correction protocols
• Accurately measuring optical signal amplitudes for AI computation readout

Quantum Error Correction

For fault-tolerant quantum computing, the photonic platform uses measurement-based error correction — where errors are detected through a pattern of measurement outcomes across the chip, and corrections are applied to subsequent operations. The CMOS feed-forward electronics implement this correction in real time, within the coherence window of the photonic state.