Optical computing is a form of computing in which light, rather than electrical signals alone, is used to represent, transmit, manipulate or process information. It exploits properties of electromagnetic waves such as amplitude, phase, wavelength, polarisation, interference and diffraction to perform computational or communication functions.
Optical computing can refer to systems in which the principal computation occurs optically, to hybrid electro-optical systems in which optical components perform selected operations while electronic circuits control the system, or to optical interconnects that accelerate the movement of information without necessarily performing the computation itself. The idea has existed for decades, with early research investigating analogue optical processors, holographic techniques and optical logic. More recent interest has been driven by the computational demands of artificial intelligence, where moving large quantities of data between memory and processing units can consume substantial energy and bandwidth.
One important property of optics is that light can propagate through a medium while carrying information in multiple physical dimensions. Different wavelengths can coexist in the same optical channel, enabling wavelength-division multiplexing, while different spatial modes, polarisations or phases can encode additional information. Optical systems can also exploit interference, in which electromagnetic waves combine according to their relative phase and amplitude.
Diffraction and the behaviour of lenses allow optical systems to perform operations that correspond mathematically to transformations such as Fourier transforms. These properties make optics attractive for certain linear-algebra operations, signal-processing tasks and high-bandwidth communications. In an optical neural-network accelerator, for example, an optical circuit can be engineered so that the propagation of light through interferometers or other structures performs weighted combinations of input signals. The resulting computation can potentially occur with very high throughput because many optical signals propagate simultaneously.
Optical computing is often discussed in connection with analogue computation. Instead of representing every intermediate result as a sequence of binary electronic operations, an optical system may encode a numerical quantity in a physical property of light and transform it continuously. This can make some mathematical operations extremely efficient, particularly linear transformations. However, analogue computation introduces precision, noise and calibration challenges. Optical devices are subject to fabrication tolerances, temperature changes, losses, nonlinearities and imperfections in optical components.
Converting information between the electrical and optical domains can also introduce overhead, meaning that the benefits of an optical accelerator depend on the entire system rather than on the optical operation alone. Digital optical logic has also been investigated, but building a general-purpose optical computer that replaces conventional digital electronics is much more difficult than using optics for specialised operations.
Optical matrix multiplication is a major modern application area because matrix operations are fundamental to neural networks. A system can encode input values as optical signals and configure an optical network so that interference and transmission through weighted components implement a matrix transformation. Mach-Zehnder interferometer meshes, microring resonators, diffractive optical elements and other photonic structures can be used to implement variants of these operations.
Optical multiplication is especially attractive when many operations can be performed simultaneously through parallel propagation of light. Some architectures use coherent light and manipulate optical phase, while others use intensity-based or incoherent approaches. Following the optical operation, photodetectors convert the result back into electrical signals if subsequent processing is electronic. The resulting architecture is therefore often a photonic accelerator or electro-optical accelerator rather than a completely optical general-purpose computer.
Optical computing has important relationships with optical communications and photonics, but these terms describe different functions. Optical communications uses light primarily to transmit information, as in fibre-optic networks, whereas optical computing uses optical phenomena to perform transformations or calculations. Photonics is the broader technological field concerned with generating, controlling, transmitting and detecting light. Optical interconnects can improve the communication capacity between electronic processors and memory without making the computation itself optical.
Silicon photonics is particularly important because it integrates optical components with semiconductor manufacturing technologies, potentially allowing compact optical devices to be manufactured alongside electronic circuits. Integrated photonics can include waveguides, modulators, photodetectors, lasers or laser interfaces and passive optical components on a chip or package.
Historically, optical computing research included analogue systems for image processing, pattern recognition and signal manipulation. Optical correlators and holographic systems exploited the natural ability of lenses and diffraction to perform certain transformations rapidly. The rise of inexpensive digital electronics reduced the commercial attractiveness of many general optical processors because electronic computers became extremely flexible and programmable. However, optical technologies retained a strong position in telecommunications because fibre optics provide enormous bandwidth over long distances.
Current interest in optical computing has therefore often emerged from the opposite direction: rather than replacing electronics entirely, researchers seek to integrate optical processing with electronic control, memory and software. This hybrid model can target specific bottlenecks such as data movement, matrix multiplication and high-bandwidth processor-to-processor communication.
Optical computing has several potential advantages, including high bandwidth, natural parallelism, potentially low energy consumption for particular operations and the ability to perform certain transformations at very high speed. Light can travel through passive optical structures without the same resistive losses associated with long electrical interconnects, although complete systems still consume energy in sources, modulators, detectors, control electronics and cooling. Optical systems can also process multiple channels simultaneously using wavelength or spatial multiplexing.
Their limitations include optical loss, component precision, thermal sensitivity, manufacturing complexity, nonlinear-device requirements, analogue noise, limited programmability and the need for electrical conversion in many practical architectures. Memory is another challenge because storing information optically is generally less straightforward than storing electrical bits in established semiconductor memory technologies. These constraints make specialised acceleration and communication particularly practical targets.
The significance of optical computing lies less in the prospect of replacing every conventional computer with an all-optical machine and more in the possibility of combining photonics with electronics to overcome particular scaling constraints. Artificial intelligence, high-performance computing, telecommunications, scientific instrumentation, signal processing and data-centre interconnects are among the areas where photonic techniques may provide useful advantages. Optical neural networks can potentially perform large linear transformations efficiently, while optical interconnects can provide high-bandwidth links between processors.
The field includes both established technologies, such as optical communication, and experimental technologies, such as programmable photonic processors. Optical computing therefore describes a broad family of techniques rather than one standard computer architecture, with the practical value of each approach depending on the computation, required precision, data movement, manufacturing technology and overall system design.
