An optical neural network is a physical implementation of an artificial neural network with optical components. Early optical neural networks used a photorefractive Volume hologram to interconnect arrays of input neurons to arrays of output with synaptic weights in proportion to the multiplexed hologram's strength. Volume holograms were further multiplexed using spectral hole burning to add one dimension of wavelength to space to achieve four dimensional interconnects of two dimensional arrays of neural inputs and outputs. This research led to extensive research on alternative methods using the strength of the optical interconnect for implementing neuronal communications. Some artificial neural networks that have been implemented as optical neural networks include the Hopfield neural network and the Kohonen self-organizing map with liquid crystal spatial light modulators Optical neural networks can also be based on the principles of neuromorphic engineering, creating neuromorphic photonic systems. Typically, these systems encode information in the networks using spikes, mimicking the functionality of spiking neural networks in optical and photonic hardware. Photonic devices that have demonstrated neuromorphic functionalities include (among others) vertical-cavity surface-emitting lasers, integrated photonic modulators, optoelectronic systems based on superconducting Josephson junctions or systems based on resonant tunnelling diodes.
All-optical nonlinear activation In a multilayer neural network, linear weighted-sum operations are generally followed by a nonlinear activation function. Optical systems can implement linear transformations using interference, diffraction, resonators, or wavelength multiplexing, but many optical neural networks convert the optical output into an electrical signal to apply the activation function before modulating a new optical signal. An all-optical nonlinear activation function instead maps an optical input directly to an optical output through an intensity-dependent or field-dependent physical response, reducing repeated optical-to-electrical and electrical-to-optical conversion between network layers. Physical mechanisms investigated for all-optical activation include two-photon absorption, the Kerr effect, free-carrier effects, saturable and reverse saturable absorption, optical bistability, gain saturation, phase transitions, and nonlinear interactions in atomic media. Implementations have used nonlinear waveguides, Mach–Zehnder interferometers, microring resonators, semiconductor lasers, two-dimensional and phase-change materials, and atomic systems. Depending on the device and its operating point, the resulting optical transfer curve can approximate rectified-linear, sigmoid, threshold, radial-basis, or saturating activation functions. Representative experiments include an all-optical neural network using electromagnetically induced transparency in laser-cooled atoms, a reconfigurable silicon-photonic device based on a cavity-loaded Mach–Zehnder interferometer, and a silicon waveguide incorporating titanium–gold split-ring resonators to produce a slow-light-enhanced two-photon-absorption response. Performance is evaluated using characteristics such as activation-curve shape, threshold optical power or energy, response speed, bandwidth, insertion loss, extinction ratio, footprint, reconfigurability, thermal stability, fabrication compatibility, fan-out, and cascadability. Resonators and slow-light structures can enhance weak optical nonlinearities but may narrow the operating bandwidth and increase sensitivity to fabrication variation or temperature. Absorptive mechanisms can reduce the optical power available to later layers, while active devices can provide gain at the cost of additional energy consumption and noise.
Electrochemical vs. optical neural networks Biological neural networks function on an electrochemical basis, while optical neural networks use electromagnetic waves. Optical interfaces to biological neural networks can be created with optogenetics, but is not the same as an optical neural networks. In biological neural networks there exist a lot of different mechanisms for dynamically changing the state of the neurons, these include short-term and long-term synaptic plasticity. Synaptic plasticity is among the electrophysiological phenomena used to control the efficiency of synaptic transmission, long-term for learning and memory, and short-term for short transient changes in synaptic transmission efficiency. Implementing this with optical components is difficult, and ideally requires advanced photonic materials. Properties that might be desirable in photonic materials for optical neural networks include the ability to change their efficiency of transmitting light, based on the intensity of incoming light.
… excerpt ends here. Continue reading the full article.
