Currently, there is a surge in network traffic. The demand for high-capacity communications is fueled by a slew of new services and applications, including cloud computing, video streaming platforms, and the IoT (Internet of Things). Optical communication systems, which use fibers to transmit data, are the foundation of today's communication networks, including fixed-line, wireless infrastructure, and data centers.
A group of researchers have developed new neural network hardware that could assist overcome this constraint by accounting for fiber nonlinearity's negative impacts. It is built on a silicon-based photonic-electronic system with a few neurons that can, in theory, beat conventional DSP processors in terms of throughput, latency, and energy consumption. The network could aid in developing the Submarine Market.
The study of 'neuromorphic photonics' began with a breakthrough. Photonic devices and biological neurons are both regulated by the same differential equations, but "photonic neurons" have a time scale of around picosecond to nanosecond. In contrast, biological neurons have a time range of roughly one millisecond.
Their previous work inspired the team to start developing high-performance photonics-based neuromorphic technology. This hardware would run artificial neural networks in nanoseconds in an ideal world, making it substantially quicker than current electronics-based systems.
Researchers built a photonic neural network using high-quality waveguides, photonic components like photodetectors, and modulators. These were initially designed for use in optical communications. This allows the network to sustain fiber transmission rates, potentially enabling real-time processing with uniquely designed optical networks.
The researchers' silicon neural network is programmable and is based on the so-called broadcast-and-weight protocol, which they first described in one of their previous works.
The team's approach modifies signal transmission through a filter by tweaking the filter's transmission edge, effectively multiplying signals with a chosen weight. The 'weighted' signals are then transferred to a photodetector that can receive numerous wavelengths in parallel and add them together.
The photocurrent created by the team drives an optical modulator during the first step. This helps convert the electrical photocurrent into optical power. The idea denotes that optical modulators in the team's photonic network take on nonlinear activation functions, acting as artificial neurons.
The new neural network developed by this academic team could help improve the performance of optical communication equipment in the future. Their network's sole purpose was to correct signal distortions in a single wavelength channel. They believe it could be used in many WDM optical fibre networks.
Related Reports:
Global Submarine Battery Market Growth 2020-2025
Global Submarine Propulsion Systems Market 2021 by Manufacturers, Regions, Type and Application, Forecast to 2026
Global Submarine Fiber Cable Market Trends, Growth, Manufacturers Analysis Report 2020 By Type, Application and Forecast To 2025
Global Submarine Optical Fiber Cables Market 2020 by Company, Regions, Type and Application, Forecast to 2025
