Press "Enter" to skip to content

Researchers Develop Potential New Design for Neural Networks

Researchers from National Institute of Standards and Technology (NIST) devised a novel neural design using light and silicon chip.

The human brain is a complex network of billions of neurons that are interconnected with each other. Several researches tried to replicate the brain’s neural network using circuits of artificial neural networks. However, current electronics that are mostly built using semiconductor circuits impede the extremely complex routing required for useful neural networks. Now researchers at NIST developed a silicon chip capable of distributing optical signals across a miniature brain-like grid with high accuracy. The research published in the journal APL Photonics on July 26, 2018, reports a potential new design for development of neural networks. The research used light instead of electricity as a signaling medium to build the neural networks that previously demonstrated remarkable power in solving complex problems such as pattern recognition and data analysis. Interference in the circuits caused by electrical charge would be eliminated by light. Furthermore, the signals would travel faster and farther than electric signals.

A neural network depends on a network of connections among neurons instead of algorithms that a conventional computer uses to process information. A neural computer comprises a large and complex system of neural networks that can be trained to recognize certain patterns of stimuli. The silicon chip developed by NIST chip vertically stacks two layers of photonic waveguides by using light signals. The waveguides confine light into narrow lines for routing optical signals similar to wires route electrical signals and enable complex routing schemes necessary to mimic neural systems. The researchers stated that the design is capable of extending to incorporate additional waveguiding layers when needed for more complex networks. A software in the design generates signal routing automatically with adjustable levels of connectivity between the neurons. A laser light is directed into the chip through an optical fiber that routed each input to every output group. This created a selected distribution pattern for light intensity or power that could be controlled in two schemes. The uniform scheme in which each output receives the same power and a bell curve distribution scheme in which middle neurons receive the most power.