Since the advent of digital computing, scientists have imagined creating artificial neural networks that would handle challenging issues by functioning like biological brains. Researchers have attempted to build neural networks that would operate at the speed of light as nanophotonic circuits became a reality. However, translating the nonlinear activation function, a crucial mathematical element of artificial neurons, from the electronic to the optical world proved challenging.
An all-optical artificial neural network has now been developed and used in a sophisticated demonstration by a team. The network combines linear functions driven by spatial light modulators with nonlinear activation functions based on electromagnetically induced transparency, a quantum interference phenomenon. (EIT).
The inputs and outputs of one layer become the inputs of the subsequent layer in artificial neural networks, which usually have interconnected layers. Of course, the goal is to mimic the intricate links between biological neurons and axons. The Hong Kong researchers used two spatial light modulators, a Fourier lens, and a flip mirror, to send an incoming coupling laser beam through for their network’s linear functions. The Fourier lens, which is programmable, changed the laser impulses.
The nonlinear optical components required powerful lasers for earlier attempts to make optical activation functions.
Related Content: Light Speed Machine Learning With Photonics