Title :
Optical implementation of Weber-Law neurons based on the dynamic behavior of electron trapping material
Author :
Athale, Ravindra A. ; Yang, Xiangyang ; Szu, Harold
Author_Institution :
Dept. of Electr. & Comput. Eng., George Mason Univ., Fairfax, VA, USA
Abstract :
An approach to realizing the neuron transfer function described by the Weber-Fechner law by exploiting the dynamical behavior of electron trapping (ET) materials is described. The interaction between the electron trap density and the blue light and near infrared light intensities closely resembles that required by the law of mass action. The scheme can realize the complex dynamical behavior of the neuron in a single ET thin film in a spatially continuous manner. This feature can result in a higher density of neurons compared to the approach which combines discrete subcells performing the constituent operations. An analytical description of the neural net STM equations, the equations describing ET dynamics, and initial experimental data is given. The overall optical system can be implemented in a compact manner and with input-output compatibility with analog optical storage encoding long-term memory storage. That, in turn, can lead to more sophisticated optical multilayer neural net systems based on adaptive resonance theory models
Keywords :
electron traps; optical neural nets; optical storage; transfer functions; ART model; ET dynamics; Weber-Fechner law; Weber-Law neurons; adaptive resonance theory; analog optical storage encoding; blue light; complex dynamical behavior; electron trap density; electron trapping material; electron trapping thin film; input-output compatibility; long-term memory storage; near infrared light intensities; neural net STM equations; neuron transfer function; optical multilayer neural net systems; optical system; short term memory; Electron optics; Electron traps; Encoding; Equations; Neural networks; Neurons; Optical films; Optical materials; Transfer functions; Transistors;
Conference_Titel :
Neural Networks, 1992. IJCNN., International Joint Conference on
Conference_Location :
Baltimore, MD
Print_ISBN :
0-7803-0559-0
DOI :
10.1109/IJCNN.1992.287217