Adapting an LCD for weight generation in an electro-optic neural processor

Abstract

This thesis discusses adapting and testing an LCD as a weight image display for use in the Hybrid Electro-optical Neural Network (HENN). The HENN project is a proof of concept prototype hybrid neural network that will be used to gather information for a more advanced project in the future. After thoroughly explaining the HENN, this thesis characterizes the LCD selected for adaption. Within the characterization is a grouping of experiments that explore different aspects of the LCD screen. Then a few experiments are conducted to evaluate the interactions of the LCD and fiber optic interconnection plate. After this, the method used to generate the weighting image is explained thoroughly. The experimental evidence is gathered to show how the LCD can be used as a weighting system. Then based on the evidence gathered several recommendations are suggested to redesign the fiber plate and further improve the HENN system.