Abstract

Multiple quantum well electrooptic (EO) devices based on the quantum confined Stark effect have been studied widely. Asymmetric Fabry-Perot (AFP)¹ devices in these materials have shown promise as normal incidence devices allowing freedom for generating device arrays. We report on an array of EO AFP reflection devices that has been designed and fabricated specifically for use in a second-order neural network system. The device structure consists of 70 gallium arsenide wells 90 Å in width and 60-Å aluminum/gallium arsenide barriers within a PIN structure and an AFP resonator with back and front reflectivities of 98% and 30%, respectively. A linear array of twenty-five devices each 100 mm wide and 2.5 mm long form a 2.5-mm square of active device area, representing a binary vector with twenty-five elements. By having a double pass on the device array and rotating the device image by 90° between passes, all the terms of the autocorrelation matrix (vector cross product with itself) are generated.² This is a basic requirement for any second-order neural network.

© 1991 Optical Society of America