Optical vector multiplier for neural networks

Abstract

An optical vector multiplier can perform linear algebra calculations by using electro-optical modulators of sandwich-type construction or in liquid crystal fields such as occur, inter alia, in the case of neural networks. It can calculate linear algebra operations as rapidly as possible, although the matrix is represented by slow liquid crystal fields. In addition, it can also use the transposed matrix, can calculate the vector product of two vectors and store the resultant matrix of the vector product directly in the region of the matrix modulator cells. To this end, two rapid electro-optical modulators of sandwich-type construction, representing a vector, are disposed offset by 90.degree. and a matrix-shaped optical modulator that represents the matrix follows this arrangement. The vector modulator arrangement that is offset by 90.degree. can form the vector product of two vectors, the result being determined by a detector matrix whose detectors are located on the matrix modulator of each modulator cell. Thus the information can be stored locally there and processed.

Dependencies are inferred from the claim text.