Skew effects due to optical path length variation in iterative neural processors

Little attention is normally given to the effects on processor performance of the different optical path lengths required within an iterative processor. The resulting clock skew can have significant degrading effects on the predicted accuracy, stability, and speed of the processor. A similar problem occurs in iterative asynchronous artificial neural networks when, for example, the time delay between two neurons is proportional to their physical separation. The authors have shown that, in the absence of temporal dispersion, certain iterative algorithms have stable steady-state solutions that are independent of clock skew. An analysis is made of the stability criterion for skewed systems and the skew obtained in some commonly used optical processors