Relating Cellular Non-linear Networks to Threshold Logic and Single Instruction Multiple Data Computing Models

This paper examines three apparently different computing models, namely, Threshold Logic (TL), Cellular Non-linear Networks (CNN) and Single Instruction Multiple Data (SIMD). TL is an area of interest in modern VLSI design and computational neuroscience. CNNs are mainly employed in image processing. Conventional SIMD architectures aim at exploiting data parallelism to speed up the execution time of computation intensive algorithms. The scope of this paper is limited to the processing of binary images. Within this scope, the paper conveys three main conclusions. First, the three computing models can be used for binary image processing. Second, not only 2D-CNNs are a sub-class of SIMD architectures, but also synchronous 2D-CNNs with a reduced set of coefficient circuits act as a classical 1-bit SIMD processing element with NEWS (North-East-West- South) for nearest-neighbor communications. Third, TL gates(TLGs) are proved to be an alternative to implement binary 2D-CNNs, leading to on-chip solutions with a very high performance.

Palabras clave: Cellular Non-linear Networks, Threshold Logic, Single Instruction Multiple Data, Hardware Implementation