Mixed-Signal Neuro-Fuzzy Processor Implementations: Sequential Architectures and Circuit-Level Description

Abstract

Mixed-signal VLSI design provides a tradeoff between the fast and compact but fixed analog implementations of neuro/fuzzy feedforward algorithms and the programmable but area and power consuming digital counterparts. In this chapter, a sequentiality study of such systems is performed. Basic blocks for sequential mixed-signal neural and fuzzy computing are proposed, and two sequential example processors are described. Feedback from the designed processors and subcircuits allowed considering the technology constraints for analysis and extension to different sequentiality degrees. IntroductionIn this chapter, we introduce architecture analysis and guidelines to develop sequential analog processing circuits. By means of switched techniques and analog memory elements, analog sequential processing circuits can be designed. These techniques are applied to the development of mixed-signal neural and fuzzy processors. For this purpose, the characteristics, constraints and properties of analog circuits, that are very different from their digital counterparts, have to be taken into account. At the architectural level, starting from sequential architectures of neural and fuzzy models, a performance analysis in terms of area and speed as a function of the sequentiality -or parallelism-degree is developed.Efficiency in terms of silicon area occupancy versus computation performance (speed) is fundamental in autonomous intelligent systems. For 93 Neural Networks and Systolic Array Design Downloaded from www.worldscientific.com by MONASH UNIVERSITY on 04/23/17. For personal use only.