, University of South Florida
Fuzzy hardware developments have been a major vehicle in popularizing the applications of fuzzy-set theory and fuzzy logic in both science and engineering. The fuzzy boom in Japan, as well as in other parts of the world, has generated a host of products and techniques demonstrating superior performance to conventional products. It is precisely that superior performance that has enhanced both the scientific label as well as the commercial label of fuzzy-set theory.
An important research trend is the design of improved fuzzy hardware. Recent investigations in the literature have shown increasing interest in both analog and digital implementation of fuzzy controllers in particular and fuzzy systems in general.
Specialized analog and digital VLSI implementations of fuzzy systems, in the form of dedicated architectures, aim at the highest implementation efficiency. This particular efficiency is asserted in terms of processing speed and silicon utilization. This speed of processing has caught the attention of developers of fuzzy hardware and researchers in the field.
This special issue on fuzzy hardware provides you with a comprehensive look at recent works: an analog chip, a processor, digital systems, rule-based systems for use on general-purpose processors, and a controller.
As early as 1980 Takeshi Yamakawa headed a team of designers interested in the design of fuzzy logic circuits fabricated in bipolar technology and assembled with discrete components. 1 In 1984 and 1985 Yamakawa and Miki 2,3 developed the first fuzzy logic chip using PMOS and CMOS technologies, at about the same time that Togai and Watanabe 4,5 developed the first fuzzy inference engine on a VLSI chip. That chip executed 250,000 rules per second. In this issue Tsutomu Miki and Yamakawa discuss first the merits of fuzzy set theory and fuzzy inference mechanisms and then describe several ways to implement them in hardware.
Witold Pedrycz and his team (C. Hart Poskar and Peter Czezowski) propose a reconfigurable fuzzy processor (RFP) for implementing both aggregative (Or, And) and referential (Matching, Difference, Dominance, and Inclusion) operations. This architecture combines both structural and parametric flexibility in the design of a network that implements these processors as fuzzy neurons. RFP neurons contain three main units: the RFP, a learning unit (incorporating a fuzzy back-propagation learning algorithm), and a local memory unit. The authors also implemented a fuzzy neural network as a bidirectionally linked series of shared buses. This implementation facilitates a modular and scalable design environment for the RFP. An appropriate interface, separate from the RFP neuron itself, promotes the reuse of the neuron design with alternative interconnection networks.
The simplicity and versatility of some successful fuzzy-inference algorithms, the advent of high-density, user-programmable logic devices, together with powerful CAD tools, make dedicated digital fuzzy hardware a feasible solution for implementing high-performance fuzzy systems. Donald L. Hung describes the process of this evolution. He also describes fuzzy systems that are based on dedicated digital hardware and thus can deliver improved performance over systems based on general-purpose computing machines.
Hartmut Surmann and Ansgar P. Ungering present several adapted concepts for the implementation of fuzzy-rule-based systems on general-purpose processors. They derived these concepts after developing a third generation of special fuzzy hardware and based them on lookup tables, optimized rule processing, and digital pulse-duration modulation. General-purpose processors can now produce solutions faster than the second generation of special fuzzy processors. Fast solutions are required during the simulation and optimization of fuzzy-rule-based systems with neural networks or genetic algorithms, especially on general-purpose hardware. For an optimally adapted implementation, the authors have analyzed the bottleneck in the fuzzy algorithm depending on the I/O complexity.
Fernando Vidal-Verdú and Angel Rodríguez-Vázquez offer a parallel architecture for fuzzy controllers and a methodology for their realization in the form of analog CMOS chips. These chips can be made to learn through adaptation of some electrically controllable parameters that are guided by a dedicated hardware-compatible learning algorithm. The proposed design methodology emphasizes simplicity at the circuit level—a prerequisite to increasing processor complexity and operation speed. This is illustrated through the use of a three-input, four-rule controller chip in 1.5-μm CMOS, single-poly, double-metal technology.
From a practical point of view, these articles clearly illustrate the importance of fuzzy hardware and its applications to the implementation of fuzzy algorithms and design of fuzzy systems.
The work of the contributing researchers to this special issue lies at the frontiers of fuzzy hardware and its applications. What stands out in their work is the thoroughness of their analysis and their success in formulating interesting and innovative solutions. Not only may we apply these solutions to the conception and design of fuzzy systems but also to a wide variety of other fuzzy problems in science and engineering.
Clearly, the visible success of fuzzy hardware emphasizes that the importance of their work is certain to grow with the passage of time, the accumulation of experience with fuzzy systems, and their implementation via fuzzy hardware. The large number of excellent submissions for this special issue forced some difficult decisions to satisfy constraints on page limitation and the focus of this issue. I am grateful to all the authors for their submissions and to the reviewers for providing excellent feedback and critique. This special issue provides you with the background to try new applications and the motivation to further explore the fascinating and rapidly growing field of fuzzy hardware.