Search

Evolvable hardware (EHW) has become a promising branch of research in the area of hardware architecture. However, there remain many issues to be tackled before EHW can be applied more extensively in real-time problems. The three major open issues are on-line adaptation, scalability and evolution termination. In this thesis, we propose the concept of an intrinsic evolvable and on-line adaptive evolvable fuzzy hardware (EFH). Adopting our proposed EFH, the three major open problems can be addressed. In particular, EFH has great potential for solving real-time control problems efficiently. In order to demonstrate our idea, we apply the EFH to tackle the problem of ATM cell-scheduling. DOCTOR OF PHILOSOPHY (EEE)

The computation bandwidth offered by current integrated circuits IC provides the capacity for an IC chip to handle multiple tasks or applications. In this paper, we present the concept and design of a multi-task fuzzy inference processor MtFIP. Earlier, a reconfigurable fuzzy inference chip RcFIP with 17 mega fuzzy logic inference per second FLIPS capable of supporting real-time context-switching has been proposed. High performance similar to that of the RcFIP is attainable by the MtFIP. Each task is handled by a RcFIP configured specific to the domain of the application. High flexibility is achieved by integrating several RcFIPs in the MtFIP. The compact architecture of the MtFIP allows for seamless online context-switching among multiple tasks. The hardware architecture of the MtFIP is described in details.

This paper describes a novel design of a fuzzy inference chip that allows for real-time online context switching. A context refers to a situation or scenario of an application requiring specific domain knowledge. In particular, our focus is on the class of applications involving embedded fuzzy control. The domain knowledge therefore refers to fuzzy rules and memberships. The kind of applications being considered is real-time in nature, which necessitates the implementation of hardware for fuzzy inferencing. The chip architecture is described and details on the design of the chip is presented. Published version

We present a multi-task fuzzy inference processor (MtFIP) in this paper. A reconfigurable fuzzy inference chip (RcFIP) with 17 mega fuzzy logic inference per second (FLIPS) capable of supporting real time context switching has been proposed in our earlier work. High performance similar to that of the RcFIP is attainable by the MtFIP. Each task is executed by a RcFIP configured specific to the domain of the application. High flexibility is achieved by integrating several RcFIPs in the MtFIP. The compact architecture of the MtFIP allows for seamless online context switching among multiple tasks. The hardware architecture of the MtFIP is described in detail in this paper. Published version

Evolvable fuzzy system (EFS) relies on dynamic adaptation of the heuristics which are coded as fuzzy rules. If the mechanisms for fuzzy inferencing are realized in hardware within the framework of the EFS in order to satisfy the time-critical requirement of the application, the system can be regarded as a form of evolvable fuzzy hardware. This chapter considers the application of evolutionary fuzzy system for cell scheduling in ATM network. Besides being able to adapt to the changing environment, the EFS also enjoys a special property whereby the cell delay can be conveniently adjusted by explicitly tuning a parameter in the fitness function. In this chapter, we explore the application of EFS to solve multiple channel scheduling. We show that any number of input channels can be accommodated while still preserving the special advantages of the EFS.

In this chapter, we describe a scheme to realize an Evolvable Fuzzy Hardware (EFH) for real-time Packet Switching problem. The common challenges of Evolvable hardware (EHW) implementation are issues pertaining to online adaptation, scalability and termination of evolution [I]. The proposed EFH addresses these issues effectively. A very interesting advantage of the proposed EFH is that the system performance can be tuned intuitively through parametric adjustment of the fitness function. This advantage gives the EFH system a very special property that conventional scheduling methods cannot fulfill easily. For the hardware implementation of the EFH, real-time fuzzy inference with high-speed context switching capability is necessary. We address this aspect through implementation based on a context independent reconfigurable fuzz9 inference chip (RFIC).

To achieve real-time performance for some applications of fuzzy systems it is necessary to realise the inference module in hardware. By the same token, real-time applications of evolvable fuzzy hardware systems require the realisation of systems as hardware that can be conveniently reconfigured. A good example of such a scenario is illustrated for the case of asynchronous transfer mode cell scheduling. Since the domain rule-set is expected to evolve during normal operation, it is imperative that the hardware supporting the fuzzy reasoning should be easily reconfigurable. To address this requirement a reconfigurable fuzzy inference chip (RFIC) is described that supports online rule-set reconfigurability. The overall methodology of designing evolvable fuzzy hardware systems based on an RFIC as the core inference processor is presented. The RFIC is an in-system context-switchable inference processor for evolvable fuzzy hardware. Hardware architectural details of the RFIC are presented and its performance simulated to verify its functionality and applicability.