Analog Integrated Fuzzy Implication System Design for Interdisciplinary Applications
Evangelos Georgakilas
Abstract
The purpose of this thesis is the design of a low power analog integrated circuit architecture for the implementation of a general purpose type-2 Fuzzy Inference System. The main objectives of the design are high implementation accuracy and low power consumption, achieved by using bias currents on the scale of nA, a supply voltage of 0.6 V and transistors operating in the sub-threshold region. The system architecture and its key structural components are analyzed, and innovative circuit architectures are proposed to enhance the response quality and reduce the power consumption of each of them. We propose a pure analog integrated type-2 Fuzzy Inference System architecture which consists of modified Bump circuits for constructing Gaussian membership functions, current-mode MIN/MAX operators for implementing the fuzzy reasoning, and OTAs in a voltage follower-aggregation technique for the center-of-gravity defuzzification block. The architecture is easily tunable at the design level with respect to the various system hyperparameters, resulting in a general-purpose fuzzy system that can be modified to suit a wide variety of fuzzy reasoning hardware applications. Finally, the system is fully electronically tunable and programmable even post-fabrication. The training procedure, operational principles and tunability of the proposed architecture are described in detail. The efficiency and accuracy of the proposed architecture are verified in fuzzy classification applications using real-life datasets. A total of six analog fuzzy classifiers were implemented in a CMOS TSMC 90nm process using the Cadence IC Suite for the electrical and physical design utilizing the proposed architecture for three classification applications. Each analog classifier is evaluated on the corresponding dataset both in terms of classification accuracy compared to the equivalent software fuzzy classifier, and in terms of the sensitivity of the analog system.