An ultra-low power fully-programmable analog general purpose type-2 fuzzy inference system

Abstract

Neuro-fuzzy systems is a soft computing technique for developing intelligent systems that solve complex real-world problems in a way that closely resembles human reasoning. In this work, a novel ultra-low power pure analog integrated type-2 fuzzy inference system architecture is proposed. The system consists of Gaussian membership function and MIN/MAX operator circuits and a center of gravity defuzzification block. The architecture is easily scalable in the design-level, regarding various system’s hyperparameters, resulting in a general-purpose system that can be modified to suit a wide variety of applications. Moreover, the system is fully programmable and electronically tunable even post-fabrication. Finally, a novel fuzzy classifier is implemented in a 90nm CMOS process using the Cadence IC Suite for the electrical and physical design utilizing the proposed architecture. The classifier is evaluated on the breast cancer Wisconsin (original) dataset and is compared with the equivalent software model and other analog classifiers. Post-layout simulation results confirm the proper operation of the design while demonstrating the optimal combination of ultra-low power and high accuracy performance relative to the other analog classifiers.