Neuro-Fuzzy Integrated System and its VLSI Design for Generating Membership Function A.Q. Ansari Department of Electrical Engg., Jamia Millia Islamia, New Delhi – 110025, India, aqansari@ieee.org Neeraj Kumar Gupta Department of Electrical & Electronics Engg., Krishna Institute of Engg. & Technology, Ghaziabad (U.P.)- 201206, India, neeraj_gupta2000@yahoo.com Abstract: In this paper, a Neuro-Fuzzy integrated system, which is based on fuzzy inference system using on-line learning ability of neural network is presented. By using on-line learning procedure, the proposed neuro-fuzzy integrated system (NFIS) can be used to construct an input- output mapping based on fuzzy if-then rules and the tuning of the parameters of membership function. The membership functions for NFIS have been realized using operational transconductance amplifier (OTA). Attention is given to design the circuits with low power consumption 2.91mW and size less than 0.65 mm 2 within the neuro-fuzzy chip. SPICE simulations showed that they are suitable to real time application. Keywords:- Neuro-Fuzzy integrated system, Operational Transconductance Amplifier, Current Mirror OTA, Multiplier Type OTA. I. INTRODUCTION Computational Intelligence combines neural network, fuzzy systems and evolutional computing [1-8]. Neurofuzzy integrated system utilizes features of both Neural and Fuzzy networks together for better results by which we can easily generalize the unseen data from seen data by forming the fuzzy rules and training. Neural networks are composed of a large number of highly interconnected processing elements (nodes), which are connected through the weights. When looking into the structure and parameter learning of neural networks, many common points to the methods used in adaptive processing can be found. The backpropagation algorithm used to train the neural network is a generalized Widrow’s least mean square (LMS) algorithm and can be contrasted to the LMS algorithm usually used in adaptive system. In this paper, we have presented neuro-fuzzy integrated system and its analog VLSI circuits for fuzzy membership functions. Neuro-Fuzzy integrated system learns system behavior by using system input-output data and so does not use any mathematical modeling. After learning the system’s behavior, neuro-fuzzy integrated system automatically generates fuzzy rules and membership functions and thus solves the key problem of fuzzy logic and reduces the design cycle very significantly. Neuro-Fuzzy integrated system then verifies the solution (generated rules and membership functions). It also optimizes the number of rules and membership functions. Finally, automatic code converter converts the optimized solution (rules and membership functions) into embedded controller’s assembly code. The first fuzzy chip was reported in 1986 at AT&T Bell Laboratories. Since then many different approaches have been suggested [9,10]. Depending on the design techniques employed they are classified into two groups: digital and analog. Generally a digital fuzzy system is either a fuzzy (co-) processor [11,12] or a digital ASIC [13], which contains logic circuits to compute the fuzzy algorithm memories to store fuzzy rules and generators or look-up tables for membership functions of the input and output variables. Compared to its analog counterpart, the digital approach has greater flexibility, easier design automation, and good compatibility with other digital systems. However, most of the digital systems require A/D and D/A converters to communicate with sensors and/or actuators. Furthermore, the digital systems are more complex and need larger chip area, e.g. the synthesis of a 4-bit maximum operation in [14] results in a CMOS unit of nearly 100 transistors. The research on analog fuzzy systems started with the pioneering work of Yamakawa [15], and was followed by many researchers [9,10,16]. With the nonlinear characteristics of the active devices in analog circuit, the fuzzy elements can be implemented in very simple structures. This brings a reduction in the circuit complexity, which implies better speed performance and reduced chip-area consumption. Until now, the main drawback of the analog approaches has been their poor flexibility [17,18]. To overcome the shortcomings encountered in analog circuit, while still keeping their advantages, analog VLSI circuits for fuzzy membership functions using operational transconductance amplifier (OTA) is presented in this paper. An operational transconductance amplifier (OTA) is a viable building block for analog hardware implementation of fuzzy logic [19-22]. An OTA can take advantages of both the voltage-mode and current-mode circuit operations since it is essentially a voltage-controlled current source. In addition, operation speed of an OTA is superior to that of an operational 1378 978-1-4673-0125-1 c 2011 IEEE