1 CMOS Nonlinear Signal Processing Circuits

In VLSI circuit design, nonlinear signals processing circuits such as minimum (MIN), maximum (MAX), median (MED), winner-take-all (WTA), loser-take-all (LTA), k-WTA, and arbitrary rank-order extraction are useful functions (Lippmann, 1987; Lazzaro et al., 1989). In general, median filter is used to filtering impulse noise so as to suppress the impulsive distortions. The MAX and MIN circuits are important elements in fuzzy logic design. With regard to WTA application, it is the major function in pattern classification and artificial neural networks. Thus, design of these nonlinear signal-processing circuits to integrate smoothly within SoC (System-on-a-chip) applications becomes an important research. Recently, complementary metal-oxide-semiconductor (CMOS) technology is widely used to fabricate various chips. In this chapter, the designs of all circuits are realized by using CMOS process. However, since CMOS transistor is continuously scaled down via thinner gate oxides and reduced device size, supply voltage is necessary to reduce in order to improve device reliability. Therefore, a high reliable WTA/LTA circuit, a simple MED circuit, and a low-voltage rank-order extractor are addressed in the chapter. The organization of this chapter is as follows. Section 1 introduces the background of these nonlinear functions, including definitions and applications. Section 2 describes conventional WTA/LTA architectures and presents a high reliable winner-take-all/loser-take-all circuit. Section 3 shows an analog median circuit, with advantage of simple circuit. Section 4 describes a CMOS circuit design for arbitrary rank order extraction. Restrictions and design techniques of low voltage CMOS circuit are also addressed. Section 5 will briefly conclude this chapter. Given a set of external input n variables a1, ..., an, the operation of MAX (or MIN) circuit determines the maximum (or minimum) value. A median filter puts out the median variable among a window of input samples. The function of a WTA network is to select and identify the largest variable from a specified set of variables. A counter part of WTA, LTA identifies the smallest input variable and inhibits remain ones. Instead of choosing only one winner, the k-WTA network selects the largest k numbers among n competing variables ( n k ≤ ), which allows for more flexibility in applications. For arbitrary rank order identification, a rank-order filter (extractor) is designed to select the k-th largest element ak among n variables a1, ..., an. Depending on application requirements, these input variables are either voltage, or current signals. In order to clearly describe these nonlinear functions, taking one example indicates these definitions. Two output responses of a circuit corresponding to a set of input currents Iin1, Source: Advances in Solid State Circuits Technologies, Book edited by: Paul K. Chu, ISBN 978-953-307-086-5, pp. 446, April 2010, INTECH, Croatia, downloaded from SCIYO.COM