Fully Programmable Gaussian Function Generator Using Floating Gate MOS Transistor

Floating gate MOS (FGMOS) based fully programmable Gaussian function generator is presented. The circuit combines the tunable property of FGMOS transistor, exponential characteristics of MOS transistor in weak inversion, and its square law characteristic in strong inversion region to implement the function. Two-quadrant current mode squarer is the core subcircuit of Gaussian function generator that helps to implement full Gaussian function for positive as well as negative input current. FGMOS implementation of the circuit reduces the current mismatching error and increases the tunability of the circuit. The performance of circuit is verified at 1.8？V in TSMC 0.18？μm CMOS, BSIM3, and Level 49 technology by using Cadence Spectre simulator. 1. Introduction Gaussian function is one of the most widely used functions in many domains such as neural network, neural algorithm, and on-chip diffusion profile. Diffusion is one of the important steps in the chip fabrication. The diffusion profile of impurity atoms is dependent on the initial and boundary conditions. When a constant amount of dopant is deposited on the surface, the doping profile is approximated by Gaussian function [1]. Another application of Gaussian function is observed in multidimensional problems like pattern matching and data classifications [2]. These cases are calculated using probability density functions and these functions can be modeled by normal distributions [3]. Madrinas et al. proposed a CMOS analog integrated circuit to implement Gaussian function [4]. They have successfully designed a five-transistor circuit in which current mirror is in weak inversion region and voltage variable resistors are replaced by two MOS transistors. But the conventional circuit has limitations of mismatching of MOS transistors and it can implement only half of the Gaussian function. The circuit proposed in [5] overcomes the limitations of the circuit proposed in [4] by using FGMOS transistors. Recently the work published in [6] implements the Gaussian functions using fourth-order approximation. The accuracy and complexity of this circuit depends upon order of approximation of the Gaussian function. So, there is always a tradeoff between circuit complexity and its accuracy. This paper presents very simple FGMOS based fully programmable Gaussian function generator that uses a single two-quadrant current mode squarer/divider to generate fully programmable Gaussian function. FGMOS has many attractive features for example it reduces the complexity of circuits and can simplify the signal processing chain of a