MESFET DC model parameter extraction using Quantum Particle Swarm Optimization

Abstract

This paper presents two techniques for DC model parameter extraction for a Gallium Arsenide (GaAs) based MEtal Semiconductor Field Effect Transistor (MESFET) device. The proposed methods uses Particle Swarm Optimization (PSO) and Quantum Particle Swarm Optimization (QPSO) methods for optimizing the difference between measured data and simulated data. Simulated data are obtained by using four different popular DC models. These techniques avoid complex computational steps involved in traditional parameter extraction techniques. The performance comparison in terms of quality of solution and execution time of classical PSO and QPSO to extract the model parameters are presented. The validity of this approach is verified by comparing the simulated and measured results of a fabricated GaAs MESFET device with gate length of 0.7 μm and gate width of 600 μm (4 × 150). Simulation results indicate that both the technique based on PSO and QPSO accurately extracts the model parameters of MESFET.

Introduction

Device simulation is one of the important steps for Integrated Circuit fabrication, verification and characterization. Each semiconductor device have models that satisfies the behavior of the device under different operating conditions. GaAs MESFET is a promising semiconductor device used in many applications in the microwave domain. Many models of GaAs MESFET device are reported in the literature [1], [2], [3], [4], [5]. Each model has a unique set of parameters that describes the underlying physical phenomena of the device. These parameters are obtained by minimizing the difference between measured drain current and modeled drain current at different gate bias voltage. This process is known as parameter extraction. Usually the model parameters are extracted using commercial software like HP IC-CAP [6], Silvaco UTMOST [7], TMA AURORA [8], etc. Parameter extraction is an crucial and difficult step for the circuit and device simulator. Until recent past, the model parameter extraction were carried out by using standard gradient based algorithm or Levenberg Marquardt (LM) algorithm [9]. LM algorithm is sensitive to initial values of parameters and is thus proned to be trapped in local minima. The complexity of the model lead the extraction algorithm to take longer computation time for resulting the solution. The traditional algorithm is not an ideal approach if more parameters need to be extracted from a complex model at a time. The major drawbacks of this approach is its poor convergence rate and non-optimal solutions. Non-optimal solutions are due to trap of algorithmic solutions in local optima. To overcome these drawbacks, genetic algorithm (GA) is being used in parameter extraction of semiconductor device model [10]. The advantages of using GA is that its solutions are independent on initial values of the parameters and it provides an optimal set of solution by avoiding local non-optimal solutions. The major drawbacks of GA based approach are that it involves more algorithmic steps and it provides inconsistent results in different simulation environment.

To overcome this problem, we have proposed two techniques namely PSO and QPSO algorithm for model parameter extraction. The PSO algorithm was first introduced by Kennedy and Eberhart in 1995 [11]. Many variants of PSO algorithm were developed by the authors [12], [13], [14] to improve the quality of solution. In recent past, PSO is being used for solving complex optimization problems [15], [16], [17]. The authors have also applied PSO algorithm for extracting small signal model parameters of MESFET [18]. The popularity of this algorithm is due to its simple form, easy implementation steps and ability to avoid local minima. The basic PSO still suffers the poor convergence rate. In this paper, a new variant of PSO known as delta potential well quantum PSO (DQPSO) based on quantum mechanics is used to extract different DC model parameters of a fabricated MESFET device. QPSO algorithm has proven to have advantages than the classical PSO due its less control parameters [19], [20]. In recent past, DQPSO is used to solve real world complex problems [21], [22]. More details about QPSO algorithm is presented in Section 5. The performance comparison of both classical PSO and QPSO algorithm for model parameter extraction is presented. In this paper, all the simulation are carried out for a single geometry device structure. The methodology proposed in this paper is to optimize the error between measured and simulated data using PSO and DQPSO algorithms. The simulated data were obtained using different popular device models [2], [3], [4], [5].

The rest of paper is organized as follows: Section 2 provides a brief description of the model parameter extraction strategies. Section 3 presents the problem formulation for parameter extraction and a brief description about classical PSO algorithm is provided in Section 4. Section 5 presents the description about Quantum Particle Swarm Optimization followed by simulation and results in Section 6. Conclusions are drawn in Section 7.