Neural network based time domain modelling of 0.18 μm MOSFETs

Abstract

In this paper a new time domain based neural network model of a 0.18 μm RF MOSFET will be demonstrated. The model consists of only three intrinsic non‐linear current sources, each being modelled by a neural network derived from large signal time domain voltage/current waveform measurements at the intrinsic device terminals. This approach negates the traditional method of multi‐bias S‐parameter measurement providing a fast and accurate method of modelling the transistor under large signal conditions. Full verification of the model will be provided against an 80 μm/0.18 μm MOSFET operating at 2.4 GHz. © 2002 Wiley Periodicals, Inc. Microwave Opt Technol Lett 35: 203–206, 2002; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/mop.10558

Non‐linear modeling of 0.18‐μM CMOS using neural network

“…Recently, the issue of non-linear modeling of CMOS in the time domain has been taken up [7]. But to the authors' knowledge a fast and accurate neural network approach to develop non-linear model of CMOS in frequency domain has not been treated previously.…”

An efficient neural network approach for nanoscale FinFET modelling and circuit simulation

“…50 nm, a physics-based compact model is very difficult to develop [2][3][4]. In recent years artificial neural network (ANN) has been used for modelling of variety of transistors [5][6][7], as it avoids repeated solving of the complex transcendental equations of a traditional compact physical model, but still offers sufficient accuracy. An ANN model can be based on either a direct [8] or an indirect equivalent circuit approach [5][6][7].…”

“…In recent years artificial neural network (ANN) has been used for modelling of variety of transistors [5][6][7], as it avoids repeated solving of the complex transcendental equations of a traditional compact physical model, but still offers sufficient accuracy. An ANN model can be based on either a direct [8] or an indirect equivalent circuit approach [5][6][7]. As the indirect approach can model both non-linear dc and dynamic (ac) indirect approach our ANNmodel is based on the equivalent circuit [9] shown in Figure 1.…”

“…The complete equivalent circuit model of FinFET shown in Figure 1 can be divided into two parts: the intrinsic part of the device corresponding to the source of the non-linearities, and the extrinsic linear, bias independent parasitic. A similar indirect approach has been used for other devices [5][6][7], but this is the first application for FinFET modelling.…”