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

The present paper demonstrates the suitability of artificial neural network (ANN) for modelling of a FinFET in nano‐circuit simulation. The FinFET used in this work is designed using careful engineering of source–drain extension, which simultaneously improves maximum frequency of oscillation ƒ_max because of lower gate to drain capacitance, and intrinsic gain A_V0 = g_m/g_ds, due to lower output conductance g_ds. The framework for the ANN‐based FinFET model is a common source equivalent circuit, where the dependence of intrinsic capacitances, resistances and dc drain current I_d on drain–source V_ds and gate–source V_gs is derived by a simple two‐layered neural network architecture. All extrinsic components of the FinFET model are treated as bias independent. The model was implemented in a circuit simulator and verified by its ability to generate accurate response to excitations not used during training. The model was used to design a low‐noise amplifier. At low power (J_ds∼10 µA/µm) improvement was observed in both third‐order‐intercept IIP₃ (∼10 dBm) and intrinsic gain A_V0 (∼20 dB), compared to a comparable bulk MOSFET with similar effective channel length. This is attributed to higher ratio of first‐order to third‐order derivative of I_d with respect to gate voltage and lower g_ds in FinFET compared to bulk MOSFET. Copyright © 2009 John Wiley & Sons, Ltd.     

A kernel approach to implementation of local linear discriminant analysis for face recognition

The multiple linear model is used successfully to extend the linear model to nonlinear problems. However, the conventional multilinear models fail to learn the global structure of a training data set because the local linear models are independent of each other. Furthermore, the local linear transformations are learned in the original space. Therefore, the performance of multilinear methods is strongly dependent on the results of partition. This paper presents a kernel approach for the implementation of the local linear discriminant analysis for face recognition problems. In the original space, we utilize a set of local linear transformations with interpolation to approximate an optimal global nonlinear transformation. Based on the local linear models in the original space, we derive an explicit kernel mapping to map the training data into a high‐dimensional transformed space. The optimal transformation is learned globally in the transformed space. Experimental results show that the proposed method is more robust to the partition results than the conventional multilinear methods. Compared with the general nonlinear kernels that utilize a black‐box mapping, our proposed method can reduce the negative effects caused by the potential overfitting problem. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

A novel empirical study of the two‐dimensional non‐linear drain‐current behaviour in sub‐micrometre MOS transistors

This paper presents a simple, quasi‐static, non‐linear (saturated mode) NMOS drain‐current model for Volterra‐series analysis. The model is based on a linear transconductance, a linear drain‐source conductance and a purely non‐linear drain‐source current generator. The drain‐current dependency on both drain‐source and gate‐source voltages is included. Model parameters are then extracted from direct numerical differentiation of DC I/V measurements performed on a 160 × 0.25 µm NMOS device. This paper presents the Volterra analysis of this model, including algebraic expressions for intercept points and output spectrum. The model has been verified by comparing measured two‐tone iIP₂ and iIP₃ with the corresponding model predictions over a wide range of bias points. The correspondence between the modelled and measured response is good. Copyright © 2005 John Wiley & Sons, Ltd.     

A cellular non‐linear network for image fusion based on data regularization

In this paper, a synthesis method developed in the last few years is applied to derive a cellular non‐linear network (CNN) able to find an approximate solution to a variational image‐fusion problem. The functional to be minimized is based on regularization theory and takes into account two complementary principles, namely, knowledge source corroboration and belief enhancement/withdrawal, both typical of data‐fusion approaches. The obtained CNN has been tested by simulations (i.e. by numerically integrating the circuit state equations) in some case studies. The quality of the results is good, as turns out from comparisons with some standard methods. Copyright © 2006 John Wiley & Sons, Ltd.     

2D time-domain electromagnetic macroscopic numerical modelling on parallel computer: application to the mm-wave silicon DIMPATT diode

A new time-domain two-dimensional (2D) electromagnetic (EM) physical modelling of non-linear distributed semiconductor devices has been developed. It is based on a numerical procedure which solves in a self consistent manner both Maxwell’s equations and a macroscopic transport model based on the drift-diffusion approximation. The software can be run on a parallel computer. It is based on finite-difference time-domain (FDTD) explicit schemes associated to the domain decomposition method. The millimetre-wave travelling-wave IMPATT diode or Distributed IMPact ionisation and Avalanche and Transit Time (DIMPATT) diode is the non linear test structure chosen to validate the model. RF simulations under amplification and CW oscillation operating modes have been performed. The results presented in this paper consist on the one hand of results that can be qualitatively compared to previously published theoretical works and on the other hand of features especially pointed out thanks to the new electromagnetic physical model capabilities.