Ann-Based Pad Modeling Technique for Mosfet Devices

Abstract

Abstract-In this paper, an approach for the pad modeling of the test structure for Metal Oxide Semiconductor Field Effect Transistor (MOSFET) up to 40 GHz is presented. The approach is based on a combination of the conventional equivalent circuit model and artificial neural network (ANN). The pad capacitances and series resistors are directly obtained from EM (electromagnetic) simulation of the S parameters with different size of pad and operating frequency. The parasitic elements in the test structure can be modeled by using a sub artificial neural network (SANN). So the pad capacitances and series resistors can be regarded as functions of the dimensions of the pad structure and operating frequencies by using SANN. Good agreement between the ANN-based modeling and EM simulation results has been demonstrated. In order to remove the impact of the parasitic elements, the de-embedding procedure for MOSFET device using ANN-based pad model is also demonstrated.

An Ann-Based Small-Signal Equivalent Circuit Model for Mosfet Device

“…Unlike the conventional training process [8,11,12], the output data of the adopted neural networks is not available, in the proposed modeling procedure, the outputs of the neural networks will be sent to the intrinsic part of equivalent circuit and conduct the S-parameter simulation with the whole equivalent circuit. As for the whole ANNbased equivalent circuit MOSFET model, the inputs are bias conditions and frequencies, the outputs are the S-parameter of equivalent circuit.…”

“…2 [3,12,20]. The electrical elements in equivalent circuit can be divided into two parts: the intrinsic parts (in dotted box) and extrinsic parts, respectively.…”

“…Through a process called training, ANNs have the ability of fitting any nonlinear behavior of passive and active component/circuit form experimental data and generating an ANN model function that can be used to describe the port-characters of component/circuit [8][9][10][11][12]. Compared to other conventional modeling techniques, such as numerical modeling method, which could be computationally expensive, or analytical method, which could be different to obtain for new devices, or empirical modeling solution, whose range and accuracy could be limited [8], the ANNs modeling technique is more efficient.…”

Design and Implementation of a Practical Direction Finding Receiver

“…ANNs have been successfully applied in artificial intelligence, device modeling [7,8], parameter prediction, cognitive science, and other scientific fields. Nevertheless, ANN models suffer from the following drawbacks: multiple local minima, overfitting, and low generalizability [9].…”

Air-gapped microaxial transmission line technology for Si-based microwave integrated circuits

“…ANNs are information processing systems that have been widely applied in the RF/microwave modeling tasks as an unconventional alternative. Through a process called training, ANNs have the ability of fitting any nonlinear behavior of passive and active component/circuit from experimental data and generating an ANN model function thatcan be used to describe the port-characters of that component/circuit[ 172,173] . The multi player perceptron (MLP) is a popularly applied neural network structure.…”