In this article, an artificial neural network (ANN)-based large-signal model (LSM) of AlGaN/GaN high electron mobility transistors (HEMTs) with accurate buffer-related trapping effects characterization and modeling is proposed. A hybrid small-signal parameter-extraction method for AlGaN/GaN HEMTs is used to acquire the parasitic parameters. To simplify the modeling procedure of the drain–source current $I_{\mathrm{ ds}}$ , an ANN-based model associated with the empirical equations taking into account the trapping effects, self-heating effects, and breakdown issue is developed. The low-frequency dispersions related to the buffer-related trapping effects have been well modeled by using a new empirical equation, which has been verified by small-signal S-parameters. Also, a new thermal factor $K_{T}$ and an improved Shockley diode equation are given in the proposed model as well. The developed LSM has been fully verified by a $2\times 100\,\,\mu \text{m}$ AlGaN/GaN HEMT with the pulsed I–V, small-signal S-parameters, power sweep, and load—pull measurements.

ANN-Based Large-Signal Model of AlGaN/GaN HEMTs With Accurate Buffer-Related Trapping Effects Characterization

Guaranteeing charge conservation of empirically extracted Gallium Nitride (GaN) High-Electron-Mobility Transistor (HEMT) models is necessary to avoid simulation issues and artifacts in the prediction. However, dispersive effects, such as thermal and charge-trapping phenomena, may compromise the model extraction flow resulting in poor model accuracy. Although GaN HEMT models should be extracted, in principle, from an isodynamic dataset, this work deals with the systematic identification of an approximate, yet most suitable, charge-conservative empirical model from standard multi-bias S-parameters, i.e., from non-isodynamic data. Results show that the obtained model maintains a reasonable accuracy in predicting both small- and large-signal behavior, while providing the benefits of charge conservation.

/pdf/charge-conservative-gan-hemt-nonlinear-modeling-from-non-58hebi2xv7.pdf

Charge-conservative GaN HEMT nonlinear modeling from non-isodynamic multi-bias S-parameter measurements

The Genetic Algorithm-Extreme Learning Machine (GA-ELM) neural network algorithm is proposed to model the relevant characteristics of GaN pseudomorphic high electron mobility transistor (P-HEMT) large signal. This algorithm solves the over-fitting problem of the Back Propagation (BP) neural network algorithm in the prediction data. It has the characteristics of fast calculation speed, so it can greatly save calculation processing time. It can also randomly generate the connection weights of the input layer, the hidden layer and the threshold of the hidden layer neurons, avoiding errors in parameter selection. In order to verify the superiority of the algorithm, the modeling effects of the BP neural network algorithm model, the Genetic Algorithm-Back Propagation (GA-BP) neural network algorithm model and the GA-ELM neural network algorithm model are compared in this paper. The results show that the proposed GA-ELM neural network algorithm model has the highest accuracy.

Large-Signal Behavior Modeling of GaN P-HEMT Based on GA-ELM Neural Network

The increasing demand of high-performance power amplifiers (PAs) for modern wireless systems has led the PA structures becoming more complex, thereby resulting in an extremely difficult optimization process for PA design. In this paper, the Bayesian optimization (BO) algorithm with a novel acquisition function, namely clustering guided Gaussian process upper confidence bound (CG-GPUCB) method, is proposed for the optimization of a multi-octave PA. To validate the developed optimization strategy, a high-performance multi-octave PA with a 10-W gallium nitride (GaN) transistor has been successfully implemented. The measured performance of the fabricated PA over the frequency band between 0.6 GHz and 2.8 GHz show that the output power is greater than 40 dBm, the power added efficiency (PAE) is over 62%, and the gain is more than 10 dB. Compared with existing BO based method, the proposed methodology is more efficient, since this method can allow achieving better performance for PAs with less optimization time. A comparison between the achieved results and the performance of other state-of-the-art PAs based on different optimization algorithms has highlighted the validity of the proposed design methodology and the obtained improvement in terms of bandwidth.

https://ieeexplore.ieee.org/ielx7/6287639/6514899/09775928.pdf

A Novel Design Methodology for a Multi-Octave GaN-HEMT Power Amplifier Using Clustering Guided Bayesian Optimization

The increasing demand of high-performance power amplifiers (PAs) for modern wireless systems has led the PA structures becoming more complex, thereby resulting in an extremely difficult optimization process for PA design. In this paper, the Bayesian optimization (BO) algorithm with a novel acquisition function, namely clustering guided Gaussian process upper confidence bound (CG-GPUCB) method, is proposed for the optimization of a multi-octave PA. To validate the developed optimization strategy, a high-performance multi-octave PA with a 10-W gallium nitride (GaN) transistor has been successfully implemented. The measured performance of the fabricated PA over the frequency band between 0.6 GHz and 2.8 GHz show that the output power is greater than 40 dBm, the power added efficiency (PAE) is over 62%, and the gain is more than 10 dB. Compared with existing BO based method, the proposed methodology is more efficient, since this method can allow achieving better performance for PAs with less optimization time. A comparison between the achieved results and the performance of other state-of-the-art PAs based on different optimization algorithms has highlighted the validity of the proposed design methodology and the obtained improvement in terms of bandwidth. 

A Novel Design Methodology for a Multioctave GaN-HEMT Power Amplifier Using Clustering Guided Bayesian Optimization

The paper presents a comprehensive charge modelling approach for field-effect transistor (FET) devices. For the first time an artificial neural network (ANN) is combined with the division-by-current approach to FET charge modelling. Using this technique a large-signal charge model is extracted for a 10 W GaN device from MACOM. It is shown through measurements that excellent results may be obtained using just a single gate charge function, integrated analytically from small-signal measurements.

Charge conservative FET modelling using ANNs

In this paper an experiment has been carried out, in MATLAB, where measured S-parameters from a 10W MACOM wafer device have been fit using an equivalent circuit model. The extracted intrinsic element values aim to keep the modelled and measured results in good agreement with one another while ensuring capacitance values do not break the principles of charge conservation. The model has been verified using measured S-parameters taken over a wide bias plane as well as at frequencies ranging up to 10GHz. The results from our experiment shows an accurate intrinsic model with smooth realistic capacitances between the gate-source and gate-drain terminals while ensuring charge conservation.

Ensuring Charge Conservation in GaN HEMT Large Signal Model

Radio frequency (RF) power amplifier (PA) design using Gallium Nitride (GaN) transistor technology requires accurate device models in order to maximise performance and reduce development time. The current state-of-the-art frequency-domain behavioural models focus on linear and quadratic approximations to the polyharmonic distortion (PHD) formalism. However, the linear approximation suffers from poor accuracy under load mismatch conditions, while the quadratic approximation suffers from poor extrapolation beyond the measured range, leading to erroneous predictions of the optimum load impedances for maximum output power and maximum drain efficiency. In this work, a rational Pade-based approximation is proposed as the model core, and it is shown, through experimental validation, that the Pade approximation-based model can provide superior results in a more scalable format. It can mitigate problems found in the existing PHD models when applied to the matching problem. Specifically, the proposed model produces fewer erroneous solutions for the optimum load points, due to the well-behaved nature of Pade approximants. In addition, for the first time, results are reported on using the behavioural model to determine the optimum impedance for maximum transducer gain in a two-port device model. All results show the Pade model has high potential when compared to the established PHD-derived models in RF PA design.

/pdf/pade-approximation-based-behavioral-modeling-for-rf-power-3fgenk8tqp.pdf

Pade-Approximation Based Behavioral Modeling for RF Power Amplifier Design

This paper introduces a simple and accurate approach to extracting an energy and charge conservative model for the displacement current in a field-effect transistor (FET). Through careful fitting of the device transconductance time-delay parameter, a symmetric capacitance matrix is obtained that may be used directly to extract a single energy function in the form of an artificial neural network (ANN). Results show excellent capacitance fits across the full bias plane along with high-fidelity S-parameter fits at multiple bias points.

Energy and Charge Conservation for FET Models

This paper describes a simple approach to dispersion modelling in GaN transistor devices. This technique accounts for the effects of trap-related dispersion on the dynamic nonlinearities i.e. the gate capacitances, in addition to the well-known dispersion of the drain-source current. Using pulsed-bias S-parameters, capacitance surfaces are extracted across the pulsed-IV plane, with the 2D capacitance surface parametrised by the quiescent bias point. Filter networks allow the model to dynamically determine the quiescent bias, ensuring the correct capacitance surface is used according to the slowly-changing state dynamics. This technique provides a model that is capable of producing two distinct sets of S-parameters, depending on whether dc or pulsed biasing is used. The modelling approach is verified up to 40 GHz on a GaN HEMT device, pulsing across the bias plane. Results show good prediction of both the dc and pulsed S-parameter measurements, from a single global model. The importance of extracting elements based on pulsed S-parameter measurements rather than dc S-parameter measurements is made clear through large-signal measurements and simulations.

Ciaran Wilson

Papers

Charge conservative FET modelling using ANNs

Ensuring Charge Conservation in GaN HEMT Large Signal Model

Pade-Approximation Based Behavioral Modeling for RF Power Amplifier Design

Energy and Charge Conservation for FET Models

Intrinsic Capacitance Extraction from Pulsed S-parameters