An advanced analytical neuro–space mapping technique with sensitivity analysis for transistor modeling

Abstract

This paper presents an advanced analytical neuro–space mapping (neuro‐SM) technique for accurate and efficient modeling of transistor devices. This is an improvement over the existing neuro‐SM, which aims to use neural networks to map a given approximate device model towards an accurate model. The proposed neuro‐SM retains the ability of the existing neuro‐SM in modifying the voltage relationship between the given approximate device model and the accurate model. The proposed technique can also map the current relationship between the given model and the accurate model. In this way, the proposed neuro‐SM can produce improved accuracy over the existing neuro‐SM. In addition, analytical formulas of mapping and sensitivities of the direct current, small‐signal S parameter, and large‐signal harmonic of the proposed neuro‐SM model with respect to mapping parameters and coarse‐model parameters are also derived. The sensitivity analysis can be used with a gradient‐based training technique to improve the model training efficiency. The validity and efficiency of the proposed approach are verified through 2 transistor modeling examples and use of the proposed neuro‐SM models in a large‐signal behavior analysis of an amplifier.

A review on the artificial neural network applications for small‐signal modeling of microwave FETs

“…Artificial neural networks (ANNs) are a well‐established and very powerful mathematical tool, finding a variety of applications as a modeling tool in the field of RF and microwaves . One of the most attractive features of ANNs is their ability to learn and generalize from a set of training data, which is suitable to be exploited for building device models from the measured characteristics.…”

Neurospace Mapping Modeling for Packaged Transistors

“…An evolutionary Neuro-SM modeling technique with high computational efficiency was proposed in literature [19] which considered not only the voltage mappings but also the current mappings. Reference [20] used a dynamic neural network as the mapping network, and two mapping networks with analytical equations were added on the existing model in [21]. These existing Neuro-SM methods mainly focus on modeling for the active cells of transistor.…”