Performance and limitations of decomposition-based parameter extraction procedures for FET small-signal models

Abstract

A recently proposed optimizer-based parameterextraction technique using adaptive decomposition is subjected to a systematic and rigorous evaluation. The technique is shown to be robust and accurate under varying starting conditions. A study of convergence performance based on decomposition theory and test results is presented. Robustness tests are used to show that commonly used statistical descriptions such as mean and standard deviation are inadequate for presenting these types of test data.

Langevin Approach to Understand the Noise of Microwave Transistors

“…The values of the circuit elements are usually determined by computer-aided modelling procedures which may follow either optimization-based or direct extraction criteria [5][6][7]. More recently, approaches based on artificial neural networks have been successfully employed [8].…”

“…More recently, approaches based on artificial neural networks have been successfully employed [8]. We had previously characterized an HEMT transistor series (NE20283A, by NEC) over the [6][7][8][9][10][11][12][13][14][15][16][17][18] GHz frequency range at the low noise bias conditions suggested by the manufacturer (V DS = 2.0 V , I DS = 10 mA). By microwave measurements, we determined both the scattering and the noise parameters of the packaged device, i.e.…”

“…By inserting the different statistics of the noise sources in Eq. (6) we can extract all the moments of the output voltage v o (t) required for a complete knowledge of the noise performance when higher order statistics occurs in the output signal of the devices. As an example we report in Appendix B the expression of the second moment and the general expression of the nth moment of the output voltage as a function of the cross-correlation functions of the noise sources.…”

A new procedure for nonlinear statistical model extraction of GaAs FET-integrated circuits

“…Both the bias-independent parameters and the fitting parameters of the nonlinear functions in the model are extracted with an automatic procedure [17,20] which makes use of simulated annealing (SA) [21][22] and gradient-optimization routines. A sensitivity analysis has been performed for each of the six nonlinear functions as suggested in [19], and the optimization routines have been modified to include parameter partition: at each step, the dimension of error function is decreased and a lower probability of entrapment in local minima is obtained. The overall extraction procedure of the MMIC model is shown in Figure 2 and will be described in detail.…”

A new adaptive multi‐bias S‐parameter measurement algorithm for transistor characterization

“…However, in many instances less bias points are added during each refinement cycle since the different S-parameter centroids might select the same bias point (see step 11). The four S-parameters are treated separately since the intrinsic elements of the device small-signal model have different sensitivities with respect to the different S-parameters [8]. A change in the value of the elements C gs and/or R i will for instance lead to a far-larger change in S 11 than in S 22 :…”