nocliper / ilt

:microscope: Numerical routines to inverse the Laplace Transform for semiconductor Deep Level Transient Spectroscopy.         To cite: Vasilev, A. (2024). Numerical Inverse Laplace Transform for Deep-Level Transient Spectroscopy. https://doi.org/10.5281/zenodo.10462383
6 stars 3 forks source link
ill-posed material-characterization regression tikhonov-regularization

Inverse Laplace Transform for Deep-Level Transient Spectroscopy

Try In Colab

This repository contains code for analysing DLTS and Laplace DLTS data using different regularisation approaches to produce reliable results. Arrhenius plot option available to plot temperature swept data.

Cite with: Vasilev, A. (2024). Numerical Inverse Laplace Transform for Deep-Level Transient Spectroscopy. Zenodo. https://doi.org/10.5281/zenodo.10462383

Classic DLTS

The time-window concept proposed by D. Lang is the reason why classic DLTS is very sensitive to the small concentration of deep traps even with a low SNR (signal-to-noise ratio). But the main drawback of this approach is the inability to deconvolute the signal of two overlapped traps.

Laplace DLTS

So instead of setting the time window and searching for the peaks, regularization is imposed and brings a full emission rate picture for one transient. This gives an increase in trap separation but higher SNR is needed. Higher SNR can be achieved by averaging many transients (SNR $\propto$ $\sqrt{N}$, where $N$ – number of averaged transients).

Used algorithms

Multiple algorithms are used to increase the reliability of results:

(L-curve is used with Contin, L2 and reSpect for regularization parameter optimisation)

Results

This notebook can be used to perform Laplace DLTS to .DLTS files in the data folder. Contin and pyReSpect will work in 99% of cases. L1, L1 + L2 and especially FISTA work well in the other 1% with different sign exponential decays (example of FISTA probing mobile ions in perovskites).


Notebook working cell



Results obtained using reSpect algorithm

Articles