Kristin M. Abkemeier
November 20, 1996
Measurements of fourth-order correlations of the flicker noise in hydrogenated amorphous silicon (a-Si:H) thin films reveal state-dependent contributions from non-Gaussian noise processes. Such processes appear only as weak features superimposed on a Gaussian background in power spectral plots and octave correlations. Applying the recently optimized ``second spectrum'' technique yields a sensitive measure of the extent of correlations between fluctuators which are found to depend upon their phase correlations only. The effects upon the noise of annealing and aging the sample as well as the time evolution of the second spectrum upon application of an electrical field suggest that sample microstructure and field induced processes may play roles in inducing correlations among fluctuators.
Thin films of hydrogenated amorphous silicon (a-Si:H), a material widely studied and increasingly used in electronic devices, [1] commonly exhibit conductance fluctuations with 1/f noise power spectra, also referred to as flicker noise. [2, 3, 4] Less universally, discrete switching referred to as random telegraph switching noise (RTSN) has also been observed in time traces of conductance fluctuations. [4, 5] Unlike 1/f noise which is completely characterized by the histogram of fluctuation amplitudes and power spectrum, RTSN also exhibits strong correlations in its higher-order statistics. [6] In this sense, RTSN cannot arise from the superposition of many independent Gaussian fluctuators. However, non-Gaussian components in a noise signal need not be so dramatic as in RTSN. Analyses of fluctuations by means of interoctave correlations in the power spectrum [7, 8, 9] and the ``second spectrum'' introduced by Restle et al. [8, 9, 10, 11] have already been performed on a-Si:H, with many samples showing strongly non-Gaussian behavior. [4, 12] The noise in other specimens has been reported to be purely Gaussian under certain conditions. [13, 14] The data published so far pose a conundrum: Why do certain samples of a-Si:H consistently manifest strongly non-Gaussian behavior including RTSN while others show only moderate amounts of apparently uncorrelated noise?
This paper addresses this question by applying a new formulation of the second spectrum analysis to conductance noise in a-Si:H thin films. We compare and contrast results from this approach with insights drawn from more conventional signal analysis methods. These complementary analyses lead us to suggest that non-Gaussian conductance fluctuations are ubiquitous in a-Si:H even during some periods when conventional analysis reveals only Gaussian processes. The differences between periods of RTSN and periods of flicker noise thus are revealed to be differences in the relative importance of phase-correlated processes to overall noise production. While the precise nature of these processes remains elusive, the methods and results described in the following sections should provide new tools for investigating the mechanisms behind spectral wandering in a sensitive time-resolved fashion.
This paper is organized as follows. Section I details the various signal processing methods used to isolate and analyze non-Gaussian noise. The details of the experimental setup for collecting noise data on a-Si:H samples are described in Section II, and the results of applying the signal analysis techniques discussed in Section I to several representative noise measurements are presented in Section III. Finally, the relationship of these results to previous work and possible interpretations of the patterns observed are enumerated in Section IV.