J.C. Dupuy

Deconvolution of SIMS depth profiles of boron in silicon

We have measured the depth resolution function of the SIMS analysis of boron in silicon for different experimental conditions and fitted this function with an analytical expression initially proposed by Dowsettet al. We use this analytical depth resolution function for the implementation of an iterative deconvolution algorithm, taking into account several properties of the signal, such as positivity and regularity. This algorithm is described precisely. The algorithm is tested on several theoretical structures and then implemented for the deconvolution of real structures of boron-doped silicon layers in silicon. In particular, a sample constituted by six consecutive delta layers and a 75 A thick layer are deconvolved. It is shown that the asymmetry of the profiles is completely removed and that the full width at half-maximum of the deconvolved delta layers can be reduced down to 41 A. It is also shown that a layer whose real thickness is smaller than the measured width of the resolution function can be easily distinguished from a delta layer, and its thickness estimated.

Toward a better reliability in the deconvolution of SIMS depth profiles

In this paper, the problem of the deconvolution of SIMS depth profiles is addressed. In particular, the hypotheses that are necessary for the deconvolution to be possible (in the actual state of the art) in the case of the SIMS signal are reviewed. Then, the principle of regularization, which is a mandatory step in the resolution of an ill-posed problem, is clarified. Two regularization methods used in the field of SIMS analysis are compared: Miller regularization and maximum entropy regularization. In a second part the study of a possible deconvolution, using a depth resolution function (DRF) that is not the DRF that has experimentally convolved the profile, is justified and theoretically addressed. Two cases arise: the DRF used in the deconvolution process is either thinner than the experimental DRF or it is thicker. It is shown that deconvolution using a DRF that is wider than the DRF that actually convolved the profile is possible, and must be taken into consideration. Some examples of simulated deconvolutions with a false DRF are given, and some tools are proposed that are theoretically able to detect a problem when the DRF used in the deconvolution process is wider than the real DRF. In the last section, an example of experimental deconvolution shows that the deconvolution process is able to provide reliable information. In particular, deconvolution of a SIMS analysis conducted at 5.5 keV O2+ and 42.4° incidence (in a Cameca IMS 3/4f) reveals some features of the sample that require at least 1 keV O2+ and 60° primary beam incidence to be detected experimentally.

SIMS depth profile correction for the study of the first step of the diffusion of boron in silicon

Abstract The analysis by secondary ions mass spectrometry (SIMS) is a powerful tool for the study of the diffusion of boron in silicon. It has been extensively used in order to determine diffusion coefficients. Most of the time, the SIMS profiles used to calculate the diffusion coefficients are sufficiently large so that the SIMS analysis do not modify the real concentration distribution in a significant manner. But more and more, the first steps of the diffusion are to be considered in order to obtain very thin diffused structures. In that case, where very thin layers are analyzed by SIMS, it is no more possible to measure directly the real concentration distribution because the SIMS analysis modify it rather significantly. When the real width or the real shape of the analyzed layers is needed, the measured profile has to be corrected in some way. We show that in the case of Gaussian original profiles, the SIMS profiles can be efficiently corrected, and the exact second order moment of the profile determined from the measured profile until the original second order moment is as low as 20 A. This can be done by the study of the properties of the convolution of the profiles by the SIMS analysis response function. In the case of non-Gaussian profiles, the real shape and width of the profiles can be determined by a deconvolution procedure that we have previously described. (B. Gautier, J.C. Dupuy, G. Prudon, J.P. Vallard, C. Dubois, Proceedings of SIMS X, The International Conference on Secondary Ion Mass Spectroscopy and Related Techniques, Wiley, Chichester, 1996, pp. 443; B. Gautier, R. Prost, J.C. Dupuy, G. Prudon, Surface and Interface Analysis 24 (11) (1996) 733). This procedure is applied to the case of the SIMS measurement of δ-doped layers of boron in silicon before and after rapid thermal annealing (RTA).