A scanning capacitance microscope (SCM) has been implemented by interfacing a commercial contact-mode atomic force microscope with a high-sensitivity capacitance sensor. The SCM has promise as a next-generation dopant-profiling technique because the measurement is inherently two dimensional, has a potential spatial resolution limited by tip diameter of at least 20 nm, and requires no current carrying metal–semiconductor contact. Differential capacitance images have been made with the SCM of a variety of bulk-doped samples and in the vicinity of pn junctions and homojunctions. Also, a computer code has been written that can numerically solve Poisson’s equation for a model SCM geometry by using the method of collocation of Gaussian points. Measured data and model output for similar structures are presented. How data and model output can be combined to achieve an experimental determination of dopant profile is discussed.

No abstract

Scanning capacitance microscopy (SCM) was used to image (1) boron dopant gradients in p-type silicon and (2) identical boron dopant gradients in n-type silicon. The bias voltage dependence of the apparent p–n junction location in the (SCM) images was measured. The theoretical bias voltage dependence of the apparent p–n junction location of the same structures was determined using a two-dimensional, numerical Poisson equation solver. The simulations confirm that, for symmetric step p–n junctions, the apparent junction coincides with the electrical junction when the bias voltage is midway between the voltage that produces the peak SCM response on the p-type side and the voltage that produces the peak response on the n-type side. This rule is only approximately true for asymmetrically doped junctions. We also specify the extent of the region on the junction high and low sides from which valid carrier profiles may be extracted with a simple model.

Scanning capacitance microscope (SCM) images, and the two-dimensional (2D) dopant profiles extracted from them, show poor reproducibility from laboratory to laboratory. Major factors contributing to SCM image variability include: poor sample surface and oxide quality, excess carrier generation from stray light, reduced sensor dynamic range from stray capacitance, and use of nonoptimal SCM operating voltages. This article discusses the sources of SCM image variability, how they affect the measured SCM images, and possible approaches for mitigating their effects. Recommended procedures for extracting quantitative 2D are discussed. Finally, a set of informal research materials is introduced consisting of a complementary metal-oxide-semiconductor transistor pair, an identical pair without metallization, and a pair of transistor-like structures with the conductivity type of the source/drains reversed. These structures are intended for use with the FASTC2D software to help improve laboratory-to-laboratory dopant profile reproducibility.

To help correlate scanning capacitance microscope measurements of silicon with uniformly doped concentrations, model capacitance curves are calculated and stored in a database that depends on the probe-tip radius of curvature, the oxide thickness, and the dopant density. The oxide thicknesses range from 5 to 20 nm, the dopant concentrations range from 10 17 to 10 20 cm Ϫ3 , and the probe-tip radius of curvature is set to 10 nm. The cone-shaped probe is oriented normal to the sample surface, so that the finite-element method in two dimensions may be used to solve Poisson's equation in the semiconductor region and Laplace's equation in the oxide and ambient regions. The equations are solved within the semi-classical quasistatic approximation, where capacitance measurement depends only on the charge due to majority carriers, with inversion and charge trapping effects being ignored. Comparison with one-dimensional-related models differs as much as 200% over the given doping range. For shallow gradient profiles satisfying quasiuniformity conditions, the database is used directly to find the doping profile. Converting a 512ϫ512 point image takes less than 2 min. ͓S0734-211X͑98͒02401-9͔