Defect Distribution in Doped Silicon Nanostructures Characterized By Means of Scanning Spreading Resistance Microscopy

Abstract

Scanning Spreading Resistance Microscopy (SSRM) is a well-known technique for electrical characterization of dopant distributions in semiconductors. The resistance between a conductive atomic force microscope (AFM) tip and a large area contact on a semiconductor sample gives direct hint of sample resistivity underneath the tip. The sample surface has to be polished to achieve a smooth surface which is crucial for a low ohmic electrical contact between tip and sample. In theory, the formula of spreading resistance Rspr = ρ/4a with tip radius a connects SSRM raw data to resistivity data which can be transformed to concentration data of corresponding dopants. In experiment, additional series resistances and electrical defects have an impact on the SSRM analysis so that the formula does not hold any more. A SSRM scan on a calibration sample with regions of different well-known dopant concentrations enables an empirical relationship between R and ρ. This relationship is only valid for the specific defect distribution and tip-sample contact of the calibration sample. Both the defect distribution and the tip sample contact, respectively, depend strongly on sample preparation. Reliable calibration of unknown dopant distributions in semiconductor samples is only possible in the case of a perfectly reproducible sample preparation which is however hard to attain. Therefore, an accurate quantification of SSRM data remains difficult. One first step towards SSRM quantification requires an improved understanding and control of electrical defects on the sample surface which affect the SSRM measurement. In this work, we apply SSRM on well characterized n- and p-type dopant profiles in silicon bulk- and nanostructures. The deviation of the experimental SSRM data from simulated spreading resistance profiles provides direct information on type, quantity and origin of electrical defects in the sample. Our analysis of bulk samples reveals two type of preparation induced defects: n-type surface defects and hydrogen traps which deactivate boron atoms. The number of surface defects is proportional to the surface roughness. An improved sample preparation with additional polishing steps reduces the impact of this defect on the SSRM analysis. Hydrogen traps are driven into the sample during water based chemical mechanical polishing steps. An annealing of the sample at 250°C for 20 minutes reactivates most of the boron atoms. A third type of defect was noticed in nanostructured samples. Silicon nano pillars with diameters between 500 and 600 nm are manufactured in a top-down approach out of a silicon wafer with a 600 nm deep boron diffusion profile. Figure 1 shows the SSRM data of the bulk sample (a), the cross-sectional prepared nano pillar (b), and the depth profiles of SSRM analysis (c). The SSRM signal at the top of the pillar is dominated by the boron profile (b). The side walls are covered with gold as second contact for SSRM analysis. Obviously, the SSRM depth profile at the shell region (#2, circles) of the pillar is deeper than at the core (#3, triangles) and in the bulk sample (#1, squares) (c). The dopant distribution itself cannot vary across the pillar as the nanofabrication is done by dry-etching at cryogenic temperatures. We explain the variance in SSRM signal by acceptor defect states at the pillar shell which increase the number of p-type charge carriers in the shell region. COMSOL simulations of the SSRM analysis, including the impact of defects on the resistance profile, is used to quantify the defect distributions. The calculated resistance profiles accurately describe the experimental data (black dotted lines in (c)) and support our assumption on the distribution of electrical defects in silicon nanostructures. Figure 1