Increasing the in-situ active doping level is a primary method for reducing source-drain resistance1. The impact of this approach is dependent on the dopant atoms spatial distribution and their local chemical environment. To measure and optimize these parameters is a characterization challenge although atom probe tomography (APT) has the potential to capture the required data. Likewise, extracting the relevant information and its interpretation still requires development. In this presentation, we show how a 3D APT dataset can be analyzed using a statistical algorithm to characterize the spatial (in)homogeneity of all the elements present. Moreover, the influence of the dopant atoms on their local chemical environment is also revealed.For demonstration purposes, a 43 nm thick in-situ highly boron-doped (~4 at.%) epitaxially grown SiGe layer with an intended site fraction of 0.7:0.3 (Si:Ge) was utilized. A custom Python code, developed upon a similar algorithm to that proposed by Stephenson et al. 2, alongside a significance test to evaluate the element spatial inhomogeneity within our dataset was employed. The code utilized a fixed number (500) of nearest neighbor atoms surrounding each dopant element, thereby establishing the local composition pertaining to the dopant element. A two-sample Kolmogorov-Smirnov test3 was conducted to validate the spatial (in)homogeneity. This involved comparing the resulting composition frequency of the experimental dataset with a randomized version of the same dataset featuring shuffled chemical identities. A significance threshold of α ≤ 0.01 was applied to reject the null hypothesis for a random distribution of elements throughout the dataset.The empirical cumulative distribution functions (CDF) for the B-B, along with the B-Si and B-Ge site fractions are presented in Fig. 1. Statistical analysis of the data supports all three elements being spatially inhomogeneous i.e., p value < 0.01. The B-B distribution also shows a notable deviation towards higher concentration relative to the randomized version and indicated ~10% of the B was aggregating. Sub-dividing the B-B composition distribution (Fig. 2a) into three regions, with the division points being where the two composition distributions (measured and randomized data) intersected, enabled further insight about the localized matrix behavior (Fig. 2b, c). The Si site fraction was found to monotonically shift from a lower to higher level with increasing B composition, while the Ge exhibited the opposite trend. This implies that the B incorporation modifies the matrix at a localized scale and leads to an enhanced Si site fraction around the B. This finding is consistent with a recent DFT simulation for B doped SiGe4. The 3D-APT metrology presented here enhances our ability to study the spatial distribution of dopants on an atomic scale and could be used, for example, to refine the epitaxial growth or link it to the performance of devices. J.-L. Everaert et al., in 2017 Symposium on VLSI Technology, p. T214–T215, IEEE, Kyoto, Japan (2017).L. T. Stephenson et al., MethodsX, 1, 12–18 (2014).J. N. Miller, J. C. Miller, and R. D. Miller, Statistics and chemometrics for analytical chemistry, Seventh edition., Pearson Education Limited, Harlow, United Kingdom, (2018).G. Rengo et al., J. Phys. Chem. C (submitted) Figure 1
Read full abstract