Answers are grounded in this paper and cite their sources.
For metrology in the semiconductor industry, the application of energy dispersive x-ray spectroscopy (EDS) in a scanning transmission electron microscope (STEM) is often limited to qualitative characterization of devices. Compositional quantification is limited by count statistics of x-ray photons within the allotted measurement time and by the standardless Cliff-Lorimer method. Quantitative metrology here means compositional mapping, at nanometre scale lateral spatial resolution, of microelectronic devices in Ultra Large-Scale Integration (ULSI) technology by an electron probe under non-channelling conditions. We note that targeting specific features in ULSI devices for preparation of TEM lamella by a focused ion beam necessarily determines the sample thickness at no more than twice the technology fin or gate pitch. Along with measuring the beam current of the electron probe, we can then apply the ξ-factor method [1]. Combined with the use of multiple x-ray photon silicon-drift-detectors (SDD) with large collection solid angles, high brightness electron sources, and automation, we evaluate STEM EDS for quantitative metrology of semiconductor device structures. Thus, the research question poised is whether STEM EDS enables high-throughput quantitative metrology of electron-beam sensitive, ultrathin structures used in advanced integrated circuits. We examined representative beam-sensitive materials used in ULSI devices in the form of thin films on a Si(001) wafer: Ti silicide, Ti, and TiN prepared by industrial physical and chemical vapour deposition, see Fig. 1. The layers were capped by Tetra Ethyl Ortho Silicate (TEOS) deposited in the Dual Focused Ion Beam. We applied a STEM equipped with a high-brightness FEG source (X-FEG) and four SDD with an overall 0.7srad solid angle (Metrios R3). We optimized the exposure time and current density of the electron beam by characterizing onset of beam damage to the various layers as described in Refs. [2,3] The contribution of the ξ-factor approach for quantification of composition was evaluated by comparing to the standardless Cliff-Lorimer approach, as well as to STEM electron energy-loss spectroscopy and to x-ray photoelectron spectroscopy. Similarly, the contribution of the high-brightness electron source and large solid-angle SDD was evaluated by comparing to measurements obtained from a comparable STEM (Tecnai F20) with a thermally-assisted field-emission gun (Schottky FEG) source and a 0.13srad solid angle Si(Li) detector. An example of a comparison between standardless Cliff-Lorimer and ξ-factor compositional quantification for linescan measurements perpendicular to the multilayers is presented in Fig. 2. Large discrepancies are observed in comparison between the Cliff-Lorimer and the ξ-factor method, which is highlighted on Fig. 2, for example for Ti:N and Ti:Si ratios. This example discrepancy is explained by N k-factor accuracy and the separation precision between N K and Ti L peaks. We demonstrate that the combination of high efficiency SDD with high-brightness FEG source in a STEM along with automation, beam current measurements, well-defined ULSI device features, and transition to the ξ-factor method, render the STEM EDS a viable method for quantitative compositional metrology in the microelectronics industry. Cross-sectional high-angle annular dark-field (Z-contrast) STEM image of the test sample with the Si(100) substrate aligned to [011] zone axis. The numbers presented denote the approximate layer thickness in nm. The layers were capped by TEOS deposited in the Dual Focused Ion Beam. Compositional quantification of EDS Linescan measurements, acquired by using 10 μsec dwell time and 4 repetitions, 0.3 nm step size and 0.3 nA beam current. Comparison of the quantification by standardless Cliff-Lorimer (top) and ξ-factor (bottom) methods. Differences between the results of the two methods are highlighted by the arrows in the top panel: Significant differences are seen for Ti and N elements, TiSi ratio and sensitivity to C and O.
No takes yet. Share an insight, caveat, or question.
Fishman et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: