On the Splitting of High Order Laue Zone Lines in CBED Analysis of Stress in Silicon

Abstract

Convergent beam electron diffraction ͑CBED͒ on cross-sectional transmission electron microscopy ͑TEM͒ specimens can map stress in the Si substrate of microelectronics devices with high spatial resolution. However, at shallow depths below the interface the diffraction lines within a CBED pattern are split, rendering pattern interpretation impossible in the classical way. This effect is believed to derive from a bending of the diffracting planes through the sample thickness. In this work, a variety of specimens, ranging from polycrystalline to epitaxial and amorphous materials, have been systematically analyzed. A similar qualitative distribution of the splitting effects is present in each case, but quantitative differences are observed which can be related to the stress induced by the different materials. Capping layers can partially avoid the splitting but seem to modify the strain distribution. The surface relaxation in the thin TEM specimen foil is modelled by finite elements simulations as well as with an epitaxial-layer model. The results predict well the experimental magnitude of the splitting but fail to simulate the intensity of the secondary lines. Possible explanations for such discrepancies are discussed and critically assessed. As the dimension of microelectronics devices continues to scale down well below the hundred nanometer range, the problem of mea-suring mechanical strain and stress on the nanometer scale is receiv-ing growing attention due to its expected effect on the device performance. 1,2 Convergent beam electron diffraction ͑CBED͒ in a transmission electron microscope ͑TEM͒ is a very sensitive method for measuring strain distributions with a strain sensitivity on the order of 0.01% and a lateral resolution on the few nanometer scale. 3,4 CBED measurements are based on the strain-induced shift of high-order Laue zone ͑HOLZ͒ lines, which are present in the central disk of a CBED pattern. The position of these lines is very sensitive to small variations in lattice parameters and therefore to the presence of local strain; a more detailed treatment can be found in Ref. 4. However, when CBED analysis is performed close to a silicon/ stressing layer interface, a splitting of the HOLZ lines inside the pattern in two stronger lines and a set of weaker fringes in between is often found. 5-9 Splitting of CBED patterns has been reported for polycrystalline silicides 5-8 and epitaxial SiGe layers. 9 Other authors showed that splitting can be avoided, e.g., in LOCOS and MOSFET structures 10,11 and under the gate in structures with SiGe grown in the source/drain regions. 12,13 The splitting effect, which hinders any stress measurements by the conventional method based on the analysis of the HOLZ lines shifts, is explained as deriving from a bending of the diffracting planes along the sample thickness. Two possible explanations have been proposed, involving either surface relaxation on the free surfaces of the thinned TEM slice 6,9 or the presence of grain boundaries within polycrystalline layers. 7 The conventional analysis of unsplit CBED patterns is based on the widely accepted assumption that stress-relaxation effects in the thinned TEM specimens are negligible and therefore plane strain approximation holds. 4,14 However, there are cases in which strain relaxation in the thin TEM foils cannot be neglected, as reported for two-beam imaging, high-resolution imaging, and selected-area dif-fraction in spinodally decomposed InGaAsP and SiGe/Si strained superlattices, 15 large-angle CBED ͑LACBED͒ of InGaAs on GaAs, 16 and for amorphous surface films on silicon analyzed by LACBED and high-resolution imaging.