Improved strain field reconstruction in a 45 nm technology by using CBED and recursive dynamical simulation A. Spessot1, R. Balboni2, S. Frabboni3, G.P. Carnevale4, I. Mica 4 and M. Polignano4 1. Micron Technology Belgium, Imec Campus, Kapeldreef 75, Leuven, Belgium 2. CNR-IMM Bologna, via P. Gobetti 101, Bologna, Italy 3. Dipartimento di Fisica, Università di Modena e Reggio Emilia and CNR Istituto di Nanoscienze-S3 Modena, Via G. Campi 213/a, Modena, Italy 4. Micron Technology Italy, R2 Technology Center, via Olivetti 2, Agrate Brianza, Italy Corresponding author:
[email protected] Keywords: CBED, strain, STI
Strain in silicon crystals is still one of the major fields of interest for the current and future integrated circuit technologies in the field of micro- and nano-electronic research. Mechanical stress build up during the device processing can have a detrimental result on the final product, such as performance and reliability degradation. On the other hand, strain engineering can be used to improve and tune the device performances [1]. Convergent Beam Electron Diffraction (CBED) proved to be a useful tool to measure the lattice strain with accuracy in the 10-4 range and spatial resolution at the 10 nm level. Quantitative measurements can be quickly obtained using a quasikinematical approach provided that the strain field is uniform in the analysed sample volume [2]. If this is not the case, as it happens with structures much smaller than the TEM lamella or when strain relaxation effects due the sample thinning are not negligible, the CBED technique can still be used but the experimental data should now be compared with dynamical electron diffraction simulations that are much more demanding in terms of strain modelling and computing time. This kind of strain analysis was performed on silicon STI structures in the orientation [3]. The crystal orientation is a limiting factor when CBED analysis is performed in oriented structures since the actual spatial resolution may be worsened by the projection effect along the TEM lamella. In this work the CBED method for strain analysis was performed using the orientation in the active areas of STI silicon structures of the 45 nm node technology, grown along the orientation; since is tilted about 8 off the direction, it allows a 30% gain in lateral spatial resolution with respect to the orientation (11.3 from the ). The samples were prepared in cross-section and thinned using FIB to about 300 nm thickness. A 5 KeV final polishing to reduce the crystal damage was used, then analysed at room temperature in a FEI Tecnai F20 ST operated at 200 KV. Fig. 1 shows a bright field image of the sample with some examples of CBED patterns taken at the points superimposed. It can be seen that the HOLZ lines in the patterns are split, due to inhomogeneous strain along the electron beam that can mainly be ascribed to strain relaxation at the surfaces after the sample thinning. To our knowledge, for the first time these diffraction patterns with split HOLZ lines were properly simulated. This task was performed by using an already described iterative fitting procedure [4] that minimizes the differences between a series of experimental and 2-D many beam dynamically simulated patterns by comparison of the profiles of the rocking curves considering a given number of HOLZ reflections. In the present case, the strongest 16 HOLZ lines were simulated, together with the (400) contours belonging to the zero order Laue zone (ZOLZ). The initial strain field in the thin lamella was calculated by using a process simulation that takes into account all the process steps seen by Silicon [5]. Fig. 2a shows the horizontal component of the calculated stress field, which represents the starting point for the recursive pattern simulation. A parametric analytical model is used to fit the initial strain field simulation, and a recursive procedure is launched by comparing a simulated split pattern to the experimental one, optimizing the parameters of the analytical model accounting for the strain field until a satisfactory agreement is reached. When the iterative simulation returns a proper description of a given experimental CBED pattern, the next one of the array is considered, repeating the iterative approach.
Fig. 2b and 2d show the experimental pattern recorded at 100 nm from the Si/oxide interface and the simulated fitting image. The matching between the HOLZ profiles can be considered quite good and after elaboration of the strain relaxation at the surfaces the measured strain values at 100 nm were XX = -610-4 and ZZ = 1-210-4, that are in acceptable agreement with those from the process simulator. In Fig. 2c and 2e the pattern and the final simulation of the datum collected at 80 nm are shown. It is now evident that the intensity distribution of the HOLZ lines is not symmetrical and in the simulation there is a contrast inversion for several lines: as such, the fit cannot be considered satisfactory. The reason for this is attributed to the high strain gradient with non negligible shear strain that cannot be managed in an easy way. In fact, in principle the software code could be extended to include shear components, but practically the introduction of additional independent variables is limited by computation time. The simulation of an array of experimental CBED patterns sets the foundation for the reconstruction of the strain field in the real device. It can therefore be concluded that the method can be presently applied to structures of the 45 nm node technology provided that the strain field can be described in term of planar strain with no shear effects. This work was performed in the framework of the ANNA project, funded by the EC [6].
Figure 1 a), b), c) CBED patterns collected at the position shown in the bright field image on the left. The patterns were taken at respectively 130 nm, 100 nm and 80 nm from the upper Si/SiO interface of the STI active area.
Figure 2 a) Simulated XX strain map for the investigated sample. In b) and c) the same patterns, with the same letters as in Figure 1, are compared with the corresponding simulations in d) and e). While at 100 nm the match is satisfactory, this is not the case for the one collected at 80 nm from the upper Si/SiO interface: the discrepancy has been related to shear strain that cannot be taken completely into account.
References [1] P. R. Chidambaram, C. Bowen, S. Chakravarthi, C. Machala, and R. Wise, IEEE TED, 53 (2006), p.944. [2] A. Armigliato, R. Balboni and S. Frabboni, App. Phys. Lett. 86 (2005), p.063508. [3] A. Spessot, S. Frabboni, A. Armigliato and R. Balboni, Nucl. Inst. and Meth. in Phys. Res. B 253 (2006), p.149. [4] A. Spessot, S. Frabboni, R. Balboni and A. Armigliato, J. of Microscopy, 226 (2007), p.140 [5] http://www.synopsys.com/TOOLS/TCAD/PROCESSSIMULATION/Pages/SentaurusProcess.a spx [6] ANNA contract no. 026134-RII3, http://www.i3-anna.org/
