Finding the Stress from the Raman Shifts: A Case Study

Abstract

Mechanical stress cannot be avoided during the processing of semiconductor devices. The growth of isolation oxides, the implantation of dopants, the deposition and growth of different films, and the fabrication and filling of trenches generate local mechanical stresses. The magnitude of these stresses depends on the geometry of the films, on their chemical properties and thermal expansion coefficient, on the deposition temperature, etc. Because stress may generate defects or indirectly affect device performance, considerable effort is spent in the microelectronics industry to find its magnitude and distribution in the device, to determine which processing steps are mainly responsible for its generation, and under what conditions the stress becomes critical. Most of these studies use finite element simulations of the stress generation during processing. Unfortunately, simulations may give a distorted image of the real situation if they are not properly validated. Raman spectroscopy is one of the few techniques that can be used for this validation. However, as explained in Sect 3.3 of Chap. 3, the complete spectroscopic information needed to compute an arbitrary stress is almost never available. A typical specimen to be analyzed is a Si-based device fabricated on a (001)-oriented Si crystal. In the backscattering configuration used for Raman microscopy one observes a single Si Raman peak, which is not sufficient to characterize the stress tensor. Broadening of the peak is sometimes observed, but this broadening is hard to interpret because it can be due to stress-induced splitting of peaks, as well as to large stress variations within the illuminated volume. This means that it is in general impossible to obtain quantitative or even qualitative information on the different strain tensor components from Raman data without assumptions or simulations. Therefore, since Raman data are needed to validate the stress simulations and the stress simulations are needed to interpret the Raman spectra, the goal of the industrial physicist is to achieve self-consistency between simulations and measurements.