|
We have developed a new Multi-domain Genetic Algorithm (MDGA) as a tool for advanced recipe development and applied it to metrology based on X-ray reflectivity (XRR) and spectroscopic ellipsometry (SE). In our MDGA approach, multiple data sets are examined with the output being an optimal set of parameters for robust and rapid measurements. The data sets usually span the expected range of variations likely to be encountered in a process to be monitored (e.g., the data sets correspond to different thickness but with the same density or dispersion).In one application involving XRR measurements, a set of Ti films with thickness of 200 Å was plasma treated for 0 sec, 10 sec, 30 sec, 50 sec, and 100 sec. Although it was expected that the plasma treated Ti film had a higher density than the non-treated Ti film, we found that the plasma treated Ti film had to be modeled as a two-layer film stack: the plasma treated Ti on top of a regular Ti. Without the MDGA, the densities of the top plasma treated Ti and the bottom Ti traded off since they are so close. With the MDGA, the densities of the top and bottom Ti films were regarded as global optimization parameters while the thickness and roughness of each layer were allowed to vary as local parameters. Our results show that the MDGA can clearly separate the plasma treated Ti from the untreated Ti film across the entire wafer set. On the other hand normal non-linear regression methods cannot distinguish the plasma treated Ti from the untreated Ti. In an application using SE measurement of a bottom anti-reflective coating (BARC) material, a linescan of 11 points across a 200mm wafer was measured with the thickness of the BARC film treated as a local parameter while the dispersion was treated as a set of global parameters. With the help of the MDGA, the dispersion modeling of the BARC film captured two main features at ~ 4.77 eV (260nm) and ~ 5.07 eV(235nm), as well as three small peaks at 3.16 eV(392nm), 3.37 eV (368nm) and 3.53 eV (351nm). In this way, the measured dispersion of the BARC film is more representative of the entire wafer than any dispersion developed from a single point measurement.
|
