Great advances have been made in the accuracy and scope of semiconductor fabrication process simulators, yet application of numerical process models to design of complex modem manufacturing processes is limited by difficulty-of-use[ 11. Specifically, differences between simulator input and manufacturing specifications, the requirement of one-dimensional simulation to manage many simulations to analyze a device, the need for calibration of process models, and the problem of interpretation of the voluminous data generated by numerical simulation programs combine to require significant expertise beyond process knowledge to effectively use simulation tools for process design. This paper reports automation of each of these tasks through application of the Process Design Aid[21 (PDA) to optimization of a 0.8pm BiCMOS process.A run-sheet describing a BiCMOS process calls for hundreds of processing operations, more than a dozen masking operations, and several hundred process parameters. Evaluation of this type of process using SUPREM I11 for one-dimensional process simulation requires that a minimum of six regions of the wafer be simulated, each by a sequence of several score simulation commands which reflect the process sequence, process parameters, and mask information. The object-oriented process representa-. tion of this run-sheet and the simulation knowledge-base in the PDA permit automatic generation of individual simulation commands based on the process recipe. Fig. 1 illustrates how the PDA automatically calculates concentration profiles for multiple regions from a single process flow based on a onedimensional representation of masking.Advancing technology pushes simulation models to, and sometimes past, their limits. Test wafers were fabricated with base and emitter implants. Calibration of the diffusion module was based on SIMS profiles measured before and after the final anneal. The poor correspondence between the default simulation results and the SIMS measurements in figs. 2 and 3 indicated the need to calibrate SUPREM's diffusion model. The utility of numerical optimization of simulation results has been reported[3]; here, for ihe first time, general multivariate optimization algorithms were used to calibrate SUPREM 111. By starting the simulation from as-implanted SIMS measurements and optimizing diffusion model parameters to fit the SIMS measurements following thermal processing, the diffusion model was independently calibrated. The calibrated simulation results agree well with measured data for both implant conditions. Calibration coefficients are incorporated into the process library for use in subsequent simulations of the same process and new recipes.With the simulator thus calibrated, the base implant could confidently be optimized based on simulation results to yield a desired structure. Base implant dose and energy were optimized for three different base widths while holding the integrated base dopant constant, illustrating the use of optimization to invert SUPREM: process parameters are computed b...

No abstract

No abstract