Agent Based Modeling of Semiconductor Yield

Summary:

Yield is of paramount importance to the multibillion dollar semiconductor industry where individual devices are fabricated as die on silicon wafers. The production of the final solid state device may take months with hundreds of process steps and tests.

Manufacturing yield loss can usually be classified or assigned to two distinct categories, often termed Systematic and Random Yield Loss. Systematic Yield Loss is usually caused by a human or machine source which continually, or at a definite frequency, causes a product to be unacceptable. Random Yield Loss results from a “more natural” cause such as dust particles, temperature fluctuations, or normal sensitivity of a process to random fluctuations in process machinery. The response to and elimination of yield loss differs by cause and can be quite difficult to elucidate in a large fabrication center with tens of thousands of die per wafer and hundreds of wafers processed per month.

The objective of this study is to develop rapid and simple methods to identify mechanisms of yield losses by studying the final yield histogram of individual die, modeled as agents in conjunction with analysis of actual yield data.