Using actual equipment data for process prediction in high-mix production

The process result prediction by means of equipment indicators is becoming essential technique for more precise process control. The difficulty in high mix production fab is to create the prediction model with insufficient sample size. It is demonstrated by experiments that the prediction model using a conventional method deteriorates its reliability or precision when sample size is smaller, so that the optimized methodologies in accordance with sample size were developed to obtain reliable prediction model. This method is verified using 45-nm Cu-interconnect via resistance prediction and the prediction result is in good match with the electrical characterization.