Process Variation Analysis of Device Performance Using Virtual Fabrication: Methodology Demonstrated on a CMOS 14-nm FinFET Vehicle

• Published in: IEEE transactions on electron devices 2020-12, Vol.67 (12), p.5374-5380
• Main Authors: Vincent, Benjamin, Hathwar, Raghu, Kamon, Mattan, Ervin, Joseph, Schram, Tom, Chiarella, Thomas, Demuynck, Steven, Baudot, Sylvain, Siew, Yong Kong, Kubicek, Stenfan, Litta, Eugenio Dentoni, Chew, SoonAik, Mitard, Jerome
• Format: Article
• Language: English
• Subjects: 14-nm fin field effect transistor (FinFET); Annealing; Calibration; CMOS; Design of experiments; Equipment costs; Fabrication; Field effect transistors; FinFETs; High temperature; Logic gates; Mathematical models; Methodology; Model testing; Parameter sensitivity; Performance evaluation; Process parameters; process variation; Semiconductor device modeling; Semiconductor devices; Semiconductor process modeling; sensitivity analysis; simulation
• ISSN: 0018-9383; 1557-9646
• DOI: 10.1109/TED.2020.3027528

A new methodology is demonstrated to assess the impact of fabrication inherent process variability on 14-nm fin field effect transistor (FinFET) device performance. A model of a FinFET device was built using virtual device fabrication and testing. The model was subsequently calibrated on Design of Experiment corner case data that had been collected on a limited number of processed fab wafers. We then performed 400 virtual experiments comprising seven sources of process variation. Using this virtual fabrication technique, we were able to identify a minimum gate-to-source/drain spacer thickness for a high-temperature post-EPI rapid thermal anneal (RTA) anneal process that avoided device subthreshold slope penalties. The model allowed us to determine the optimal Si recess depth target and process window prior to source/drain epitaxy. We obtained these results by reviewing device performance as a function of statistical process sensitivity and highlighting key process parameters requiring variation control. These experiments would have been impractical to perform in an actual fab, due to the time, cost, and equipment requirements of running 400 fab-based process variation experiments for each process parameter. This methodology can be used to avoid wafer-based testing during early technology development.