Leveraging LER to minimize linewidth measurement uncertainty in a calibration exercise

Many semiconductor metrologists are aware that line edge roughness (LER), and thus linewidth variation (LWV), can be a significant contributor to measurement uncertainty. More generally, the impact of measurand variation and proper sampling is becoming a major player in nearly every area of semiconductor metrology. This paper describes a simple technique of using the LWV of a feature as a fingerprint to uniquely characterize the measurement target in such a way to make the LER contribution negligible in a linewidth calibration exercise. A single crystal critical dimension reference material (SCCDRM) was the calibration artifact used to calibrate the tip width of a critical dimension atomic force microscope (CD-AFM). These samples were released by the National Institute of Standards and Technology (NIST) to SEMATECH member companies in 2004. The specific SCCDRM used for this work had six calibrated linewidths ranging from 100 nm to 270 nm. Our paper shows in detail the overlay of the CD-AFM linewidth data with that of the data used to calibrate the SCCDRM for each linewidth. With the aid of this linewidth fingerprinting, Mandel regression is used to assess the quality of correlation of the CD-AFM to that of the NIST-derived calibration data. An uncertainty budget is presented as a conclusion of the tip width calibration exercise. A combined expanded uncertainty of less than 2 nm with a k = 3 coverage factor is achieved.