Effect of bias variation on total uncertainty of CD measurements

Measurement precision to process tolerance ratio (P/T) is an essential indicator of metrology readiness for a specific technology. A smaller ratio improves the process control achievable with a given metrology. The International Technology Roadmap for Semiconductors is calling for P/T less than or equal to 0.2. The "precision" (P) represents total uncertainty of the measurement. In practice P is estimated as the total variation of measurements made on a reference sample(s) over an extended period of time. This procedure leaves sample-to-sample variation of measurement bias outside of P (measurement bias is the difference between reported average and true value). We report on sample-to-sample CD SEM bias variation as characterized by CD AFM. CD AFM is virtually immune to material, layout and line profile variations and, therefore, is expected to have negligible sample-to-sample. bias variation. We found that sample-to-sample CD SEM measurement bias variation (full range up to 15 nm) is often comparable to or even exceeds CD SEM reproducibility (full range of similar to 3 nm). Therefore, the current methodology of the "precision" measurement is leaving a significant component of the total measurement uncertainty unaccounted. The sample-to-sample bias variation (GB) measured on a set of samples representative to the specific technology needs to be corrected or added to the tool reproducibility (OR) in order to estimate total uncertainty of measurement (sigma(p)): sigma(p)(2) = sigma(B)(2) + sigma(R)(2). This may noticeably change the "precision" of CD SEM (3sigma(p)) and move P/T for the current 130 nm and 100 nm technologies well over the limit of 0.2. Should the industry keep CD SEM as a major in-line CD across chip metrology the sample-to-sample bias variation has to be significantly improved. Otherwise, chip manufacturers will likely fail to deliver required across chip gate CD uniformity.