Skip to content

THE COURSE / M9 MEASURE AND YIELD / M9.1

Measure, then decide

After this lesson you can:

  • Tell metrology, inspection and electrical test apart, and say what each is for.
  • Say why a fab measures a sample, and how a measurement corrects the step that made it.
  • Judge whether a gauge is precise enough for a window, and tell precision from bias.

Loading the lesson…

Sources

  1. N. G. Orji, M. Badaroglu, B. M. Barnes, C. Beitia, B. D. Bunday, U. Celano, R. J. Kline, M. Neisser, Y. Obeng and A. E. Vladár, "Metrology for the next generation of semiconductor devices," Nature Electronics 1, 532–547 (2018), author manuscript. For some products more than half the manufacturing steps can involve measurement; CD-SEM, scatterometry and image-based or scatterometry overlay, CD-SAXS, TEM (destructive), AFM; hybrid metrology; precision and stability matter more than absolute accuracy for most fab uses, traceability when instruments are compared; methods divergence, such as 2.7 nm between CD-AFM and EUV scatterometry; under 5 nm, a deviation under 10% can affect device performance.
  2. MIT OpenCourseWare, 2.830J/6.780J/ESD.63J Control of Manufacturing Processes, lecture 2, "Semiconductor Process Variation" (Spring 2008), CC BY-NC-SA. Defects against parametric variation; temporal variation, wafer to wafer and run to run; spatial variation at the wafer, chip and feature scales; feedback control of variation.
  3. D. E. Hardt, "Cycle to Cycle Control: The Case for using Feedback and SPC," lecture 20, MIT OpenCourseWare 2.830J/6.780J/ESD.63J (Spring 2008), CC BY-NC-SA. Output feedback: measure the product, and change the process input to compensate for disturbances.
  4. NIST/SEMATECH e-Handbook of Statistical Methods, chapter 6, "Process or Product Monitoring and Control" (read 25 October 2026). SPC compares what is happening today with what happened previously, against control limits from the process's history; 3σ limits, outside which a normal process falls 0.27% of the time; inspecting every product is costly and inefficient.
  5. NIST/SEMATECH e-Handbook of Statistical Methods, chapter 2, "Measurement Process Characterization" (read 25 October 2026). Accuracy is qualitative; bias is the difference between the average of measurements and the true value, found by calibration against references and check standards; gauge studies separate repeatability, reproducibility and stability.
  6. J. Oh, J. Son, C. Yoon et al., "Ultra-wide-field imaging Mueller matrix spectroscopic ellipsometry for semiconductor metrology," Nature Communications 16, 8512 (2025). Spectroscopic reflectometry and ellipsometry give thickness and CD non-destructively but point by point, limiting spatial density and throughput; SEM's field of view is small; an imaging instrument measuring over ten million spectra across 20 × 20 mm.
  7. KLA, Metrology (product page, read 25 October 2026). SpectraFilm F10 and F20: a high-brightness light source and expanded-wavelength broadband spectroscopic ellipsometry; the F10 for gate-all-around logic at 2 nm and below and DRAM, with sub-angstrom precision for superlattice and HKMG layers and a tight threshold voltage distribution; applications include inline process and tool monitoring and process tool matching. Archer 800: imaging-based overlay, feedback of on-product overlay error for scanner corrections.