The need for absolute accuracy is increasing as semiconductor-manufacturing technologies advance to sub-65nm
nodes, since device sizes are reducing to sub-50nm but offsets ranging from 5nm to 20nm are often encountered. While
TEM is well-recognized as the most accurate CD metrology, direct comparison between the TEM data and in-line CD
data might be misleading sometimes due to different statistical sampling and interferences from sidewall roughness. In
this work we explore the capability of CD-AFM as an accurate in-line CD reference metrology. Being a member of
scanning profiling metrology, CD-AFM has the advantages of avoiding e-beam damage and minimum sample damage
induced CD changes, in addition to the capability of more statistical sampling than typical cross section metrologies.
While AFM has already gained its reputation on the accuracy of depth measurement, not much data was reported on the
accuracy of CD-AFM for CD measurement. Our main focus here is to prove the accuracy of CD-AFM and show its
measuring capability for semiconductor related materials and patterns. In addition to the typical precision check, we
spent an intensive effort on examining the bias performance of this CD metrology, which is defined as the difference
between CD-AFM data and the best-known CD value of the prepared samples. We first examine line edge roughness
(LER) behavior for line patterns of various materials, including polysilicon, photoresist, and a porous low k material.
Based on the LER characteristics of each patterning, a method is proposed to reduce its influence on CD measurement.
Application of our method to a VLSI nanoCD standard is then performed, and agreement of less than 1nm bias is
achieved between the CD-AFM data and the standard's value. With very careful sample preparations and TEM tool
calibration, we also obtained excellent correlation between CD-AFM and TEM for poly-CDs ranging from 70nm to
400nm. CD measurements of poly ADI and low k trenches are also reported, and both show good correlation to in-line