by Katherine Derbyshire, Contributing Editor, Solid State Technology
July 31, 2008 – For years, fabs have depended on optical inspection to find all kinds of defects: particles, scratches, stringers of incompletely removed metal. Optical inspection is a robust, reliable, well-understood technique. Yet shrinking feature sizes and more demanding surface cleanliness requirements are demonstrating the limits of optical techniques. Contamination in a cleaning bath might leave a haze of copper ions, less than a monolayer thick. Evaporating immersion lithography fluids can leave watermarks, little islands of resist compounds. Advanced gate stacks measure desired thickness in atomic layers and are sensitive to any disruption of the carefully controlled interfaces. Yet such thin films leave little or no optical signature.
Any contamination on the surface does, however, change the work function of the material. For several years now, Qcept Technologies’ ChemetriQ inspection system has measured changes in work function in order to identify defects such as cleaning residue. As CEO Erik Smith explained in an interview at SEMICON West, the technique works by scanning a metal probe across the wafer. When the work function changes, so does the charge coupling between the probe and the wafer surface. More precisely:
where Cair is the capacitance of the air gap, v is the relative velocity, and VSPD is the surface potential difference. Changes in surface composition affect VSPD.
Let’s look at an example. This spring at the SPCC’08 conference, Spansion reported that their NOR flash memory fab had been suffering yield loss due to “bullet hole” defects appearing after the trench etch (see Fig. 2). Though such defects are usually due to arcing, additional defects appeared during wet strip, not trench etch. These defects could have been missed in the pre-wet strip inspection, but the defect morphology was also unique.
Figure 2: Typical etcher defect (left) versus post-wet strip defect (right). Note the lack of melting around the hole in the right-hand image. (Source: Qcept)
ChemetriQ wafer maps taken at various points in the strip process showed significant surface charging associated with the DI rinse step. A correlation with spin speed indicated that friction between the high resistivity deionized water and the wafer caused surface charging. Adding CO2 to the rinse water reduced charging and eliminated the abnormal yield loss.
In another study with Semitool, the technique identified a via outgassing problem. Post-clean residues trapped inside vias outgassed to the wafer surface. Use of a sweep, rather than stationary, rinse nozzle fixed the problem, probably by improving the efficiency of residue removal.
Since its introduction in 2005, the Qcept technique has shown that it can identify contaminants in the 109 atoms/cm2 range. Measurements do not give the work function of the contaminant, so quantitative techniques like TXRF may still be needed for troubleshooting. Still, the ability to identify surface contaminants at levels below the visual threshold gives fab owners a powerful new monitoring capability. — K.D.