Journal of Vacuum Science & Technology B, Vol.19, No.5, 1931-1941, 2001
Run to run control in tungsten chemical vapor deposition using H-2/WF6 at low pressures
Run to run control with an Internal Model Control (IMC) approach has been used for wafer state (thickness) control in the tungsten chemical vapor deposition (CVD) process. The control implementation was preceded by establishing a stable wafer state thickness metrology using in situ mass spectrometry. Direct reactor sampling was achieved from an Ulvac ERA-1000 cluster tool module during the H-2/WF6 W CVD process at 0.5 Torr for temperatures 350-400 degreesC using a 300 amu closed ion source Inficon Transpector system. Signals from HF product generation were used for in-process thickness metrology and compared to ex situ, postprocess thickness measurements obtained by microbalance mass measurements, providing a metrology accuracy of about 7% (limited primarily by the very low conversion efficiency of the process used, similar to2%-3%). A deliberate systematic process drift was introduced as a -5 degreesC temperature change for each successive wafer, which would have led to a major (similar to 50%) thickness decrease over ten wafers in an open loop system. A robust run to run (RtR) control algorithm was used to alter the process time in order to maintain constant HF sensing signal obtained from the mass spectrometer, resulting in thickness control comparable to the metrology accuracy. The efficacy of the control algorithm was also corroborated by additional experiments that utilized direct film weight measurements through the use of the microbalance. A set of simulations in Matlab (R) preceded the control implementation and helped in tuning the controller parameters. These results suggest that in situ chemical sensing, and particularly mass spectrometry, provide the basis for wafer state metrology as needed to achieve RtR control. Furthermore, since the control was consistent with the metrology accuracy, we anticipate significant improvements for processes used in manufacturing, where conversion rates are much higher (40%-50%) arid corresponding signals for metrology will be much larger.