Monitoring Wafer-Pad Interfacial Conditions Via Platen Motor Current in Copper, Cobalt, and Shallow Trench Isolation Chemical Mechanical Planarization

Abstract

The experimental analyses in this thesis are presented in two parts (see Chapters 5 and 6). The first study seeks to uncover the relationships that exist between the coefficient of friction (COF) and platen motor current (PMC), as well as shear force (SF) and PMC at highly non-steady-state conditions during chemical mechanical planarization (CMP). The second work expands on the conclusions and realizations made from the first study and focuses on whether in-situ PMC monitoring is reliable at steady state for inferring process conditions or abnormalities. The basis for both studies consists of 12 experimental data sets gathered from Stribeck+ curves (10 for copper, one for cobalt buff, and one for shallow trench isolation). As presented in greater detail in Chapter 5, it is demonstrated that real-time PMC data are strongly correlated with SF data at transient conditions. This becomes evident from high average values of correlation coefficient and coefficient of determination for all 12 cases (0.913 and 0.835, respectively). The average correlation coefficient and coefficient of determination for COF and PMC in Cases A through D were 0.949 and 0.900, respectively. However, for the remaining 8 cases, correlation coefficients ranged from zero to 0.425, with coefficients of determination ranging from zero to 0.180. This was explained based on the predominant contact mechanism which varied between cases. In high-volume manufacturing (HVM) it is desired to run CMP processes at steady-state conditions to reduce variability between polished wafers. Therefore, the follow-up study presented in Chapter 6 investigated whether PMC was a reliable indicator of fundamental tribological phenomena at steady state. From the 12 cases that formed the basis of this thesis, 60 distinct steady-state steps were isolated from their associated Stribeck+ curves and examined. Data averaging, in conjunction with a new trend matching algorithm, signified that PMC was a reasonably good indicator of both SF and COF over large time spans. This was apparent from the average percent match value of 68.5% (ranging between 62% to 86%) for PMC and SF trends. Regarding PMC and COF, the average value was 68.8% (ranging between 62% and 86%). However, at small timescales (approximately for one second) correlations between SF and PMC and COF and PMC were found to be poor or in some cases non-existent. Here, PMC was not sensitive enough to capture the instantaneous stick-slip occurrences and other important tribological phenomena present in CMP.