Paper PS-TuP34
Fault Detection in Radio-frequency Plasma Processing using Voltage-current (VI) Probes and Statistical Models

Tuesday, October 31, 2017, 6:30 pm, Room Central Hall

Radio-frequency (RF) voltage-current (VI) probes, mounted between the matching network and plasma reactor, can accurately detect plasma impedance changes at the fundamental and harmonic frequencies. The plasma impedance is very sensitive to any electrical, mechanical and/or geometrical changes that may occur in the reactor. The nonlinear nature of the plasma impedance generates a rich harmonic spectrum, accurate measurement of which can be used to precisely detect faults that affect the electrical, geometrical or mechanical integrity of the plasma process. Examples of such faults include; electrical integrity of the RF connections, misplaced substrates, broken focus rings, leaking valves, plasma confinement issues and parasitic plasma formation.

In this poster, we summarize the results of fault detection studies across several RF processes and across a number of industry sectors. It was found that for certain gross faults, such as plasma not igniting, the fundamental voltage, current, and phase angle are sufficient for detection. However, many fault types are more subtle. For example, a wafer misplaced by less than a millimeter off-centre does not show any significant deviation in the fundamental parameters. Higher order harmonics, on the other hand, can show significant variation. It was found that the exact harmonic properties (voltage, current or phase) that are sensitive to the fault are also process and reactor dependent.

The statistical approach used relies on a preliminary baseline measurement of a faultless process across the full spectrum of parameters detected by the RF VI probe. The analysis runs continuously and a Z-score technique is applied to all parameters. A deviation from the “normal” is considered significant if it exceeds six standard deviations (Six Sigma). Data from live semiconductor wafer production will be presented showing fault detection due to wafer misplacement. Charts of various other faults and their significance in terms of sigma-number will also be presented for other processes. This approach can be used to implement alarms on plasma processing tools when faults are detected, preventing costly product scrappage events.