Method for fault detection in a plasma process

Details Method for fault detection in a plasma process Method for fault detection in a plasma process

2002-11-15

2004-08-24

Pert, Evan (Department: 2829)

Electricity: measuring and testing

Using ionization effects

Reexamination Certificate

active

06781383

ABSTRACT:

BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method of fault detection in a plasma process chamber powered by an RF source.
2. Prior Art
Plasma processing of materials is used in a large number of industrial applications, which include the manufacture of semiconductor devices, flat panel displays, optical components, magnetic storage devices and many more. These plasma processes include the deposition and etching of dielectrics, conductors and semiconductors on a substrate, for example, a silicon wafer. The plasma process usually involves placing the substrate in a vacuum chamber, introducing process gases and applying radio-frequency (RF) power, typically 0.1 to 200 MHz, to create a plasma. The plasma consists of ions, electrons, radical gas species and neutral gas, all of which permit the desired reaction to proceed. The plasma reaction has many inputs, including RF power, gas flows, chamber pressure, substrate and wall temperatures, chamber wall conditions, electrode spacing and so on.
Control of the complex plasma process is the key to improved manufacturing, i.e. to have accurate and repeatable processing the plasma itself should be repeatable. Unfortunately there are few schemes in existence for direct plasma process monitoring and control. It is more usual to monitor or control gas flow, power output from RF generator, chamber pressure or temperature, etc., i.e. the process inputs. However, since the plasma process depends directly on the plasma parameters, measuring these indirect variables is generally not sufficient.
Very often the control scheme employed in, for example, semiconductor manufacturing relies on Statistical Process Control (SPC), whereby many if not all of the process inputs are recorded and control charts are monitored for out-of-control events. It has been, noted many times in the industry that monitoring all these control screens is both problematic and undependable. Inputs can stray outside control limits, with no apparent effect on the process output (i.e. a false positive) and/or process inputs can remain within control limits but process output can drift out-of-control (due to changes in the process conditions).
Improvements in fault detection can lead to greater manufacturing efficiency by detecting problems in a timely manner and reducing yield loss. When the fault is detected the process is terminated and the problem is then fixed.
There is a need to improve the methods for monitoring and controlling plasma assisted manufacturing. In particular improved techniques for fault detection are needed.
The plasma represents a non-linear complex load in electrical terms. This results in distortion of the fundamental RF driving signal. Using a Fourier Transform, the RF signal can be separated into its Fourier components. It is generally accepted that monitoring the Fourier components of the RF power signal provides a useful way to monitor the plasma process. These components are a more direct measurement of the plasma process since they are more directly related to fundamental plasma parameters.
It is known to use an RF sensor to monitor and control RF plasmas by measuring the Fourier components of voltage, current and phase. The sensor can be used in closed or open loop control, as for example, in etch end-point control or as in-situ monitoring of the plasma process. In either case the plasma can be terminated when one or more of the Fourier components reaches pre-determined limits.
U.S. Pat. No. 5,576,629 describes a method for plasma fault detection using a standard SPC approach to monitoring RF components. This approach has a fundamental limitation; namely, only the variance of the individual RF components is considered. This gives rise to several problems. Firstly, there is no indication of which RF components should be monitored. Without this the user is faced with more control charts to monitor. Secondly, the RF components are very sensitive to all process variations, hence their suitability for process control. However, it is found that events such as tool maintenance and normal chamber conditions drift (such as layer formation during normal lifetime) make SPC control using RF components very difficult. Chamber condition drift means the SPC limits need to be dynamic and must be learned in advance.
For example, consider
FIG. 1
which shows a plot of the normalised fundamental voltage V0 recorded before and after a preventative maintenance (PM) event as successive wafers (or wafer batches) are processed. The V0 parameter shows a significant change after the PM. Control limits are calculated from the set of data using the 3-sigma rule and are shown as parallel dashed lines. The change in the nominal value of V0 after the PM necessitates the broadening of the control limits. This is the limitation of applying the classical SPC technique to the RF components.
FIG. 2
is a table showing an experiment where three process input settings of a semiconductor wafer process are varied in turn to replicate possible faults that may occur in a plasma process chamber. The first two wafers are carried out with all process input settings at their nominal set-point. For subsequent wafers, coil generator power, bias generator power and chamber pressure are each varied by the amount shown in the table.
FIG. 3
shows a plot of V0 during the experiment. As expected, the nominal data-points are in control; however six of the fault conditions are also within the control limits, and therefore not detectable. Tool maintenance often requires chamber hardware rebuilds, which has the effect of changing measured RE components dramatically, meaning old SPC limits cannot be reapplied.
For this reason, multivariate statistical techniques have often been used in an attempt to offset these problems. Multivariate techniques take into account not only the individual variance of the control parameters, but also their covariance. This addresses some of the shortfalls of SPC techniques in that the multivariate statistic can be used to compress the data and thus reduce the number of control charts. Also, by including the covariance, problems with data drift can be countered, since the fault alarm now depends not only on how any individual parameter changes but also how they change together. Using these techniques the number of false positives and the number of missed faults have been shown to be greatly reduced.
U.S. Pat. No. 5,479,340 shows a method of applying multivariate statistics to plasma control using inputs from an RF sensor. The technique takes all RF sensor data and compresses to a single Hotelling T
2
statistic. This single statistic is then used for fault detection. A problem with the method as described is that all data is assigned equal significance. The variance and covariance of all parameters are used in the construction of the statistic. There is no way to determine which parameters are meaningful to collect, resulting in a large amount of redundant data being processed. This introduces unnecessary noise and results in broader process control limits. For example, in the case of the preventative maintenance event shown in
FIG. 1
, the global multivariate approach will learn how the SPC limits typically jump from PM to PM and widen the SPC limits accordingly.
Therefore, there is a need to develop a fault-detection technique that is sensitive to fault conditions, but not to the normal changes that occur in a chamber after a preventive maintenance.
SUMMARY OF THE INVENTION
Accordingly, the present invention provides a method of fault detection in a plasma process chamber powered by an RF source and subject to periodic standard preventive maintenance events, comprising the steps of:
prior to a production run using a predetermined plasma process, determining the changes in magnitude of a plurality of Fourier components of the RF source resulting from known changes in a plurality of process conditions and constructing a single parameter which is a linear combination of a selected subset of said components, said combination bein

Affiliated with

Also associated with

Rating

Method for fault detection in a plasma process does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Method for fault detection in a plasma process, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for fault detection in a plasma process will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3349664