|
In recent years, many process control techniques have been
developedin the scenario of wafer fabrication, one of which is
the MIT's EWMA (exponentially weighted moving average)
controller which compensates for process drifts by adjusting the
process inputs on a run-by-run basis. The goal of this thesis is
to further enhance the MIT's approach by self-tuning the control
parameter dynamically so that process outputscan be maintained
on the targets as close as possible. In the EWMA controller,
process models are assumed as linear polynomials,in which
constant terms are tuned run-by-run based on the EWMA
algorithmwhile slope terms are fixed and determined off-line.
The optimal controlparameter, which determines the effectiveness
of the controller, is obtained by minimizing the equation of
mean square error from target. Based on the statistical
derivation, it is dependent on the estimation of the drift rate
and model slope terms. In this thesis, we assume the slope
terms are correct and propose two modules, one to estimate the
drift rate and the other one to decide when to change the
control parameter, so that the optimal control parameter can be
self-tuned dynamically. In the drift rate estimation module,
both biased and unbiased estimates are derived using statistical
theory and characterized using Monte Carol simulations. Results
show that both methods can estimate the true process drift rates
within 50 runs. In the decision module, the control parameter
is updated only when there is a statistically significant change
in the drift rate estimate. Two decision rules under reset and
no reset conditions are established and their performances are
evaluated using Monte Carol simulations. Results show that the
proposed self-tuning EWMA controller can capture the process
characteristics dynamically and achieve a better performance
than the original EWMA controller.
|