10:15 AM - 10:30 AM
[16a-Z34-6] Beam tuning parameter optimization by Neural Network on the
NISSIN BeyEX medium current ion implanter
Keywords:Ion Implantation, Machine Learning
In the ion implantation process on mass production of semiconductor devices, beam
tuning parameters are important settings that directly affect the productivity. However, since the
parameters are verified by physically complex interactions such as ion source conditions,
installation accuracy, and beamline contamination, the parameters gradually mismatch against
the optimized process, resulting in the need for periodic manual beam setup interventions to
acquire adequate parameters.
We have developed automatic parameter tuning system to verify optimal parameters for
the current equipment status from the past parameter tuning results using neural network system.
In order to study neural network system, we have to consider two objective variables. One is
beam tuning time and on the other hand is beam tuning parameters. In order to adjust two
objective variables we have developed the combination of two neural networks.
First we have studied that tuning time can be accurately predicted from the transition of
the equipment status, we conducted an analysis with an inference model. As a result of original
random forest inference for the neural network, we succeeded in creating a model with an
accuracy rate of more than 90%.The system was created and simulated, and the results showed
that the beam setup time was reduced in certain condition.Results show that the tuning time is
significantly reduced over 500 seconds conditions in the past.Other than that, using neural
network the variation of each parameter is reduced. This result suggests an improvement in the
process quality.
tuning parameters are important settings that directly affect the productivity. However, since the
parameters are verified by physically complex interactions such as ion source conditions,
installation accuracy, and beamline contamination, the parameters gradually mismatch against
the optimized process, resulting in the need for periodic manual beam setup interventions to
acquire adequate parameters.
We have developed automatic parameter tuning system to verify optimal parameters for
the current equipment status from the past parameter tuning results using neural network system.
In order to study neural network system, we have to consider two objective variables. One is
beam tuning time and on the other hand is beam tuning parameters. In order to adjust two
objective variables we have developed the combination of two neural networks.
First we have studied that tuning time can be accurately predicted from the transition of
the equipment status, we conducted an analysis with an inference model. As a result of original
random forest inference for the neural network, we succeeded in creating a model with an
accuracy rate of more than 90%.The system was created and simulated, and the results showed
that the beam setup time was reduced in certain condition.Results show that the tuning time is
significantly reduced over 500 seconds conditions in the past.Other than that, using neural
network the variation of each parameter is reduced. This result suggests an improvement in the
process quality.