Corinne BERGES poster 050115

Implementation of logistic regression and partitioning techniques
with JMP PRO, for a wafer issue in automotive semiconductor industry
Corinne Bergès, Yves Chandon, Pierre Soufflet
Objective
Context / Issue
• Context:
–electronic die manufacturing for automotive
industry
–specific context: fully automated wafer handling at
test station
Request from the customer for a Design Of
Experiments (DOE)
Objectives: a statistical analysis was started:
–to highlight the key parameters involved in the
issue from the available data
–to check the corrective actions firstly implemented
–to find a possible alternative to a costly DOE
• Issue: at extraction of the dies from the wafers,
backside metal peeling issue on a significant
quantity of wafers
First corrective actions
Available data
• First corrective actions: removing of the pincette
time, and choice for a pincette with less contact
• Available data: all the process and wafer
handling parameters already implemented for
zero defect strategies
Difficulties
• Data not always relevant for this defect
• Due to measurement difficulties, some potential key parameters not monitored
Implementation of logistic regression and partitioning techniques
with JMP PRO, for a wafer issue in automotive semiconductor industry
Corinne Bergès, Yves Chandon, Pierre Soufflet
Method
• Preliminary study: data cleaning and
correlation of all the measurement values
• Logistic regression to see the effect of the
parameters individually: Is there one
parameter for which a specific adjustment will
solve the peeling issue ?
JMP PRO: Analyze  Fit Y by X
• Partitioning techniques: decision tree based
on a G² likelihood ratio Chi-Square test: Is
there a combination of parameters to solve
the peeling issue ?
JMP PRO: Analyze  Modeling  Partition
RT: Room Temperature
HT: Hot Temperature
Logistic regression and partition factors:
• Pincette type
• Pincette time at Room and Hot Temperature
• Probing time at Room and Hot Temperature
Response:
With or Without peeling
Implementation of logistic regression and partitioning techniques
with JMP PRO, for a wafer issue in automotive semiconductor industry
Corinne Bergès, Yves Chandon, Pierre Soufflet
Logistic regression results
Partition results
Before or after corrective actions: Probing Time
at Hot Temperature has the largest impact on
peeling
Logistic regression
Partition: main parameters to be addressed to decrease peeling rate:
• before corrective actions: Probing Time HT and Pincette Time RT/HT
• after corrective actions (Pincette Time removal and Pincette change):
existence of a safe path for no-peeling incidents with specific actions on
Probe Time HT (corrective actions validated)
Partition before corrective actions
Partition after
corrective actions
Implementation of logistic regression and partitioning techniques
with JMP PRO, for a wafer issue in automotive semiconductor industry
Corinne Bergès, Yves Chandon, Pierre Soufflet
Conclusions
References
• Efficiency of the implemented corrective actions
measured
• Chapter #10 (‘Logistic Regression with Nominal or
Ordinal response’), in ‘Fitting Linear Models’ jmp book
• DOE avoided: validity of corrective actions confirmed
later on production data
• Chapter #3 (‘Partition Models’), in ‘Specialized Models’
jmp book
• Finally, discrimination of some parameters involved in
the issue, and help for their new finer adjustment to
definitively solve the problem