PLEASE REGISTER FIRST !
-
Wafer manufacturing is highly complex and takes a long time
-
To improve predictability and product yield, we have added a lot of sensors and measurements in the manufacturing line
-
The sensors and measurements are noisy, expensive and take a long time to manage
We want to save the time for measurement and get better insights into the predictive value of these sensors for product quality by using deep learning models to predict the measurement results based on the sensing data
Sthitie Bom, Seagate Technology, sthitie.e.bom@seagate.com
- Sthitie Bom heads the global analytics, reporting and controls organizations for the Seagate Wafer factories.
Chao Zhang, Seagate Technology, chao.1.zhang@seagate.com
- Chao Zhang is currently a researcher working on the soft sensing problem in Sthitie’s team with various deep learning models. He is currently pursuing a PhD in Molecular Engineering at the University of Chicago.
Jaswanth Yella, Seagate Technology, jaswanth.k.yella@seagate.com
- Jaswanth Yella is currently a researcher working on the soft sensing problem in Sthitie’s team focusing on deep neural networks. He is currently pursuing a PhD in Computer Science at the University of Cincinnati.
Yu Huang, Seagate Technology, yu.1.huang@seagate.com
- Yu Huang Zhang is currently a researcher working on the soft sensing problem in Sthitie’s team with various deep learning models. He is currently pursuing a Ph.D. Degree in Electrical Engineering at Florida Atlantic University.
The proposed Data Challenge belongs to the business problem/research problem sector.
The data released contain the sensor data coming out of vaccuum tools.
First Place : $2000.00 USD
Second Place: $1,000.00 USD
Third Place: $500.00 USD
- Important dates
Registration: Jul. 12 - Aug. 31, 2021
Submission: Jul. 12 - Nov. 4, 2021
Judgement: Nov. 5 - Nov. 12, 2021
- Submission
Create a folder under submission/ with your email address, and put your code and prediction results there.
- Task and the evaluation metrics
The task is to use the provided data sets to develop a classification model that works the best. The evaluation metric will be the ROC-AUC of your prediction on the testing set (validation set will be available to you, while testing set not)
The dataset contains train/val/test splits, each has a 3-dimensional input with axis (sample, time-step, feature) and a 2-dimensional label with axis (sample, measurement). The testing labels will be hidden. Check the data/ folder for more details
Note that the train/val/test data are from different time periods of the manufacturing factory, and it’s not guaranteed that the distributions are similar among them.
Participants will need to submit their source code with the prediction values for the testing data, and in backend a score will be calculated based on it, and prizes will be awarded based on the scores.