The 2010 ENBIS Challenge
(Sponsored by JMP)
Using An LPCVD Simulator for Fun and Profit
This year's ENBIS Challenge is about optimization, more specifically about maximizing profit when you have limited resources.
Statistical experimentation is involved, and although it can be tempting to see undertaking an experiment as a singular activity, in the industrial setting experimentation is almost always conducted iteratively to serve the wider goal of improvement. So knowing when and how to experiment is an important skill, but one that can only be appreciated in a broader, yet specific, context.
Also, and perhaps contrary to many text-book examples, it is not at all hard to find real-world experiments that generate a lot of data, so understanding the pattern of response before and after an experiment is run, and general data management issues, can be non-trivial.
This challenge is designed to allow you to explore these issues, comparing different strategies to (try to) meet a well-defined goal.
Manufacturing working semiconductor devices is non-trivial. Many steps are involved and one of the key aspects is that although it's the final, individual, chips that count, in the early stages these have to be fabricated on 'wafers'. When each wafer is cut up, it forms many chips (ideally all identical).
For operational reasons wafers are often arranged into groups ('Lots' or 'Batches'), and can be processed at a particular manufacturing step as a 'Run' consisting of either part of a Lot, or several Lots.
This scenario is an example of the latter situation: Low Pressure Chemical Vapour Deposition (LPCVD) is conducted in a furnace tube, which holds four Lots (each of twenty-four wafers). The engineering objective is to grow a new layer of Silicon Nitride on the surface of each wafer that has properties such that the final device works as designed.
The furnace tube is heated, and gases flow from one end of the tube to the other, passing over and between the wafers. The process is executed according to a recipe that controls things like the rate of heating, temperature, and gas flows and pressures.
The scenario has been simplified, but is still somewhat realistic. Detailed prior knowledge of how real furnaces work will not give you an unfair advantage in attempting the challenge.
You are given 200 Runs, and you have to make as much profit as possible. You can use any sequence of experimentation and production that you care to, but any wafers used during experiments will be given zero profit. The way you use the 200 Runs makes up a 'Scenario'.
Submissions can be made by anyone, and can be uploaded at https://www.enbis.org on or before 24th December 2010. Upload will require you to create an ENBIS profile if you do not already have one. As part of the upload process you will also need to answer the following questions in a web form:
- Name of software used for design and analysis.
- Self-assessment of software proficiency ("Novice", "Proficient", "Expert").
- Frequency of software usage ("Several times a day", "Several times a week", "Several times a Month").
Submissions should be uploaded as a ZIP file named 'ZQJ_<YourNameHere>.ZIP'.
Your submitted ZIP file should contain:
- The JMP table called 'Scenario' containing all the information from the 200 runs in your scenario. (This table can be found in the 'Scenario/' folder, see ‘Installing the Simulator' below).
- A report of no more than five pages showing key findings, and the interplay between these and your strategy to maximize profit.
Submissions will be assessed according to the profit you managed to make in your scenario, and the report you provide.
Paid travel to and free accommodation and attendance at the 10th ENBIS Annual Conference in Antwerp:
Please mind that you need to be ENBIS member, if you are not yet! Membership is free, so you cannot afford not to be part of it. If you are already ENBIS member please login!
The Zipfile containing all the necessary documents and the simulator can be downloaded here.
Installing the Simulator
You can use any software to design experiments, and analyze the data that result from these, and from running production. But the simulation scenario itself has to be conducted and managed via JMP.
Please proceed as follows:
- If you do not already have JMP installed, download the fully functional trial version (which lasts for 30 days) from http://www.jmp.com.
- Download the file ZQJ_ENBIS_Challenge_2010 and unzip to a location of your choice.
- This should make an "ENBIS 2010" folder on your hard drive with the contents shown below:
- Note that ‘ENBIS_2010_Main.JRN' is a JMP Journal file, and (with JMP installed) double clicking will open this in JMP. This is the only file that you need to work with to conduct and manage your scenario.
- If for some reason, the folders inside the ‘Application' folder above are not created successfully, you can simple create them ‘by hand' with the names shown below:
(The other JMP files ‘ENBIS_2010_XXX' shown above are scripts that are called from the main Journal file and should not be moved or deleted).
6. If you do want to use other software for design and/or analysis, the ‘Transfer' folder shown above is the location that JMP will export to and import from.
7. Your final submission must include the completed ‘Scenario.JMP' file that resides in the ‘Scenario' folder shown above.
A Word About JMP
As mentioned, you are free to use any software to meet the challenge.
But if you are unfamiliar with JMP, and would like to road test it, the following comments may help you.
Three important points about using JMP:
- When you select a menu item (for example, Analyze > Distribution), you are presented with a dialog to launch the platform you have chosen. If you are not sure what the various column roles and options are intended to do, hit the Help key (or use the "?" tool in the toolbar) to get context-sensitive information.
- The specific output that appears in your report is determined by the JMP preferences you currently have set. Theses can be changed by selecting File > Preferences and then clicking on the Platform icon.
- As mentioned, JMP is very functional and nimble. Therefore you can make mistakes and learn quickly. Do not be afraid to click around in the report windows with a left or a right mouse click (the latter will usually bring up a context menu that depends on what you clicked on).