ZMT

Intel manufacture process includes many steps in the way from raw material to a complete product.

The key material in the fabrication of semiconductors is wafers which are grouped in lots.

Every lot is processed by many production and testing machines, in more than 400 steps.

The lots path through the machines is determined by a scheduling algorithm.

In order to simulate the fabrication factory production we use the AutoSched system.

Intel’s current scheduling algorithm is the MPLM which plans ahead a shift of twelve hours.

Forced changes and problems in real time causes by the workers that change the lots path manually, using their own reasoning.

A planed shift consequence is that only 70% of the scheduling process is determined by the algorithm.

Today a lot is ready in more than two months, though can be ready within two weeks.

When a user wants to run a simulation, he has to give the AutoSched a huge excel file.

The output of the simulation’s run is also an excel file. The user has to copy and paste the data into MATLAB in order to analyze it.

To reduce the cycle time of the lot in the Fab we want to reduce the time a lot spends in waiting for metrology machines.

An important issue is to maintain the overall yield of the Fab though the algorithm changes.

To examine the different aspects we will develop algorithm component that will interface with the AutoSched.

We will write GUI application which will make running the simulations much easier for the user.

The application will enable experiments management for multi users, and intuitive data analysis.

Configuration file(STD-S1)

Simulation System

AutoSched

Algorithms implementation

Algorithm implementation:› Implement a scheduling algorithm that sends lots

to metrology based on real-time events.› Same, but with taking yield into consideration.

Simulation running environment:

› User can enter the variables for the algorithms we have implemented through an application.

› Saving and loading simulation variables and results.

› Analyzing simulation results – making graphs, tables, etc…

› Running the simulations from our application instead of AutoSched.

Algorithms implementation:› Speed: should give an answer for each lot

within less than a second.› Portability: if anyone would like, he can write

extensions to our algorithms using an API we will provide.

Simulation running environment:› Speed: saving & loading simulations shouldn’t

take more than 3 seconds.› Usability #1: it should be easier now to run

simulations.› Usability #2: the user will not need to do any

analysis by himself or with a 3rd party program.

UC1: Request for a new AP from the simulation.

When an AutoSched simulation runs, an event happens, that is part of the event that needs a new AP solution. The simulation sends information about its current state to the system and after doing its calculation it returns the result (the new AP solution) to the AutoSched simulation.

UC2: Researcher runs experiments.

When the researcher wants to examine new inspection policies he enters new configuration with the right parameter throw the GUI application and run the experiment. After the system run he can analyze the experiment results.

The algorithm wouldn’t give better cycle time results than the current position.

The yield would be harmed as a result of the changes.