Optimization of diffusion batch size based on real-time information

Abstract

Diffusion is one of the operations that are run on batching basis in semiconductor

fabrication. Due to its long processing time and being non-pre-emptive, operators and

supervisors on the shop floor are frequently facing the dilemma that whether to launch

the next batch promptly with available lots to avoid idle time, or to wait for more lots to

arrive so as to make better utilization of the equipment. Hence, the goal of the research is

to derive a new control strategy to optimize the diffusion batch size based on real-time

information in manufacturing system.

Many similar efforts have been made to deal with the problem previously. Compared."

to them, this research features a couple of improvements. First, it employs integer linear

programming instead of heuristics to solve the problem so that the solution is guaranteed

to be optimal. Second, move rate (throughput) is selected to be the objective function as it

is a more practical KPI than the long-time used cycle time in prior research.

Simulation results prove the new method to be productivity boosting under all traffic

intensities. Compared to benchmark, the improvement is about 150/0 under low

utilization, while 2-50/0 under medium to high.

Recommendations for future study are provided at the end.