Lifter Assignment problem in multi-floor FAB
As semiconductor manufacturing processes become complicated and require more production steps, a single-floor FAB is not sufficient to meet the increasing demand. Therefore, a multi-floor FAB was developed in recent years and the efficient operations of inter-line (inter-floor) lifter emerged as one of the important challenges in semiconductor manufacturing. Due to an increase in inter-line transfer, a bottleneck can occur in lifters, leading to production loss. Hence, it is crucial to assign lots to each lifter properly.
Since traffic control is one of the significant challenges in an AMHS, the efficiency of inter-line delivery has emerged as an important issue. The number of lifters is limited and the increase in inter-line delivery leads to a bottleneck if an excessive number of FOUPs are assigned to a single lifter. Therefore, it is essential to assign interline TRs to lifters properly in order not to cause production loss.
Our project aims to design an algorithm that can be applied to a multi-floor FAB with heterogeneous lifters. In contrast to previous papers, in our approach, a lifter is assigned and scheduled simultaneously using reinforcement learning. However, if the cost function is calculated solely with the total delivery time, utilization of lifters becomes unbalanced and lifters that are nearby bottleneck machines will suffer from excessive loads. Therefore, we proposed a heterogeneous lifter assignment algorithm balancing utilization.
Lifter operation procedure
A lifter is composed of input and output ports including physical and virtual ports, a rack-master, and shelves. Once an inter-line transfer request is generated, a nearby vehicle moves to the machine (side-track-buffer) and delivers a FOUP from a source machine (side-track-buffer) to a lifter. The FOUP is assigned to an available lifter by the manufacturing execution system (MES) according to its policy and waits in an input port until the rack master is available. An ‘available lifter’ means that the number of scheduled lots is less than the maximum capacity of input ports. The FOUP is delivered to a different floor by a rack-master (transporting robot) and waits for a vehicle in an output port to be transported to the destination machine. If all the output ports are occupied, the FOUP is sent to a shelf. If all the lifters are occupied, a vehicle carrying the FOUP moves along the railway not to block other vehicles, which is called a “retrial’. If the assigned lifter is still unavailable after several retrials, the FOUP is reassigned to another lifter, which is called an alternative delivery.
Heterogeneous Lifter Assignment
In multi-floor FABs, while each floor is connected via lifters, heterogeneous lifters exist. Each lifter runs only on certain floors or in a certain direction. If lifter A only runs between the first and the second floor and lifter B runs between the second and fourth floor, a TR, which is scheduled to be transported from the first floor to the fourth floor, can be assigned to both lifter A and B in order. Using two different lifters increases net delivery time. However, if all other lifters are occupied or the waiting time of other lifters is longer, delivery via two different lifters should be considered.
Benchmark Solution approaches
1) Round Robin rule
Each lot is assigned equally to each lifter in a circular order. While the round robin rule can reduce a gap between utilization of lifters, the rule cannot guarantee transfer efficiency as it does not consider waiting time and distance from machine to lifter.
2) Shortest Distance First
A transfer request is assigned to the nearest lifter from the source machine. The rule reduces delivery time from the source machine to the lifter. However, as it does not consider the delivery time from the lifter to the destination machine, the rule cannot minimize the actual delivery time. Moreover, if excessive lots are waiting in the input ports of the assigned lifter, the sum of the delivery time and the waiting time increases.
3) SEAT (Shortest Expected Arrival Time)
Na et al. [1] proposed Shorted Expected Arrival Time (SEAT) rule and compared with four different rules to assign lifters. A transfer request is assigned to a lifter that minimize the expected arrival time derived from the following equation:
4) MDP-based algorithm
Shin et al. [2] designed a two-step algorithm for inter-line transports.: 1) an MDP-based algorithm for inter-line transports and 2) an algorithm using clustering, partitioning, and tournament methods to reduce the dimension of the problem. Using the Markov decision process, they considered future states as well as current states.
[1] Na B, Woo JE, Lee J (2016) Lifter assignment problem for inter-line transfers in semiconductor manufacturing facilities. Int J Adv Manuf Technol 86:1615–26.
[2] Shin, K., Jang, H. & Kim, H. (2022) An MDP-based lifter assignment algorithm for inter-floor transportation in semiconductor fabrication. Int J Adv Manuf Technol 119, 6583–6598