Generally speaking, scheduling means the temporal assignment of activities to resources where several constraints have to be fulfilled [13]. A lot of scheduling problems occurred in production environment and several approaches have been proposed. However, the scheduling problem of today’s semiconductor wafer fabrication is more difficult than other production systems because of more complex and dynamic manufacturing environment. Since the traditional scheduling solution is difficult to incorporate with human expertise, it can not be easily utilized. Therefore, we propose a knowledge-base system solution.
In a mass production front-end semiconductor fabrication factory (referred to as wafer Fab hereafter), Lot scheduling plays a very important role in wafer Fab, where the product is processed by a batch size, says ‘Lot’. According to the up-to-date customer-oriented market situation, a good Lot scheduling mechanism should help the wafer Fab to meet numbers of constraints and allocate reasonable resources to gain higher profit.
The manufacturing process flows are controlled by an existing information system, manufacturing execution system (referred to as MES hereafter), in most wafer Fabs. As we know, there are more than 300 process steps for a product and over 20 products in a wafer Fab at the same time. This brings a big challenge for Lot scheduling. As shown in Figure 1.1, every step is served by a machine group. Maybe more than one machine is setup in one machine group. For instance, one machine is
shutdown for preventive maintenance. It is possible that Lots are piling up for waiting process. Furthermore, numbers of re-enter flows are in a product flow and different products are using the same machine group in some steps and the recipes are maybe different. Many concerns, like product delivery date, machine utilization, Fab throughput, line balance, shared resource, manpower arrangement, process causal restrictions, etc., make the wafer Fab Lot scheduling problem be more and more complicated.
Figure 1-1 Flow Control and Lot Scheduling Problem
The overall complexity of wafer Fab Lot scheduling comes from the various process flows, dynamic and uncertain manufacturing environment and changeful market requirements.
There are a lot of performance criterions in wafer Fab manufacturing, including:
On-time delivery
High machine utilization
High throughput
Low product cycle time
Low material and gas usage
Well manpower arrangement
It is difficult to consider all the criterions in the same time. Especially, criterions are conflicted in some situations. For example, the criteria of product on-time delivery maybe decrease the utilization of machines which is one of the criterions. Thus, different situations adopt different criterions normally.
In order to increase average performance, in most wafer Fabs, there would be a specific human expert, called Production Controller who is responsible for the Lot scheduling job in the manufacturing management department, utilizes on her/his knowledge, including sales order requirement, inventory management, manufacturing status, flow control and process machine usage, and experience to make an adaptive Lot scheduling and generates a suite Lot scheduling logics to prioritize Lots. Of course, it is hard to train a comprehensive production controller, who must have enough industry experience and real Fab practice through a couple of years.
In order to avoid any product tardiness, resource waste and machine utilization lost, Lot scheduling is a daily operational planning in semiconductor wafer Fab.
According to the short-term production requirement, production controller aggregates overall relative factors to do the Lot scheduling. The factors are:
1) Delivery due date of processing products: It is a commitment for customers.
2) Bottleneck machines: A machine will be identified as a bottleneck machine, if the queuing Lots of the machine are over its 2 days throughput.
3) Current manufacturing policy: It is a particular manufacturing policy for current business strategy. For example, a cost down policy is to reduce the times of recipe change. The policy will increase machine/manpower utilization and decrease material usage.
4) Engineering Lot control: It is an interrupt to process engineering Lot for some experiments. For a new process development, sometimes engineering department will borrow some machines for experiments and may affect the normal production.
5) Machine preventive maintenance arrangement: A periodic maintenance of machine is needed in order to guarantee the process quality of the machine
6) Process restriction: For example, the reticles which are used in photo-lithograph stage have a high cost and are unique at each layer. The process step must use its corresponding reticle.
7) Inventory handling: The outsourcing manufacture exists in some products process flows.
8) Uncertainty events: For example, unexpected machine down occurs. It means some process steps are stopped. Lot scheduling must respond this kind of special event.
Production controller makes good use of his/her expertise to decide which constraints or heuristics she/he should take. As shown in Figure 1-2, there are 5 Lots are waiting processes by bottleneck machine X. Production controller just concerns
the commitment of due date and product cycle time constraints. She/He adopts two sorting procedures - the most recent due date first and the shortest queuing time first.
After series calculations, the prioritized result comes out.
Expertise Ex:
Lot_DueDate + FIFO
B D E A C A B C D E
Waiting lots for machine X Waiting lots for machine X
Figure 1-2 The Progress of Lot Prioritizing
For the purpose of solving complex Lot scheduling issue, we propose a LS-KBS (Lot Scheduling Knowledge-Based System) approach to utilize human expertise appropriately. First of all, we define a machine-readable XML-based knowledge representation to represent Lot scheduling logic for sharing knowledge. Secondly, depending on our knowledge representation, we provide a visual knowledge acquisition mechanism to elicit human expertise. More detailed reasons of adopting above approach will be shown in coming chapters.
The scope of this thesis is outlined here. In Chapter 2, some related solutions are classified and introduced. Besides, the motivation of this thesis is also described. In Chapter 3, the system architecture is explained. Chapter 4 introduces the knowledge representation and knowledge acquisition mechanism of our Lot scheduling knowledge-based system. Two cases are demonstrated for problem solving in Chapter
5. Evaluation is done in Chapter 6 and conclusion is given in Chapter 7.