Dr. Peter Lendermann
Chief Business Development Officer D-SIMLAB Technologies
Peter Lendermann is a Co-Founder and the Chief Business Development Officer of D-SIMLAB Technologies, a Singapore-headquartered company providing simulation-based decision support solutions to Semiconductor Manufacturing companies. Prior to this he worked at the Singapore Institute of Manufacturing Technology where he led related R&D activities until spinning them off into D-SIMLAB. Peter has been engaged in the field of production logistics, supply chain management and related decision support technologies and solutions since the early 1990’s. He holds a PhD in Physics from Humboldt University in Berlin (Germany) and an MBA in International Economics and Management from SDA Bocconi in Milan (Italy).
Presentation Title
Scheduling Of Wafer Fab Operations – How to Make It Smart While Juggling Multiple Objectives
Presentation Abstract
In the competitive environment of today’s semiconductor manufacturing, especially in wafer fabrication facilities, it is of high importance to make use of production capacities as effectively as possible while keeping cycle times low. The question of how to schedule lots to production tools in a sequence such that a defined set of Key Performance Indicators (KPIs) is likely to be met in the best possible manner is therefore of utmost importance. Because of the complexity of a wafer fab, even for a single Equipment Group, the optimisation problem to be solved for the generation of a production schedule is a challenging task.
In this setting, the simulation-based approach of the D-SIMCON Scheduler enables generation of schedules for a wafer fabrication (or similar kind of) facility through scheduling lots to a set of several subsequent Equipment Groups (called a Scheduling Group), also taking into consideration actual and forecasted WIP upstream and downstream. To make the schedule ‘smart’ and yet acceptable to production managers and (in less automated fabs) operators, specific domain-specific attributes and constraints have to be considered for schedule generation. The optimisation carried out is a combination of several weighted local objectives such as minimise cycle time, maximise the number of wafer moves, or maximise batching efficiency. Straightforward scheduling decision control is achieved through consideration of special characteristics for operations or even environmental conditions. Certain scheduling parameters can also be adapted automatically based on the WIP situation in the fab.
The key advantage of the D-SIMCON optimisation engine is its ability to guide the search for an optimal schedule using a rigorously tested Genetic Algorithm with domain-specific heuristics, supported by a customised Greedy Search. The underlying paradigm is the focus on creating maximum possible value with as few iterations as possible and leveraging parallel computing infrastructure to achieve this within minimum time.
The solution also incorporates a feature to test revised scheduling objectives (i.e. revised weightages for each of the objective function elements) with snapshots taken at a time when an earlier schedule was generated with a previous, already implemented set of weightages. This is of particular relevance when the fab situation changes, for example in a transition from high to low fab load or vice versa.
The reference deployment for the D-SIMCON Scheduler is in a 200mm wafer fab in Japan. Before the implementation, because of the lack of visibility into upstream (and downstream) Equipment Groups, the fab had been facing considerable challenges at two particular Equipment Groups in deriving dispatch rules that would bring the performance beyond what operators had already been able to derive manually on the spot.
The first area entails a cleaning process comprising a four-lot batch operation for which an equal number of wafers in lots placed oppositely are required. Dummy wafers therefore need to be inserted into those lots containing fewer wafers compared to the opposite lot which involves manual effort. Hence for this Scheduling Group the scheduling procedure cannot just aim to maximise the batching efficiency but also needs to limit the number of dummy wafer insertions as much as possible. On top of that, in the specific case of this implementation the travel time between buildings also had to be considered.
The multi-objective approach led to an increase of the batching efficiency from below 84% to above 90%, corresponding to a capacity increase accordingly. Dummy wafer insertion was required for more than 20% fewer batches, and for those batches requiring dummy wafers the number of such dummy wafers was reduced by more than 35%. At the same time, the number of travel movements in between buildings was reduced by about 10%.
In turn, the Wet – Diffusion area entails a complex routing structure through 23 Equipment Groups as shown with approximately 70 tools. For this Scheduling Group, after the implementation of the D-SIMCON Scheduler it was possible to increase the number of wafers that can be processed within the critical time windows by close to 7%.
Overall, a consistent operator compliance (schedule adherence) of above 95% is now observed, compared to below 50% compliance with schedules generated during test periods when such schedules were actually shielded from operators.