An Introduction to Fab-Broad Scheduling
The semiconductor business is without doubt one of the largest and most complicated industries on the earth. The vital elements in semiconductor manufacturing are the flexibility to quickly develop and check novel applied sciences, enhance manufacturing processes to cut back rework and waste, in addition to meet manufacturing targets by way of prescribed volumes and due dates. On this context, top quality scheduling is of paramount significance.
Because of the lengthy cycle instances, the place a wafer is processed over a span of months, decision-making in semiconductor fabrication vegetation (fabs) is often framed as a two-level downside. On one hand, international scheduling (or fab-wide) is tasked with the strategic administration of manufacturing facility property whereas contemplating all work-in-progress, incoming and outgoing flows throughout the fab, anticipated useful resource availability and different constraints. Then again, native (or toolset-level) scheduling focuses on the operation of particular person work centres. It’s sometimes tasked with figuring out the most effective rapid dispatch selections i.e. which jobs ready for dispatch needs to be assigned to which obtainable machine.
Most growth efforts to this point have targeted on the shorter timeframe dispatch selections i.e. native scheduling. It is a extra manageable downside since there may be little look-ahead and the scope is proscribed to a single or just a few toolsets. Regardless of quite a few analysis efforts, to this point there has not been a broadcast case examine of a fab-wide scheduler efficiently deployed in a big semiconductor manufacturing facility. However, the potential for enchancment on the fab-wide stage is large; there are quite a few alternatives to enhance all through and have a step change in efficiency. For instance:
- Bottlenecks happen attributable to repetition of course of loops, high-cost machines with low capability, and different bodily or operational constraints. To handle them, a strategic strategy is required that appears on the larger image and avoids early dispatch of wafers that can find yourself in a bottleneck space.
- WIP move management mechanisms (kanbans) are essential for high quality management however can block high-priority wafers. Fab-wide scheduling can tremendously enhance this facet of operation.
- Timelinks (also referred to as timeloop, time lag, or qtime constraints) are difficult as a result of they outline the minimal or most period of time between two or extra consecutive course of steps, resulting in a conundrum of preserving downstream machines idle or not. Fab-wide scheduling can tremendously help by precisely predicting arrival instances and deciding when to set off timelinked heaps.
Methodology
The scheduling framework proposed on this weblog is hierarchical and consists of two foremost elements which run independently and at completely different frequencies — the Toolset Scheduler (TS) and Fab-Broad Scheduler (FWS).
The Toolset Scheduler considers the presently in-process and/or upcoming course of step of all wafers within the cluster.
FWS takes a view of your entire fab directly and considers a number of future steps for every wafer. It focuses on bettering schedule high quality by contemplating the move of wafers by the fab, one thing the toolset scheduler can’t do attributable to its singlestep, toolset-level nature. The principle objective is to redirect move by the fab and thereby enhance move linearity, scale back bottlenecks, enhance WIP move management administration, and scale back timelink violations.
Our FWS strategy achieves this by predicting wait/cycle instances for a number of future steps, analysing these predicted wait/cycle instances with respect to the completely different areas of potential enchancment, and re-prioritising wafer steps in a means that ensures improved (weighted) cycle instances. Briefly, FWS combines two foremost components: (i) an operational module that captures in full element all related constraints e.g. detailed course of time modelling, machine upkeep, shift adjustments, dynamic batching constraints, kanbans and so forth. (ii) a search module that identifies helpful precedence adjustments given the evolving fab situations and state options.
FWS communicates with the toolset schedulers by way of precedence weights (and another predicted timing data) for particular person steps of a wafer, as proven in Determine 2. A bonus of our strategy is that, whereas FWS all the time schedules all instruments within the fab, customers can specify which toolsets are topic to steerage; FWS adjusts its search accordingly. That is notably helpful for steadily rolling out FWS in a fab and evaluating its affect. As well as, the steerage power is controllable – though full steerage is the optimum selection, tuning down steerage permits for a extra gradual deployment.
Seagate Deployment
Seagate is a world chief in information storage expertise, with greater than 40% share of the worldwide Onerous Disk Drive (HDD) market. The Springtown facility in Northern Eire produces round 25% of the entire international demand for recording heads, the vital part in a HDD. Flexciton’s FWS / TS scheduling system was trialled in Seagate Springtown between March-Might 2022. After profitable testing, the system has been operational 24/7 since June 2022; a timeline is proven in Determine 3.
You will need to be aware that deploying and testing a novel piece of expertise in a big manufacturing facility that runs across the clock presents many sensible challenges to be overcome:
- Controllability (scope): essential to make sure that the brand new growth is deployed in a managed method. The FWS-TS steerage scheme permits for localised trials, the place focus could be positioned on problematic areas and steadily enhance scope.
- Controllability (magnitude): it’s helpful to solely give attention to instances with apparent benefit first. That is achieved by controlling steerage power.
- Explainability: essential to have the ability to detect and cause concerning the adjustments. That is achieved by a mixture of UI options and help instruments which have been designed to provide operators and managers situational consciousness.
Outcomes and Learnings
Quantifying the advantage of an alternate scheduling strategy stays a difficult activity. When deployed in an actual plant, conventional A/B testing between pre and post-deployment undergo from (i) dynamic fab situations (ii) an ever-changing product combine and (iii) evolving capabilities of the fab e.g. elevated/decreased labour capability and new instrument commissioning/decommissioning.
As such, it was determined to have a look at the affect from completely different angles – a statistically important affect can be anticipated to end in a considerable shift in quite a few enterprise processes and metrics. Specifically, three completely different elements have been examined.
Deep dives on particular toolsets and metrics.
Comparability towards inside simulation and planning instruments.
Observing the affect on guide interventions.
Notably, all three approaches indicated a change in fab efficiency between pre and post-deployment; extra particulars will probably be shared in future articles. Within the Winter Sim Convention paper offered in December 2022, we targeted on the latter level; A proxy we will use for this profit is the quantity of advert hoc management move guidelines activated/deactivated within the fab. Day-after-day, specialists need to outline quite a few, in some instances even lots of, of advert hoc management move guidelines to higher handle operations given the prevalent situations.
For instance, setting a ”arduous down” rule, the place heaps are manually positioned on maintain in order to not proceed to a downstream bottleneck. In Determine 5, we present the variety of advert hoc operational guidelines carried out within the Seagate Springtown fab between weeks 2 and 26 of the 12 months 2022 (i.e. from early January till late June). As could be seen within the ultimate weeks, the variety of advert hoc rule transactions averaged lower than 150 per week, a lower of over 300% in comparison with the pre-deployment interval. That is sturdy proof that FWS deployment lowered massively guide interventions required to successfully management flows throughout the fab.
Conclusions
The principle takeaway of the Winter Sim paper is that the elevated horizon look-ahead and international nature of FWS presents quite a few alternatives for a step change in manufacturing facility KPIs. The Flexciton FWS was efficiently trialled at Seagate Springtown over 3 months in 2022 and has been totally enabled throughout the fab since June 2022. It resulted in a radical lower of interventions beforehand used to manually management wafer flows. Additional evaluation means that Flexciton’s TS and FWS schedulers have achieved substantial enhancements in throughput and cycle instances.