Wip (work in Progress)
Work in progress (WIP) is the inventory of unfinished wafers inside a defined manufacturing boundary. It includes wafers being processed, waiting for a tool, moving between stations, or held for a decision, provided those states belong to the chosen boundary. WIP answers “how much material is inside?” Throughput answers “how quickly does material leave?” They are different quantities.
A fab-wide count might include all released wafers until they leave the defined fab route. A tool-group count might include only its queue and material being processed. State the boundary before comparing numbers: a smaller reported WIP can reflect a narrower definition rather than faster manufacturing.
Use inventory units for WIP. Wafers and lots are counts; wafers per day and wafer starts per week are rates. Lot counts can hide a change in wafers per lot, so they are not interchangeable without a conversion. Workload measures such as expected processing hours answer another useful question: how much work is waiting? Two queues with the same wafer count can impose very different loads on a tool group.
Little's Law connects the averages. For a consistently defined population and boundary, with suitable long-run averages, average WIP equals average throughput multiplied by average time inside the system. In fab terminology that residence time is often called cycle time. Here it includes waiting and holds, not just the time spent under active processing.
For example, suppose a stable, simplified route has no scrap or diversion, averages 12,000 wafers inside, and completes 200 wafers per day. Its implied average cycle time is 60 days: 12,000 wafers divided by 200 wafers per day. These are invented numbers for arithmetic, not a benchmark for a technology node.
The equation is an accounting relationship, not a promise that cutting releases will preserve throughput. If too little eligible material reaches a constrained tool, it can become starved and output can fall. Nor does today's WIP divided by yesterday's shipments reliably predict a newly released lot's completion date. Changing product mix, downtime, release rates, and holds can make short measurement windows misleading.
Scrap and diversion require care. If the counted inventory includes wafers that later leave through scrap or other exits, using only good-wafer shipments as the denominator mixes populations. Define the exits and the associated residence-time population consistently, or analyze the populations separately. Similarly, internal process moves are not final wafer departures: the same wafer can generate many moves before leaving the fab.
More WIP can feed a bottleneck, then mostly feed a queue. Extra material can initially reduce starvation. Once output is constrained by available capacity, adding more inventory cannot make that capacity unlimited. Variability in arrivals, processing times, availability, and eligibility changes how much buffering is useful and where it should sit. There is no universal WIP number that optimizes every fab.
Where the WIP sits matters. A wafer fab is a re-entrant system: wafers can revisit the same equipment family at different stages of the route. A large queue may contain material that cannot use the currently available chamber because of recipe qualification, processing history, or a hold. A high total count therefore does not prove that a tool has runnable work. Track eligible work and remaining workload alongside raw inventory.
An aging queue also deserves attention even if its size is stable. The oldest material may be trapped behind repeated priority changes or a persistent engineering hold while newer lots keep moving. Average inventory alone does not reveal that distribution. Useful views separate processing, runnable waiting, blocked waiting, and hold states, with age and product family available for investigation.
Release control and dispatching solve different problems. Release control decides when new work enters the boundary. Dispatching decides which eligible lot runs next. Limiting release can help contain inventory, while reducing downtime, unnecessary holds, or avoidable waiting can improve flow. A sound policy checks output and cycle time together rather than celebrating a lower WIP count that was achieved by starving production.
The simplest inventory reconciliation is beginning WIP plus entries minus exits equals ending WIP. Internal transfers do not change fab-wide WIP, although they change area counts. Splitting one lot into two changes the lot count but need not change wafer inventory. These checks help distinguish a real operational improvement from a change in reporting or grouping.
The key idea: WIP is stored unfinished work, not production speed. Interpret it together with output, time in the system, material eligibility, and a clearly stated boundary. The goal is reliable flow with appropriate buffering, not the largest inventory or the smallest number in isolation.
Sources: INFICON FabTime cycle-time tutorial and queueing formulas; FabTime forward-looking cycle-time metrics and fab trade-offs; MIT Lean Thinking teaching material. Numerical examples and graphs are illustrative, not fab performance claims.