# MES, dispatch and process release Canonical: https://semiaifoundry.com/learn/atlas/manufacturing-execution-dispatch/ Revision: 2026-10-07 Rights: https://semiaifoundry.com/ai-atlas/licenses.html Fabrication / Factory systems · Atlas v1.3.0 # MES, dispatch and process release Coordinates material routes, equipment eligibility, holds and measurement decisions. [Explore in Atlas](https://semiaifoundry.com/ai-atlas/#manufacturing-execution-dispatch)Chamber, wafer carrier and fab floor; schematic routes Learning objectiveSeparate dispatch, qualification, process control and material release. ## The mechanism A manufacturing execution system records where material is and whether the next operation is permitted. Dispatching chooses among eligible lots and tools, while engineering holds prevent suspect material from continuing. These are different decisions from selecting a process recipe or statistically controlling its result. Re-entrant fabrication makes scheduling difficult because wafers revisit tool families many times and share them with other products. Prioritizing one lot may delay another without increasing total output. The useful factory view therefore connects route state, tool qualification, queue time, inspection and genealogy, and distinguishes an urgent order from a physically feasible production path. A model recommendation can be an input to dispatch without becoming the authority that releases a lot or selects equipment. Route eligibility, holds, qualification, current state and authorized decisions remain distinct control conditions. The fictional incident workflow illustrates this boundary using inventory and work orders; it should not be mistaken for an implementation of a real fab’s MES or private dispatch algorithm. 1. Represent the actual route, including repeated visits and required measurement gates. 2. Determine which tools are qualified and currently available for each operation. 3. Apply holds and queue-time constraints before ranking eligible material. 4. Dispatch and record the operation, retaining recipe and equipment identity. 5. Reconcile completed work with inspection and downstream demand rather than optimizing isolated tool counts. 6. A recommendation retains its model/data provenance and the asset or lot identity to which it applies. 7. An authorized dispatch evaluates current holds and qualification before creating a concrete operational effect. ## Science and assumptions ### Flow conservation $$ \mathrm{WIP}_{\mathrm{next}}=\mathrm{WIP}_{\mathrm{now}}+N_{\mathrm{arrivals}}-N_{\mathrm{departures}} $$ Next work in process equals current work in process plus arrivals minus departures, all in the same material unit. Symbols and units WIP\_next Material in process after the interval same declared material units WIP\_now Material in process before the interval same declared material units N\_arrivals Material entering during the interval same declared material units N\_departures Material leaving during the interval same declared material units All terms count the same material unit over one time interval; WIP means work in process. Assumptions Count split/merge events explicitly. Wafer, die and package counts cannot be added without conversion. ## Engineering example Synthetic shift: 120 wafers arrive at a qualified step and 90 depart. Its queue grows by 30 wafers unless material is diverted or held elsewhere; raising arrival priority does not remove the imbalance. Suppose a classifier flags a hypothetical chamber for inspection while a lot is held pending qualification. The flag may create an inspection recommendation, but it does not release the hold. A workflow may seek the authorized review and record a work order; it still cannot mark the lot eligible merely because a model prediction was confident. This separates prediction quality from the controlled operational action. ## Distinctions and tradeoffs A scheduling priority cannot create qualified capacity in a tool family that is already the binding constraint. Local utilization can conflict with cycle time and time-sensitive surface requirements; a busy tool is not automatically improving system output. Queueing and release decisions affect when logic dies and HBM stacks become available for package convergence. ## Specifications and interfaces Route state Eligible, processing, queued or held A material state at a timestamp, not a final yield prediction. Tool eligibility Qualified product/process combination Installed equipment may not be released for every route. ### Interfaces Orders, route and tool state Input Required operations, priorities and qualification constraints Dispatch and genealogy events Output Authorized movement and a traceable process history ## Check your understanding ### Why can maximizing every tool’s utilization increase cycle time? 1. All fabrication steps have identical capacity 2. Extra work-in-process can build queues ahead of a bottleneck 3. More queued wafers always mean more finished packages Answer and reasoning Extra work-in-process can build queues ahead of a bottleneck Upstream releases can exceed downstream throughput, increasing waiting without increasing completed output. ### A model recommends processing a held lot. What establishes whether dispatch may proceed? 1. The confidence score alone 2. Current route/hold/qualification and authorization conditions 3. Whether the recommendation is fluent Answer and reasoning Current route/hold/qualification and authorization conditions Dispatch authority and eligibility remain external operational constraints, not model-generated permissions. ## Evidence and scope TSMC describes public capabilities, not its proprietary rules, live queues or customer priority policy. The lesson uses generic operations reasoning. The scenario is fictional and qualitative. No manufacturing recipe, production dispatch scheduler, equipment interface or real lot release is executed or inferred. No current calculator directly models this lesson’s main mechanism. Use the library for related bounds while retaining their stated omissions. ## Sources - [TSMC — Intelligent Packaging Fab](https://www.tsmc.com/english/dedicatedFoundry/services/apm_intelligent_packaging_fab) - [Edwards — Semiconductor vacuum and abatement](https://www.edwardsvacuum.com/en-us/semiconductor) - [Temporal Activities](https://docs.temporal.io/activities) [Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/manufacturing-execution-dispatch-bd3550e94e978ca6.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html) ## Continue exploring ### Foundations - [Wafer handling, carriers and genealogy](https://semiaifoundry.com/learn/atlas/wafer-handling-genealogy/) ### Related lessons - [Cycle time, ramp and effective capacity](https://semiaifoundry.com/learn/atlas/fab-ramp-cycle-time/) - [HBM stack assembly, repair and staged test](https://semiaifoundry.com/learn/atlas/hbm-stack-assembly-test/) - [User authority and service identity](https://semiaifoundry.com/learn/atlas/tool-authority-identity/) - [Workflow graphs and explicit state](https://semiaifoundry.com/learn/atlas/workflow-graphs-state/) - [Evaluation & the AGI frontier](https://semiaifoundry.com/learn/atlas/evaluation/)