IGBT production capacity planning is, at its core, a multiproduct and multistage resource-allocation problem: front-end wafer-out must be matched to back-end assembly/test throughput, with order priority and varietal compatibility driving the schedule [S2]. The Springer paper frames production capacity as the *available* capacity after virtual occupancy, not the nameplate nameplate — and that framing is what separates a defensible IGBT plan from a back-of-the-envelope guess [S2].
For an IGBT line, that means three binding constraints typically govern the plan: epitaxial/diffusion furnace hours, dicing-and-die-bond line balance, and the test-floor paralleism for high-current switching tests. The JD Edwards EnterpriseOne capacity-planning module groups this work as Resource Requirements Planning (RRP) and Rough-Cut Capacity Planning (RCCP) feeding the MPS and MRP outputs [S8]. On a 6-inch or 8-inch IGBT line, those steps must be reconciled against the equipment's rated wafer-out per shift before any delivery promise is quoted [S1][S8].
What "Capacity" Actually Means on an IGBT Line
Production capacity is the maximum output of a production facility over a defined period, measured in finished parts — not in wafers-started, not in die-bond placements [S3]. The MRPeasy definition is explicit: capacity is the theoretical upper limit given installed machines, labor, and current routings, and it must be compared against the actual output of past periods to derive a utilization ratio [S3].
Two metrics are non-negotiable for IGBT planning. The first is machine-hour capacity: the product of available machines and the hours labor can actually run them, calculated per process step [S3]. The second is throughput time, which on an IGBT module line routinely runs 3–6 weeks from epitaxial wafer start to tested, marked, and packed module. Without both numbers, the MPS is decoration. The Oracle NetSuite Advanced Manufacturing module restates the same idea as "the maximum amount of work your work centers can complete" inside a fixed window, then layers route and start/end dates on top [S7].
The Three Planning Horizons: MPS, MRP, RCCP
Capacity planning for an IGBT fab is not a single document — it is a stack of three horizons, each with a different decision it is allowed to drive. The MPS sets the rate at finished-goods level and is the only layer that should move customer commitments. MRP explodes the MPS into component and raw-material requirements, flagging shortages before they hit the line. RCCP (Rough-Cut Capacity Planning) is the coarse check that the key work centers — diffusion furnaces, implant, dicing saws, die-bonders, wire bonders, and the test floor — can absorb the MPS at the work-center family level [S8].
The Oracle WebLogic reference uses a different vocabulary (concurrent sessions, SSL overhead, network load) but the structural lesson transfers: capacity is the product of configuration × workload, and a single mis-sized layer (bandwidth, database, or in our case a wire bonder) caps the entire system [S1]. A 15% performance penalty — the figure Oracle cites for tunneling T3 over HTTP — is the same order of magnitude as the throughput hit an IGBT line takes when die-bond or test-floor capacity is under-specified relative to wafer-out [S1]. The MBA智库 PP overview confirms the same three-tier structure: Basic Data (PP-BD) → MPS (PP-MPS) → MRP (PP-MRP) → CRP (PP-CRP), with CRP being where IGBT-specific work-center calendars must be honored [S6].
Selection Criteria: Which Bottleneck Binds the Plan
For an IGBT line, the four criteria that decide whether the plan is feasible are: (1) diffusion/implant utilization versus rated wafer-out per hour, (2) dicing saw throughput versus wafer-out, (3) die-bond + wire-bond combined cycle time, and (4) test-floor parallelism for high-current switching tests. Whichever of those four sits at the highest sustained utilization over a 13-week rolling window is the RCCP flag that has to be addressed before the next MPS freeze. [S2]
The selection logic mirrors what the steel-plant paper recommends: varietal compatibility between different products, production capacity of variety per hour, minimum or maximum batch size, and transfer time all feed the available-capacity network [S2]. On an IGBT line, "variety" maps to voltage class (600 V / 1200 V / 1700 V) and current rating (e.g., 75 A / 150 A / 300 A modules), and "batch size" is bounded by die-bond pitch and the lead-frame strip layout. The same paper's genetic-algorithm-plus-local-search approach reduced order delay rate in simulation — directly relevant to IGBT planners who want a tractable optimizer for short-term replans [S2].
Use Cases: New-Line Build-Out vs. Running-Line Debottleneck
Two use cases dominate IGBT capacity-planning work in 2026. The first is a greenfield 8-inch or 12-inch IGBT line build-out, where the plan answers: how many diffusion tubes, how many implanters, how many die-bonders, and what test-floor footprint are needed to hit a target wafer-out per week? This is a capital-spend question and the open-pit coal-mine production-capacity-planning method in Scientific Reports is the closest published analog for fitting a throughput target to a fleet of equipment [S5]. The second use case is running-line debottleneck, where the question is narrower: given the existing tool set, where is the next 5–10% of throughput going to come from? That is the steel-plant use case, and the multiproduct/multistage available-capacity-network model is the published reference [S2].
For the greenfield case, the answer is dominated by the diffusion-furnace count and the test-floor parallelism, because those two are the longest-lead tool purchases and the most expensive to expand later. For the running-line case, the answer is usually die-bond cycle time, because the wire-bonder fleet is sized to a 600 V product mix and a 1200 V/300 A module pulls more bond cycles per strip. Both use cases demand a written capacity plan, not an Excel back-of-the-envelope — and the plan must show assumptions, not just outputs [S7][S8].
Failure Modes and Limits of the Plan
The plan fails in three recognizable ways. The first is RCCP green at the work-center family level but red at the individual tool level — a coarse check passes, but a single bonder is the choke point, and the MPS makes promises the line cannot keep [S8]. The second is ignoring the "available" part of available capacity: virtual occupancy from WIP between operations is not subtracted before RCCP runs, so the plan sells capacity that is already spoken for [S2]. The third is locking the MPS to a single product mix and ignoring varietal compatibility — a 1200 V/300 A module and a 600 V/75 A module cannot share a die-bond setup without re-tooling, and the plan must account for that changeover time [S2].
Capacity planning is not an exact science and every application is different, which is Oracle's own admission in the WebLogic capacity-planning chapter and it applies unchanged to an IGBT line [S1]. The corollary is that the plan must be re-baselined on a fixed cadence (typically weekly for MRP, monthly for RCCP, quarterly for the long-range plan) and must be re-baselined out-of-cycle whenever a tool goes down for more than 48 hours or a new product mix exceeds 10% of the MPS [S7][S8].
What to Track From Here
Two trackable signals tell you whether the IGBT capacity plan is holding. First, the rolling 13-week RCCP utilization on the binding constraint — diffusion furnace, die-bonder, or test floor — should stay below the red-line threshold (commonly 85%); above that, the plan starts to slip on the first non-trivial disturbance. Second, the order-delay rate from the most recent replan cycle should be the headline KPI for the planning team, not wafer-out, because the steel-plant evidence is that an improved genetic-algorithm-plus-local-search replanner measurably cut order delay in simulation [S2]. For an IGBT planner evaluating power-semiconductor production line design decisions, both signals should be wired into the monthly capacity review before the next MPS freeze.
Spec-level background on the components involved: pressure transmitter, flow meter, and industrial valve.