Semiconductor Fab Transformer: How to Size the MV/LV Distribution for a Wafer Fab, Why Voltage Sags and Harmonics Decide the Design

The answer first: size a fab's transformers from connected tool load, then let sag immunity and harmonics rewrite the specification

A 300 mm wafer fab is built around three transformer layers: utility service transformers of 60–100 MVA at 230/34.5 kV or 132/33 kV for a 90–150 MW load, MV/LV unit substations of 2.0–2.5 MVA at 13.8/0.48 kV or 11/0.4 kV repeated across every cleanroom bay, and small isolation or 400 Hz units at tool level. Size each layer from connected tool load multiplied by a demand factor of 0.55–0.70, never from the sum of nameplates — a fab's connected load is normally 1.4–1.7 times its actual peak, so nameplate summation oversizes the whole chain by a third. Two factors then override the arithmetic: a 200 ms voltage sag can scrap every wafer in process, which makes impedance, redundancy and transfer scheme a process-yield decision rather than a cost decision; and switch-mode tool loads with 25–45% current THD make K-factor selection and harmonic derating mandatory instead of optional. The result is a specification that looks expensive per kVA and is in fact the cheapest 20-year option a fab can buy.

Key takeaways

• Utility layer: three 60 MVA units at 230/34.5 kV, %Z 12%, on-load tap changer ±8 × 1.25%, fed from two independent grid connections. For a 105 MVA peak that is 58% normal loading and 87% on loss of one unit — N+1 without buying a fourth transformer.
• Substation layer: 2,000–2,500 kVA cast-resin units at 13.8/0.48 kV or 11/0.4 kV, %Z 5.75–6.5%, loaded to 60–65% and K-13 minimum. A 100 MW fab lands on 55–75 unit substations — the count is driven by loading target and 2N on critical buses, not by megawatts.
• SEMI F47-0706 requires tools to ride through sags to 50% retained voltage for 0.2 s, 70% for 0.5 s and 80% for 1.0 s. No transformer alone can deliver that; it is met by a static transfer switch under 4 ms, a flywheel or UPS, and by keeping substation impedance low.
• Harmonics are the second design driver: tool front ends run 25–45% current THD with strong 5th and 7th, so apply IEEE C57.110 derating (0.85–0.92 at K-13) and hold IEEE 519-2022 limits at the point of common coupling — never with capacitor banks, which resonate on a fab bus.
• A fab energises 8,760 hours a year at 58–87% load, so loss capitalisation reaches 30–40% of purchase price. For 65 unit substations that is USD 2.5–3.5 million of evaluated losses, larger than the annual maintenance budget of the whole distribution system.
• Lead time is a schedule risk, not a commercial detail: 12–18 weeks for cast-resin substation units, 30–50 weeks for the 60 MVA utility transformers, plus 4–8 weeks for seismic calculation, FAT witnessing and third-party inspection. Order before the cleanroom steel goes up.

How much power does a wafer fab actually draw, and why does that decide the transformer size?

Fab power demand is set by wafer starts, technology node and — more than most people expect — by the facility systems that keep the cleanroom alive. A leading-edge logic line consumes 1.8–2.5 times the electricity of the same floor area built for 28 nm, because extreme ultraviolet lithography, more process steps and tighter contamination control all add load without adding wafer output.

The split between process tools and facility is the number that surprises project teams: process equipment accounts for only 40–55% of demand. Chillers and process cooling water, clean dry air and vacuum, exhaust and abatement, ultrapure water, and the cleanroom air-handling units together consume the other 45–60%. Those loads are large induction motors with variable-frequency drives, they run continuously, and they are the reason a fab's transformer specification must be written for a motor-heavy bus rather than for electronic load alone.

Two consequences follow directly. First, the utility transformers are sized on a load that is known to within perhaps ±10% at design stage, so a 15–25% capacity margin for tool hookup growth is normal and cheaper than adding a transformer to a live 34.5 kV bus three years later. Second, because the load grows in steps of 5–15 MW as each tool set is commissioned, the substation layer should be modular: identical 2,500 kVA units added bay by bay, with spare 34.5 kV or 13.8 kV breaker positions designed in from day one.

How many transformers does a fab need? Five duties that are constantly confused

A fab is frequently tendered as if it needed “one main transformer”. It needs five distinct duties, and each one is specified from a different input. Buying them as one package without separating the duties is the most common origin of a technical mismatch discovered during commissioning.

The fifth duty is the one most often forgotten in tenders. Ion implantation, some metrology and several legacy 200 mm tool sets operate at 208 V or 400 Hz, and they need either a dedicated isolation transformer or a solid-state frequency converter with its own transformer. These units are small in kVA and disproportionately expensive per kVA, and they are installed inside the sub-fab where noise and heat rejection matter more than efficiency.

How do you size the substation transformers without oversizing them?

Work bottom-up from tool lists and top-down from the utility connection, then reconcile the two. The bottom-up build gives the number the electrical design actually needs; the top-down figure is the contractual limit. When the two differ by more than 15%, one of the assumptions — usually the demand factor or the facility load — is wrong.

At a 60–65% loading target on 2,500 kVA units, 105 MVA of peak demand needs about 162 MVA installed, which is 65 unit substations. Load them to 80% instead and the count falls to 53 — but each unit then runs 15–20 K hotter, loses 8–12% more energy, and has no headroom when the next tool set arrives. In a fab energised continuously for twenty years, the 60% target pays for the extra transformers in loss savings within roughly seven years, and it keeps the 2N critical buses genuinely independent.

The utility layer then follows the same logic with fewer, larger units: 105 MVA peak on three 60 MVA transformers gives 58% normal loading and 87% with one unit out of service. Two independent grid connections, each capable of carrying the full load through the 34.5 kV bus tie, are what make that 87% acceptable. A single radial feed with three transformers is not N+1; it is N with spare iron.

Why does a voltage sag cost more than the transformer that could prevent it?

Semiconductor process tools are the most sag-sensitive industrial load in existence. A dip that a cement plant would never notice will interrupt an etch chamber mid-process, and every wafer in that chamber, plus the ones queued behind it in the same run, is scrapped. Industry experience puts the cost of an unplanned one-hour fab outage at USD 1–10 million once scrapped work-in-process, requalification and restart time are counted — which is several times the price of the substation transformers serving that area.

SEMI F47-0706 defines what tools must tolerate: 50% retained voltage for 0.2 s, 70% for 0.5 s, 80% for 1.0 s, measured at the tool input terminals. IEC 61000-4-11 and IEC 61000-4-34 define the equivalent immunity tests. In practice 70–90% of sags originate on the utility network — lightning, adjacent feeder faults, capacitor switching, construction damage — and the fab cannot remove them by specification. What the electrical design can do is shorten and shallow them.

• Keep substation impedance low and transformers close to the load, so a fault on a parallel feeder produces a shallower dip at the tool.
• Use a static transfer switch with a transfer time under 4 ms (a quarter cycle) between the two 2N sources; mechanical automatic transfer switches at 100–300 ms are far too slow for lithography and etch.
• Add ride-through for the critical bus: a flywheel or lithium-titanate UPS covering 2–10 s, sized for the tools that cannot tolerate even a 4 ms transfer.
• Coordinate the 34.5 kV protection so that a downstream fault clears in 3–5 cycles instead of 10–20; faster clearing is cheaper than more copper.

This is why the two 2N substations in a critical bay are specified as a pair with identical impedance. Mismatched impedance between the normal and alternate source produces a voltage step at the moment of transfer, and a 5% step is enough to trip some tool power supplies that had already survived the sag.

How should the MV network be arranged for redundancy and arc-flash safety?

The standard arrangement is a dual-source main-tie-main 34.5 kV switchgear lineup, operated open-tie, with each half fed from a different utility transformer and each capable of carrying its own load plus the critical block of the other. Closed-transition transfer is possible but requires the two sources to be synchronised, and on a fab bus with 30–40 kA fault levels that synchronisation check is itself a risk item.

Fault level is the constraint that drives equipment cost. A 105 MVA load fed at 34.5 kV from three 60 MVA transformers with 12% impedance produces a three-phase fault current around 25–30 kA at the 34.5 kV bus, and 40–50 kA at the 480 V substation busbars. That is why 480 V switchgear in a fab is nearly always rated 65 kA or 100 kA, and why transformer impedance is sometimes raised deliberately to hold the low-voltage fault level below the rating of existing switchgear. The calculation is set out in our short-circuit current guide, and it should be done before the substation transformer impedance is frozen.

Grounding deserves equal attention. Fab MV systems are normally resistance-grounded at 200–400 A for 10 s: enough to detect and clear a feeder fault, low enough to limit damage at the fault point and to keep stray current away from sensitive metrology. Solid grounding on a 34.5 kV fab bus produces very high single-line-to-ground currents and a correspondingly severe arc-flash hazard. At 480 V, the convention is solid grounding with high-resistance grounding reserved for selected control and emergency buses — see the low-voltage grounding guide for the decision tree.

Arc flash is not a theoretical concern on a fab site: 480 V switchgear and motor control centres are where most incident-energy calculations land in the “dangerous” band, because the transformers sit close to the loads and the available fault current is high. IEEE 1584-2018 calculations, current-limiting fuses or breakers, remote racking and arc-rated labelling all belong in the electrical package, not in a later safety study.

Which transformer type belongs inside a fab and which belongs outside?

The dividing line is fire load and water. Oil-immersed transformers are acceptable in an outdoor yard with blast walls and oil containment; cast-resin dry-type units are what belong inside the electrical rooms, the sub-fab and anywhere above or below a cleanroom. A 2,500 kVA oil-filled unit holds roughly 1,200–1,800 litres of mineral oil, which is both a fire load and, if it leaks, a contamination event in a building where a single particle excursion can cost a production week.

Two specification details are specific to fabs. First, cast-resin windings should be vacuum-cast rather than open-wound and impregnated, because the filled epoxy is what gives the low partial-discharge behaviour that a 20-year continuous duty needs; specify partial discharge of 10 pC or less at 1.2 times rated voltage and require it as a routine test on every unit, not a type test on one. Second, ask for low-flammability and low-smoke materials with a declared halogen content: a cable or transformer fire inside a cleanroom building contaminates the space, and the clean-up cost exceeds the equipment cost by an order of magnitude. The wider comparison of construction types is in the oil versus dry-type selection guide.

Why are harmonics the second design driver in a fab?

Almost every load in a fab is a power-electronic converter: tool power supplies, VFDs on chillers and pumps, UPS rectifiers, LED drivers and 400 Hz converters. The current they draw is not sinusoidal, and the transformer that supplies them must be specified for the current it will actually carry rather than for its nameplate.

Apply IEEE C57.110 to convert that distortion into a derating factor. For a dry-type unit at 30–35% THD-i the practical factor is 0.85–0.92, meaning a 2,500 kVA K-13 transformer delivers 2,125–2,300 kVA of usable capacity. Ignoring the derating means running the unit above its thermal limit for years — the classic failure mode in a fab is a K-4 substation transformer with a 12-year life instead of 25. Our K-factor selection guide and the harmonics guide give the calculation and the measurement method.

At the point of common coupling, IEEE 519-2022 governs: 5% total voltage distortion and 3% for any single order at voltages up to 69 kV, with current distortion limits scaled by the short-circuit ratio. A fab with 480 V buses at 65 kA available fault current has a short-circuit ratio around 40–60, so the current limits are tight and TDD allocation must be managed tool group by tool group. Do not solve residual distortion with capacitor banks: on a bus with 25–30% THD-i, the 5th and 7th harmonic resonances will amplify voltage distortion rather than reduce it. Use detuned reactors, active filters on the 480 V bus, or 12-pulse and active-front-end drives instead. The power-factor correction that a fab actually needs is modest — modern front ends already run 0.95–0.98 — and that is fortunate, because capacitors are the wrong tool here.

What does the cleanroom environment do to the transformer specification?

Three environmental factors change the mechanical and thermal design, and all three are routinely omitted from Chinese and Southeast Asian fab tenders because they do not appear on a nameplate.

Temperature and airflow. A 2,500 kVA cast-resin unit dissipates 20–24 kW of loss as heat into the electrical room. That heat is removed by the cleanroom air-conditioning system, which means the transformer is simultaneously a load and a cooling problem. Specify 40 °C ambient as a minimum, verify the room's actual air change rate, and derate the nameplate by about 1% per K above 40 °C — a plant room at 46 °C costs 6% of capacity before any harmonic derating is applied. Where the room is tight, an IP31 enclosure with forced ventilation is better than an oversized transformer in still air.

Noise and vibration. Core magnetostriction produces 100 Hz noise and vibration on a 50 Hz system, 120 Hz on 60 Hz. A 2,500 kVA dry-type unit at 58–62 dB(A) is unremarkable in a plant room but unacceptable next to a metrology or lithography bay, where floor vibration criteria of VC-D (6.25 µm/s) to VC-E (3.1 µm/s) apply. The fix is location and isolation, not quieter iron: keep transformers out of the structural bays that carry lithography tools, mount them on spring isolators where they must be close, and specify sound pressure level as a guaranteed tested value with a tolerance of +2 dB. Our noise level guide covers the measurement standard and the specification wording.

Particles and outgassing. Anything inside a cleanroom envelope must not shed. That means sealed windings, no open cooling ducts in ISO Class 5 areas, gasketed enclosures, and cable entries from below. It also means the transformer room is normally outside the cleanroom proper, on a separate air system, with the sub-fab serving as the buffer — a layout decision that must be taken with the architectural team, because a transformer that arrives after the waffle slab is poured may not fit through the available opening.

How do seismic, floor loading and transport into a fab change the order?

Fabs are built in seismic regions as often as not, and a transformer is one of the heaviest items in the electrical package. A 2,500 kVA cast-resin unit weighs 6–8 t; a 60 MVA oil-immersed unit with oil weighs 90–120 t. Three checks decide whether the specified unit can actually be installed.

The seismic package is a delivery item, not paperwork: a certified anchorage calculation and a shake-table or analysis-based qualification report are required before the equipment can be released for installation in most Asian and North American fabs. Budget 4–8 weeks for it and make it a contractual milestone, because a transformer that arrives on site without its qualification report sits in storage while the schedule slips.

How should you specify efficiency and losses for a 20-year asset?

A fab transformer runs 8,760 hours a year at 58–87% load, which is the most severe duty in industrial power distribution. Loss capitalisation is therefore not a refinement; it is the dominant part of the owning cost, and it belongs in the tender as an explicit evaluation formula rather than a vague request for “low loss”.

Two practical points. First, the A and B factors should reflect the fab's real energy cost and loading profile: at USD 0.09/kWh and 8,760 hours, the theoretical no-load factor is about USD 7,900/kW before discounting, so USD 6,000/kW is a defensible twenty-year present-value figure. Second, the EU Ecodesign regulation for transformers (2019/1783, amending 548/2014) sets minimum efficiencies only up to a few megavolt-amperes; a fab's 60 MVA units fall outside its scope, which means loss evaluation is a commercial decision the buyer must make explicitly. Our total cost of ownership guide and the efficiency standards comparison give the full calculation, including the interaction between loss capitalisation and the 60% loading target.

The interaction is worth stating plainly: if losses are capitalised at USD 1,200/kW, running a substation at 60% instead of 80% loading cuts load loss by 44% and saves roughly USD 8,000–10,000 per transformer over twenty years. That is the arithmetic that justifies the extra 12 units in the 100 MW example.

What tests prove the design before it ships?

A fab's inspection regime is heavier than a normal industrial project because the equipment cannot be easily replaced once the cleanroom is operational. Specify tests, not just standards, and require witnessed factory acceptance testing with a third-party inspector appointed by the owner.

The single most valuable item is the type test report from an accredited independent laboratory, dated and traceable to the design — not to “a similar unit”. A short-circuit withstand test and a temperature-rise test on the actual design are what separate a specification from a hope. Our pre-shipment inspection checklist and supplier audit checklist list the documents to request before the purchase order, not after.

What does a fab transformer cost in 2026 and how long does it take?

Prices below are indicative FOB China ranges for single units, excluding freight, duties, installation and seismic qualification, and they move with copper and grain-oriented steel. Treat them as a budget check, not a quotation.

The schedule risk is not the transformer itself but the sequence around it: design freeze, seismic calculation, type test availability, FAT witnessing and ocean freight add 8–14 weeks to the quoted lead time. For a fab, the utility transformers must be ordered before the structural steel package, because a 60 MVA unit that arrives four weeks late delays energisation of the entire facility. See the lead time guide for the full critical-path view, and the FOB quotation breakdown for what should and should not be inside the price.

What are the five specification mistakes that cost fab owners the most?

• Summing tool nameplates. It oversizes the utility transformers by 40–70%, and an oversized transformer runs at low load with poor efficiency and a larger footprint that the plant room may not have. Use demand factors of 0.55–0.70 and reconcile bottom-up against the utility limit.
• Buying on price per kVA. A 15% cheaper unit with 20% higher load loss costs more than its saving within four years at fab duty. Put the capitalisation formula in the tender and evaluate the quoted losses, not the quoted price.
• Treating SEMI F47 as the tool supplier's problem. Tool immunity is a specification, but the sag depth and duration at the tool terminal are set by the distribution design. Without a sub-4 ms static transfer switch and a defined ride-through block, compliant tools still trip.
• Specifying K-4 substation transformers. On a 25–30% THD bus, K-4 units lose 12–15% of usable capacity and age prematurely. K-13 is the fab standard, and the cost difference is 6–10% of the unit price.
• Ignoring the installation path and the seismic package. The cheapest transformer is useless if it cannot pass the door or if its anchorage report arrives after the slab is poured. Check rigging, floor loading and qualification deliverables at the design stage.

None of these five is a manufacturing defect. All of them are specification errors that no factory test will catch, and each one costs more than the difference between a good transformer and a cheap one. If you are sizing a fab distribution system now, the sizing and budget formula gives the first-pass arithmetic, the selection guide covers the technology choices, and our engineers can review a tool list and a single-line diagram against the five points above before the purchase order is placed.

Sources / 资料来源

Source: IEC 60076-1 — Power transformers, Part 1: General.
Source: IEC 60076-2 — Temperature rise for liquid-immersed transformers.
Source: IEC 60076-3 — Insulation levels, dielectric tests and external clearances in air.
Source: IEC 60076-5 — Ability to withstand short circuit.
Source: IEC 60076-10 — Determination of sound levels.
Source: IEC 60076-11 — Dry-type transformers.
Source: IEC 60076-20 — Energy efficiency.
Source: IEC 61000-4-11 — Testing and measurement techniques: voltage dips, short interruptions and voltage variations immunity tests.
Source: IEC 61000-4-34 — Voltage dips immunity tests for equipment with mains current above 16 A per phase.
Source: SEMI F47-0706 — Specification for semiconductor processing equipment voltage sag immunity.
Source: IEEE 519-2022 — Recommended practice and requirements for harmonic control in electric power systems.
Source: IEEE C57.110 — Recommended practice for establishing transformer capability when supplying non-sinusoidal load currents.
Source: IEEE C57.12.01 — General requirements for dry-type distribution and power transformers.
Source: IEEE C57.12.90 — Test code for liquid-immersed distribution, power and regulating transformers.
Source: IEEE 1584-2018 — Guide for performing arc-flash hazard calculations.
Source: IEEE 693-2018 — Recommended practice for seismic design of substations.
Source: IEC 60068-3-3 — Seismic test methods for equipment.
Source: UL 1561 — Dry-type general purpose and power transformers (K-factor ratings).
Source: NFPA 70 — National Electrical Code, Article 450 (transformers).
Source: ISO 12944 — Corrosion protection of steel structures by protective paint systems.
Source: Commission Regulation (EU) 2019/1783 — Ecodesign requirements for transformers (amending Regulation 548/2014).
Source: QDTB Engineering Team — dry-type and oil-immersed transformer supply to industrial, data-centre and electronics projects, 2015–2025 (demand factors, load densities, harmonic derating factors, price bands, lead times and outage-cost ranges are industry experience values, not published standards).