The software stack
Five Systems Run a Fab's Facilities, Usually From Five Vendors
Thousands of tools, hundreds of pumps, dozens of air handlers and UPW, gas, power and particle sensors all stream data at once, and the control systems must correlate events across them in milliseconds where yield is at stake.
| Layer | Function | Data | Platforms |
|---|---|---|---|
| Manufacturing execution system (MES) | Lot tracking, recipe execution, tool states, work in process, lot genealogy | Lots, recipes, tool states, inline metrology | Applied Materials SmartFactory; Siemens Opcenter Semiconductor (Camstar); Critical Manufacturing (ASMPT); in-house systems at TSMC and Samsung |
| Equipment automation | SECS/GEM and SEMI E30, E40, E87, E90 and E94 interfaces between tools and MES; E116 for equipment performance tracking | Tool state, alarms, recipe parameters, carrier tracking | Native interfaces from ASML, Lam, Applied Materials and Tokyo Electron; Cimetrics (PDF Solutions) connectivity |
| Building management system (BMS) | Cleanroom HVAC, chiller plant, fire, access and general electrical | Temperature, humidity, pressure, fan and damper states, substation status | Siemens Desigo CC; Honeywell; Schneider Electric EcoStruxure Building; Johnson Controls Metasys |
| Facility management system (FMS) | Semiconductor utilities in one view: UPW, gases, chemicals, vacuum, abatement | UPW resistivity and TOC, cylinder status, chemical concentration, pump current, foreline pressure | No dominant product; built on AVEVA PI System or Honeywell, Aspen and Inductive Automation tools |
| Computerized maintenance management (CMMS) | Preventive maintenance, work orders, spares and failure history for facility assets | Asset registry, maintenance records, MTBF and MTTR by asset | IBM Maximo; SAP Plant Maintenance; Hexagon EAM; Oracle |
| Energy management system (EMS) | Battery dispatch, microgrid control, demand response, load shedding | Power by zone, state of charge, tariffs, grid signals | Schneider Electric EcoStruxure; Siemens; custom systems tied to microgrid controls |
Why no DCIM for fabs
Data Centers Have DCIM; Fabs Have Point-to-Point Links
Data center infrastructure management (DCIM) puts power, cooling, assets and capacity in one product, and Schneider Electric EcoStruxure IT, Vertiv and Sunbird run it in thousands of data centers. Fabs have no equivalent:
- Incremental build-out: fabs grew over decades with tools from many makers, facility systems from many vendors and one-off links to ERP.
- Unrelated utilities: UPW, specialty gas, chemicals, vacuum and HVAC each have their own sensors, protocols and failure modes.
- Standards stop at the tool: SEMI defines equipment interfaces but not facility utility monitoring, so each operator built its own layer.
The result appears when UPW quality drifts: the FMS alarms, the CMMS opens a work order and the MES holds tools, as three separate events. Nobody is told which lots were exposed or how long the hold will last. Closing that gap is the case for integrated facility platforms and facility digital twins.
Digital twins
Process and Yield Twins Are Mature; Facility Twins Are Not
| Twin | Models | Use | Maturity | Platforms |
|---|---|---|---|---|
| Process | Etch profiles, deposition conformality, implant and CMP physics | Recipe development with fewer test wafers | Mature; in production at TSMC, Intel and Samsung | Applied Materials; Lam Research Semiverse; Synopsys Sentaurus TCAD; Silvaco |
| Tool | Chamber conditions, plasma, gas flow, stage motion | Drift and failure prediction; chamber matching; remote diagnostics | Emerging; ASML remote EUV diagnostics is the leading example | ASML; Lam; Applied Materials; Tokyo Electron; KLA |
| Yield | Process variables, tool states and materials against yield across hundreds of steps | Excursion root cause; lot disposition | Mature as yield management systems; machine learning is the frontier | KLA Klarity; PDF Solutions Exensio; Onto Innovation Discover; Synopsys Yield Explorer |
| Facility | UPW, HVAC, power, gas pressures and pump curves, matched to sensor data | What a failure does to production; energy optimization; capital planning | Nascent; no off-the-shelf product at leading-edge scale | NVIDIA Omniverse; Siemens Xcelerator; Synopsys (Ansys Twin Builder); AVEVA |
| Full fab | Process, yield and facility in one simulation | New fab design; technology transfer; multi-fab risk | Aspirational; no fab runs one at production scale | No single vendor |
TSMC and NVIDIA
TSMC Runs NVIDIA Libraries in Production and Is Exploring a Virtual Fab
At GTC Taipei in June 2026, NVIDIA and TSMC described what is in use: cuLitho for computational lithography (20% to 50% better cost or cycle time than CPU runs), cuEST for materials simulation (about 50 times faster on average), cuML for advanced process control across hundreds of thousands of parameters, GPU-accelerated fab scheduling on H200 systems, and Metropolis vision AI for nanometer-scale defect inspection. A facility twin built on Omniverse, called FabTwin, is still being explored for testing tool layouts before installation. More on AI in Fabs.
Predictive maintenance
Predictive Maintenance Is the Most Mature AI in Facility Management
A pump swapped in a planned four-hour window costs the same as an emergency replacement but avoids two to eight hours of unplanned tool downtime and the risk of scrapping lots in process.
| Application | Inputs | Predicts | Maturity |
|---|---|---|---|
| Vacuum pump health | Motor current, foreline pressure, body temperature, vibration spectrum | Failure 7-30 days ahead; deposit buildup | Deployed; Edwards and Ebara sell pump telemetry and service contracts |
| HVAC filters | HEPA and ULPA differential pressure; fan current; particle counts | Replacement before shutdown threshold; filter bypass | Differential pressure monitoring universal; remaining-life models being added |
| UPW components | RO pressure and salt rejection; EDI resistivity; UV lamp output; TOC | Membrane fouling, resin exhaustion, lamp end of life | Emerging; Veolia, Kurita and Organo offer monitoring services |
| EUV scanner health | Thousands of internal sensors: source power, optics, stages, vacuum | Drift before yield impact; module replacement | Deployed; ASML streams telemetry to its field service under the scanner contract |
| Power quality | Waveform capture at UPS outputs and panels; battery events | UPS battery fade; recurring grid-linked events | Emerging; correlating power events with yield is spreading |
Platform vendors
Platform Vendors
| Vendor | Platforms | Role |
|---|---|---|
| Applied Materials | SmartFactory MES and automation; process simulation | MES and process control; widely used outside the in-house systems at TSMC and Samsung |
| Siemens | Desigo CC; Opcenter; Xcelerator; Altair (acquired 2025) | The broadest facility portfolio: BMS, MES, simulation and digital twins |
| NVIDIA | Omniverse; CUDA-X libraries; Metropolis | GPU simulation and vision AI; TSMC deployments announced June 2026 |
| AVEVA (Schneider Electric) | PI System | The de facto data historian under most fab facility systems |
| IBM | Maximo Application Suite | CMMS widely used at large fab sites |
| KLA | Klarity and yield analytics | The yield twin; as hard to replace as KLA inspection hardware |
| Synopsys (including Ansys since July 2025) | Sentaurus TCAD; Yield Explorer; Fluent CFD; Twin Builder | Process and device simulation plus physics for facility twins |
Greenfield fabs
New Fabs Can Design the Data Layer In From the Start
US greenfield fabs at Arizona, Texas, Ohio, New York and Idaho can design sensors, data architecture and system integration for a unified facility platform before construction. Retrofitting the same capability onto a campus with 20 years of mixed systems is far harder. The same simulation platforms are spreading to gigafactories and AI data centers, so fabs, gigafactories and data centers are converging on shared tools.
Record
Page Record
- Covers
- Fab facility software layers, the missing fab DCIM, digital twin types, TSMC and NVIDIA deployments, predictive maintenance and platform vendors
- Sources
- NVIDIA and TSMC announcement, GTC Taipei, June 2026; SEMI E-series standards; Synopsys (Ansys acquisition closed July 2025); vendor product literature
- Last verified
- October 2026