Outages that tested fabs
Fires, Storms, Contamination and Malware Have Cost More Downtime Than Earthquakes
Engineering targets are written in nines, but the record is written in events: the longest documented recent outages came from a fire, a winter storm and a contaminated material, and the costliest from a storm and an earthquake.
| Event | Cause | What stopped | Recovery and cost |
|---|---|---|---|
| TSMC, August 2018 | A WannaCry variant spread when a supplier's tool software was installed without a virus scan | Tools at fabs in Hsinchu, Taichung and Tainan | Most tools back within about three days; TSMC estimated about 3% of third-quarter revenue |
| Samsung Austin, February 2021 | Winter Storm Uri; ERCOT ordered the fab off the grid | The whole Austin fab, with wafers in process | Full output by about the end of March; loss reported at about $270 million |
| Renesas Naka, March 2021 | Electrical fire in a plating tool on the 300mm line | Automotive MCU production | Production restarted within a month and returned to pre-fire output in about three months |
| Kioxia and Western Digital, February 2022 | Contaminated process material at the Yokkaichi and Kitakami NAND fabs | BiCS flash output | Western Digital reported at least 6.5 exabytes of flash lost; output restored by early March |
| TSMC, April 2024 | Hualien earthquake, M7.4 | Fabs in Hsinchu, Taichung and Tainan | Most tools back within hours, facilities within three days; about $92 million |
| TSMC, January 2025 | Chiayi earthquake, M6.4 | Fab 14 and Fab 18 in Tainan | Up to 20,000 wafers in process affected; about $160 million |
Uptime targets
Uptime Targets by System
The industry has no single published uptime standard. Fabs set targets per system, separate planned from unplanned downtime, and size redundancy to a recovery time objective for each utility.
| System | Target | Recovery objective | If exceeded |
|---|---|---|---|
| Power to process tools | 99.999%; under 5.3 minutes unplanned a year | Zero-transfer UPS; generators in under 30 seconds; full fab two to four hours | Lot scrap, tool faults, requalification |
| Ultrapure water | 99.99%; under 53 minutes a year | Production hold in minutes; restore in 2-8 hours | Cleaning, CMP and wet etch stop at once |
| Cleanroom HVAC | 99.99% | Minimum airflow in seconds; full air changes in 30 minutes | A classification breach can take 24-72 hours to requalify |
| Critical gases | 99.99%+ per gas; zero at cylinder changeover | Restore in 15-60 minutes | Mid-recipe aborts |
| Vacuum pumps | 99.5%+ per pump; fleet above 99.9% | Pump swap and recovery in 2-8 hours | One chamber down per failure |
| Process tools | 85-95% overall equipment effectiveness (OEE); EUV 90%+ | Mean time to repair under four hours | Wafer start shortfall |
| MES and process control | 99.999% | Hot standby in under 30 seconds | Lot traceability gaps |
Resilience KPIs
Resilience KPIs
| KPI | Definition | Target |
|---|---|---|
| Power events | Interruptions to tool loads longer than 10 ms | Zero at the tool; fewer than two a year at facility distribution |
| UPW uptime | Time UPW meets spec at tools; quality excursions count as downtime | Above 99.99% |
| Mean time between failures | Operating hours between unplanned tool failures | Plasma etch 500-2,000 hours; furnaces 2,000-5,000 |
| Mean time to repair | Fault to production-ready, including requalification | Under four hours for common failures |
| OEE | Availability times performance rate times quality rate | 85-95% fleet-wide; below 80% triggers review |
| Supply continuity | Days of critical chemicals and gases on hand | 30-90 days of strategic stock |
Risk categories
Risk Categories and the Architecture Against Each
| Risk | Mechanism | Architecture | What remains |
|---|---|---|---|
| Grid failure | Outages, sags, frequency excursions | Zero-transfer UPS, dual feeds, battery storage, N+1 generation, microgrids | Multi-day outages beyond fuel stock |
| Water | Municipal interruption; drought | One to five days of raw water storage; N+1 UPW trains; 80-90% recycling; trucking contracts | Multi-week drought at water-stressed sites; Taiwan's 2021 drought had TSMC trucking water |
| Earthquakes | Ground motion, piping rupture, tool misalignment | Isolation layers; automatic shutdown; early warning | An M7.0 under Hsinchu; see Seismic and Vibration |
| Extreme weather | Winter storms, typhoons, heat | Winterization; flood-safe siting; cooling margin | Compound heat, grid and water events |
| Materials supply | Single-source gases, resists, slurries, spares; helium | Safety stock; dual qualification; long-term agreements | 12-18 month requalification; Qatar helium halted in 2026 |
| Cyber | Ransomware on MES or ERP; infected tool software | Segmented OT networks; verified tool updates; offline backups | Supplier software and insiders |
| Geopolitics | Export controls; a Taiwan blockade or conflict | Capacity in more countries; multi-jurisdiction sourcing | No engineering answer to a Taiwan Strait crisis |
Redundancy
N+1 Is the Floor; Power, UPW and HVAC Go Further
Every critical system survives the loss of any one component without a production hit, and the highest-criticality systems run dual feeds or N+2.
| System | Standard | Covers | Switchover |
|---|---|---|---|
| Utility feeds | Dual feeds on independent transmission paths | Substation or line outage | Automatic transfer in under 100 ms, bridged by UPS |
| UPS | N+1 modules; A and B feeds to critical tools | A module or feed failure | Instant A/B transfer; static bypass for maintenance |
| Generation | N+1 sets sized for critical loads | Outages past UPS and battery ride-through | Start in 10-30 seconds; 72-168 hours of fuel |
| UPW train | N+1 on RO, EDI, UV and final filters | Any stage failure | Automatic stream isolation and diversion |
| Cleanroom HVAC | N+1 recirculation air handlers and chillers; overlapping fan filter units | A unit or fan failure | Automatic load transfer without a production hold |
| Gas supply | A/B cylinder banks | Cylinder depletion | Pressure switch changes banks in under a second |
| Vacuum | N+1 spare pumps per bay; predictive replacement | Pump failure | Swap and qualify in under four to eight hours |
CHIPS Act conditions
CHIPS Act Awards Tied Subsidy to Resilience Conditions
CHIPS and Science Act awards carried conditions that reach into resilience: limits on expanding leading-edge capacity in China and other countries of concern for 10 years (the guardrail rule, with clawback of the full award), supply chain reporting, workforce and childcare plans for large awards, and cybersecurity requirements. The Commerce Department has since renegotiated several awards, including converting Intel's grants into a roughly 10% federal equity stake in August 2025. Program detail is on CHIPS and Science Act.
Record
Page Record
- Covers
- Documented outages, uptime targets, resilience KPIs, risk categories, redundancy and CHIPS Act conditions
- Sources
- TSMC 2018 and 2024 statements and annual reports; Samsung and Korea Times reporting, April 2021; Renesas Naka fire updates, 2021; Western Digital 10-Q, 2022; TrendForce on the January 2025 earthquake; Commerce Department
- Last verified
- October 2026