FDC System Explained: Fault Detection & Classification
Averroes
May 18, 2026
Modern fabs generate more sensor data in an hour than most engineering teams could manually review in a year.
Temperature, pressure, gas flow, electrical signal.
Every tool on the line is broadcasting state data in real time, and the fab’s ability to act on it before scrap appears separates a healthy yield curve from a costly one.
That’s the job of an FDC system.
We’ll cover what an FDC system does, how it differs from SPC and APC, and how to implement one without the usual pitfalls.
Key Notes
An FDC system continuously monitors equipment and process parameters to detect, classify, and respond to anomalies in real time.
FDC complements SPC (which tracks process drift) and APC (which adjusts process parameters) – the three work together, not in isolation.
Implementation success depends more on integration with MES, CIM, and tool management systems than on the FDC platform itself.
What Is An FDC System?
An FDC system (short for Fault Detection and Classification) is the monitoring layer that watches your equipment and process data in real time.
It flags anomalies, identifies probable causes, and triggers responses fast enough to prevent defective wafers from being produced.
The 3 Jobs Of An FDC System
Every modern FDC system runs the same three functions on every critical tool in the fab:
Detect (spot deviations from expected performance in real time)
Classify (identify the probable root cause behind the anomaly)
Respond (trigger the appropriate corrective action automatically)
Strip any one of those out and you’ve got a partial system (SPC alone can detect but not classify and a classification engine alone can’t act).
Integration is what makes FDC work as a closed loop.
The Core Principle Behind FDC
Equipment health directly drives product quality.
A tool drifting out of spec doesn’t always announce itself – it gradually produces more marginal output until a yield review catches up to it weeks later.
An FDC system closes that gap in two ways:
Treats equipment state as a leading indicator of yield, not a lagging one
Acts on data continuously rather than waiting for periodic review cycles
How Do FDC Systems Work?
FDC systems run a continuous four-step loop: collect data, detect anomalies, classify them, and trigger a response.
Each step uses different technology, but they only work as an integrated chain.
Step 1: Data Acquisition
The foundation of any FDC system is high-quality real-time data collection from strategically placed sensors across your manufacturing equipment.
Sensors Typically Monitored:
Temperature sensors: Equipment and material temperatures (crucial in wafer fabrication).
Pressure sensors: Gas pressures in etching, deposition, and CVD processes.
Electrical sensors: Current, voltage, and power signals indicating tool functionality.
Optical sensors: Imaging for surface condition monitoring and inline defect detection.
Data Acquisition Typically Runs Through Three Methods:
Direct measurement: Sensors transmit readings straight to the FDC platform.
SECS/GEM communication standards: Standardized tool-to-FDC data exchange.
Continuous data logging: Historical archives for trend analysis and root-cause investigation.
Step 2: Fault Detection
Sophisticated algorithms analyze the incoming data stream in real time, incorporating Statistical Process Control (SPC) techniques and increasingly machine learning models.
What Modern Detection Looks Like:
Real-time monitoring: Deviations flagged as they occur, not in next morning’s report.
Subtle pattern detection: ML models flag drifts that look normal on any individual chart but combine into a pre-failure signature.
Step 3: Fault Classification
Once an anomaly is detected, a classification engine determines the probable cause.
This is where machine learning earns its place – pattern recognition across historical fault data lets the system improve accuracy over time.
Standard Classification Categories:
Tool malfunction: Equipment failure or component degradation.
Process drift: Gradual parameter changes pushing the process out of its window.
Environmental factors: External influences like temperature swings or contamination events.
Step 4: Corrective Action
Based on the classification, the system triggers the appropriate response (anywhere from a soft alert to operators, to automatic parameter adjustment, to emergency tool shutdown for critical faults).
Effective Corrective Action Depends On Integration With Broader Manufacturing Systems:
MES for workflow coordination
CIM for control strategy execution
Tool management for direct parameter adjustment
Without those integrations, the FDC system flags problems but can’t fix them.
FDC vs. SPC vs. APC: How They Work Together
FDC, SPC, and APC overlap in places but they do different jobs (and modern fabs run all three together).
Statistical Process Control (SPC)
SPC monitors process parameters with control charts to detect drift before it produces defects.
It’s the foundational layer: track the parameter, set upper and lower control limits, intervene when the process drifts toward those limits.
What it does: Monitors and alerts on parameter drift
What it doesn’t do: Classify root causes or trigger corrective actions
Where it sits: The diagnostic baseline that everything else builds on
Fault Detection and Classification (FDC)
FDC sits on top of SPC.
It uses SPC techniques for detection but adds two things SPC alone doesn’t have: classification of probable causes and automated response triggering.
It’s also better at multivariate analysis – SPC charts watch one parameter at a time, while FDC can watch dozens simultaneously.
What it does: Detects, classifies, and responds to faults in real time
What it doesn’t do: Actively adjust process parameters mid-run
Where it sits: The monitoring and response layer
Advanced Process Control (APC)
APC is the active controller.
It uses data from SPC and FDC to make real-time adjustments to process parameters – feed-forward control where upstream measurements inform downstream settings, run-to-run control where each batch tunes the next.
What it does: Actively adjusts process parameters to keep output in spec
What it doesn’t do: Detect or classify faults on its own
Where it sits: The control execution layer
How They Combine in Practice
A wafer moves through a deposition tool:
SPC charts track film thickness across the wafer in real time.
FDC flags when the readings drift outside expected variance and classifies the cause as gas flow drift.
APC adjusts the gas flow setpoint on the next wafer to bring the film thickness back into target.
Run any one of these without the others and you get partial coverage.
Run all three together and you get a closed-loop system where drift is detected, classified, and corrected before scrap accumulates.
Integration With Manufacturing Systems
FDC systems aren’t standalone.
Their value comes from integration with the broader manufacturing stack – without those connections, FDC is just a more sophisticated alarm.
The five integration points that matter:
Critical Applications of FDC in Semiconductor Manufacturing
FDC systems act as the primary monitoring and control mechanism across multiple stages of semiconductor manufacturing.
Four applications dominate:
Wafer Fabrication Monitoring
Wafer fabrication runs hundreds of process steps with tight tolerances on temperature, pressure, and chemical concentrations.
FDC monitors these parameters across every stage where drift can produce scrap.
Where FDC Adds The Most Value In Wafer Fab:
Thermal processes: Oxidation, annealing, and diffusion where temperature drift directly hits transistor performance.
Deposition: Film thickness and composition monitoring in CVD, PVD, and ALD steps.
Etching: Gas chemistry, plasma stability, and end-point detection.
Wet processing: Chemical concentration and contamination control.
When a parameter drifts in any of these stages, FDC flags the deviation in real time and triggers corrective action before the defect propagates downstream.
Equipment Health Tracking
FDC analyzes performance and operational data to identify signs of wear or malfunction before tools fail. This shifts maintenance from reactive to predictive – a meaningful operational and financial difference.
The Three Benefits That Show Up Consistently:
Predictive maintenance: Tools get serviced before they fail.
Reduced unplanned downtime: Production schedules stay on track.
Extended equipment life: Capital assets generate more output per dollar invested.
Process Drift Detection
Process drift is the silent killer of yield.
Parameter changes look fine in any single shift’s data but produce cumulative quality degradation over weeks.
How FDC Catches Drift Before It Costs You Yield:
Continuous parameter comparison: Process variables tracked against established control limits in real time.
Multivariate trend detection: Drift patterns visible across multiple parameters that single-variable charts miss.
Threshold-based alerts: Corrective action triggered as soon as drift exceeds tolerance, not at the next shift handoff.
Quality Control Integration
FDC feeds real-time equipment and process data into quality management systems, turning monitoring data into closed-loop quality improvement.
The Integration Enables 3 Things Downstream Quality Teams Couldn’t Do Before:
Automated defect reporting: No manual data transcription between systems.
Faster decisions: Quality issues surface with their root-cause context already attached.
Continuous process improvement: Root-cause analysis feeds back into process windows and control limits.
FDC Vendors: How The Major Platforms Compare
Several established vendors offer FDC systems for semiconductor manufacturing. They differ meaningfully on depth of analytics, integration breadth, and AI capability.
Resource allocation: Budget, personnel, and technology requirements (accurate scoping prevents scope creep later).
Stakeholder engagement: Engineering, operations, quality, and management buy-in before kickoff.
Phase 2: System Selection
Match the platform to your operating reality:
Scalability: Can it grow with you?
Integration capabilities: Clean MES, CIM, and tool management connections.
Vendor support: Training, maintenance, and troubleshooting depth.
Feature set: Real-time monitoring, predictive analytics, advanced reporting as needed.
Phase 3: Integration
Architecture design: Map data flows, communication protocols (SECS/GEM), and integration points before installation.
Installation: Hardware and software deployment with vendor coordination.
Data integration: Connect to existing databases and analytics platforms so data aggregates cleanly.
Phase 4: Testing & Validation
System testing: Verify all components function as expected.
Validation: Confirm fault detection, classification, and response work against specified scenarios.
User acceptance testing (UAT): End-user feedback before full deployment.
FDC System FAQs
How often should FDC system algorithms be retrained?
FDC system algorithms typically need retraining every 3–6 months, or whenever significant process changes occur. More frequent updates are required during new product introductions, major equipment modifications, or after process recipe changes. The retraining frequency should balance detection accuracy against system stability (too frequent and you introduce noise / too infrequent and accuracy degrades).
Can FDC systems work on legacy semiconductor equipment?
FDC systems can work on legacy semiconductor equipment through retrofitted sensors, external data acquisition modules, and protocol converters. Older tools often lack built-in SECS/GEM communication, but modern FDC platforms support adapter layers that bridge the gap. The retrofit cost is almost always less than replacing the tool itself.
What team is needed to run an FDC system?
A standard FDC team includes a process engineer, a data scientist or analytics specialist, and an equipment specialist. Smaller fabs often cross-train existing staff rather than hire dedicated roles. Modern no-code FDC platforms reduce the data science requirement significantly – process engineers can build and tune models without a separate ML team.
What’s the difference between FDC and predictive maintenance?
FDC and predictive maintenance overlap but solve different problems. FDC monitors process and equipment data in real time to catch faults as they occur. Predictive maintenance uses historical patterns to forecast when equipment will fail before any anomaly appears. Most modern FDC platforms include predictive maintenance capabilities, but the two are distinct functions worth understanding separately.
Conclusion
An FDC system earns its place by closing the gap between sensor data and corrective action – detect the deviation, classify the cause, trigger the response, all in real time.
Pair it with SPC for parameter monitoring and APC for active control, and you’ve got a closed-loop process control stack that catches drift, classifies it, and corrects it before scrap accumulates.
The platforms that win the next decade combine AI classification with integration breadth, because the FDC platform itself matters less than how cleanly it talks to MES, CIM, and your existing tool management infrastructure.
Book a free demo to see how AI-powered FDC plugs into the fab stack you already run.
Modern fabs generate more sensor data in an hour than most engineering teams could manually review in a year.
Temperature, pressure, gas flow, electrical signal.
Every tool on the line is broadcasting state data in real time, and the fab’s ability to act on it before scrap appears separates a healthy yield curve from a costly one.
That’s the job of an FDC system.
We’ll cover what an FDC system does, how it differs from SPC and APC, and how to implement one without the usual pitfalls.
Key Notes
What Is An FDC System?
An FDC system (short for Fault Detection and Classification) is the monitoring layer that watches your equipment and process data in real time.
It flags anomalies, identifies probable causes, and triggers responses fast enough to prevent defective wafers from being produced.
The 3 Jobs Of An FDC System
Every modern FDC system runs the same three functions on every critical tool in the fab:
Strip any one of those out and you’ve got a partial system (SPC alone can detect but not classify and a classification engine alone can’t act).
Integration is what makes FDC work as a closed loop.
The Core Principle Behind FDC
Equipment health directly drives product quality.
A tool drifting out of spec doesn’t always announce itself – it gradually produces more marginal output until a yield review catches up to it weeks later.
An FDC system closes that gap in two ways:
How Do FDC Systems Work?
FDC systems run a continuous four-step loop: collect data, detect anomalies, classify them, and trigger a response.
Each step uses different technology, but they only work as an integrated chain.
Step 1: Data Acquisition
The foundation of any FDC system is high-quality real-time data collection from strategically placed sensors across your manufacturing equipment.
Sensors Typically Monitored:
Data Acquisition Typically Runs Through Three Methods:
Step 2: Fault Detection
Sophisticated algorithms analyze the incoming data stream in real time, incorporating Statistical Process Control (SPC) techniques and increasingly machine learning models.
What Modern Detection Looks Like:
Step 3: Fault Classification
Once an anomaly is detected, a classification engine determines the probable cause.
This is where machine learning earns its place – pattern recognition across historical fault data lets the system improve accuracy over time.
Standard Classification Categories:
Step 4: Corrective Action
Based on the classification, the system triggers the appropriate response (anywhere from a soft alert to operators, to automatic parameter adjustment, to emergency tool shutdown for critical faults).
Effective Corrective Action Depends On Integration With Broader Manufacturing Systems:
Without those integrations, the FDC system flags problems but can’t fix them.
FDC vs. SPC vs. APC: How They Work Together
FDC, SPC, and APC overlap in places but they do different jobs (and modern fabs run all three together).
Statistical Process Control (SPC)
SPC monitors process parameters with control charts to detect drift before it produces defects.
It’s the foundational layer: track the parameter, set upper and lower control limits, intervene when the process drifts toward those limits.
Fault Detection and Classification (FDC)
FDC sits on top of SPC.
It uses SPC techniques for detection but adds two things SPC alone doesn’t have: classification of probable causes and automated response triggering.
It’s also better at multivariate analysis – SPC charts watch one parameter at a time, while FDC can watch dozens simultaneously.
Advanced Process Control (APC)
APC is the active controller.
It uses data from SPC and FDC to make real-time adjustments to process parameters – feed-forward control where upstream measurements inform downstream settings, run-to-run control where each batch tunes the next.
How They Combine in Practice
A wafer moves through a deposition tool:
Run any one of these without the others and you get partial coverage.
Run all three together and you get a closed-loop system where drift is detected, classified, and corrected before scrap accumulates.
Integration With Manufacturing Systems
FDC systems aren’t standalone.
Their value comes from integration with the broader manufacturing stack – without those connections, FDC is just a more sophisticated alarm.
The five integration points that matter:
Critical Applications of FDC in Semiconductor Manufacturing
FDC systems act as the primary monitoring and control mechanism across multiple stages of semiconductor manufacturing.
Four applications dominate:
Wafer Fabrication Monitoring
Wafer fabrication runs hundreds of process steps with tight tolerances on temperature, pressure, and chemical concentrations.
FDC monitors these parameters across every stage where drift can produce scrap.
Where FDC Adds The Most Value In Wafer Fab:
When a parameter drifts in any of these stages, FDC flags the deviation in real time and triggers corrective action before the defect propagates downstream.
Equipment Health Tracking
FDC analyzes performance and operational data to identify signs of wear or malfunction before tools fail. This shifts maintenance from reactive to predictive – a meaningful operational and financial difference.
The Three Benefits That Show Up Consistently:
Process Drift Detection
Process drift is the silent killer of yield.
Parameter changes look fine in any single shift’s data but produce cumulative quality degradation over weeks.
How FDC Catches Drift Before It Costs You Yield:
Quality Control Integration
FDC feeds real-time equipment and process data into quality management systems, turning monitoring data into closed-loop quality improvement.
The Integration Enables 3 Things Downstream Quality Teams Couldn’t Do Before:
FDC Vendors: How The Major Platforms Compare
Several established vendors offer FDC systems for semiconductor manufacturing. They differ meaningfully on depth of analytics, integration breadth, and AI capability.
What To Evaluate When Comparing?
Four criteria separate the strongest vendors from the rest:
Scalability: Can the platform grow with new tools, new process steps, and new product lines without rebuilds?
Want FDC That Learns From Your Data?
AI hits 99% accuracy where static rules miss.
Implementing An FDC System: A Practical Guide
Successful FDC implementation requires careful planning across four phases.
Phase 1: Planning
Phase 2: System Selection
Match the platform to your operating reality:
Phase 3: Integration
Phase 4: Testing & Validation
FDC System FAQs
How often should FDC system algorithms be retrained?
FDC system algorithms typically need retraining every 3–6 months, or whenever significant process changes occur. More frequent updates are required during new product introductions, major equipment modifications, or after process recipe changes. The retraining frequency should balance detection accuracy against system stability (too frequent and you introduce noise / too infrequent and accuracy degrades).
Can FDC systems work on legacy semiconductor equipment?
FDC systems can work on legacy semiconductor equipment through retrofitted sensors, external data acquisition modules, and protocol converters. Older tools often lack built-in SECS/GEM communication, but modern FDC platforms support adapter layers that bridge the gap. The retrofit cost is almost always less than replacing the tool itself.
What team is needed to run an FDC system?
A standard FDC team includes a process engineer, a data scientist or analytics specialist, and an equipment specialist. Smaller fabs often cross-train existing staff rather than hire dedicated roles. Modern no-code FDC platforms reduce the data science requirement significantly – process engineers can build and tune models without a separate ML team.
What’s the difference between FDC and predictive maintenance?
FDC and predictive maintenance overlap but solve different problems. FDC monitors process and equipment data in real time to catch faults as they occur. Predictive maintenance uses historical patterns to forecast when equipment will fail before any anomaly appears. Most modern FDC platforms include predictive maintenance capabilities, but the two are distinct functions worth understanding separately.
Conclusion
An FDC system earns its place by closing the gap between sensor data and corrective action – detect the deviation, classify the cause, trigger the response, all in real time.
Pair it with SPC for parameter monitoring and APC for active control, and you’ve got a closed-loop process control stack that catches drift, classifies it, and corrects it before scrap accumulates.
The platforms that win the next decade combine AI classification with integration breadth, because the FDC platform itself matters less than how cleanly it talks to MES, CIM, and your existing tool management infrastructure.
Book a free demo to see how AI-powered FDC plugs into the fab stack you already run.