Averroes.ai

Achieve Zero Defect Escapes.
Every Single Time.

WatchDog™ Catches What Traditional Inspection Cannot

Your fab produces millions of wafers annually. One undetected defect batch costs millions in scrap and lost yield. WatchDog™ is the only AI inspection system that detects both known AND unknown defects—delivering absolute zero escapees through dual-layer cascading architecture. No blind spots. No surprises. No excuses.

91.5%
Anomaly Detection Rate
10%
False Positive Rate
100%
Real-time Processing
Hours
Deployment Time

The Critical Gap

Unknown defects from unexpected sources lead to defect escapes, compromising wafer yield and increasing downstream costs.

Equipment Changes

Tool modifications and component replacements introduce novel defect patterns that traditional supervised learning systems cannot detect.

Process Drift

Gradual parameter shifts over time create new defect modes that historical AI models have never encountered, leading to blind spots.

Electrical Variations

Power fluctuations and grounding issues can trigger defects that emerge unpredictably and vary from standard patterns.

Environmental Factors

Temperature, humidity, and vibration changes in fab environments introduce defect modes that require continuous monitoring.

Averroes WatchDog Solution

AI-powered solution that continuously monitors semiconductor fab processes to detect anomalies without requiring pre-labeled defect libraries

Minimal Data Requirements

Starts with few data points and learns from your fab environment. No extensive historical data needed.

Real-time Monitoring

Continuously analyzes wafer data to detect anomalies as they occur, preventing defect escapes.

Rapid Deployment

Deploy advanced AI inspection applications in hours, not months, with minimal data requirements.

Cascading Architecture

Unsupervised to Supervised Learning Pipeline

WatchDog runs alongside your existing supervised models to catch unknown defects that would otherwise escape detection. This creates a continuous learning loop that enhances both systems over time.

Production Data (Real-time wafer/sensor data)

1

Supervised Engine

Fast, precise classification and detection of known defect types

  • Low computational overhead
  • Detailed defect categorization
  • Production-ready speed
2

Unsupervised Engine

Anomaly detection flags unknown patterns and anomalies

  • Discovers unknown defects
  • No historical data required
  • Continuous monitoring
3

Human Validation

Expert review by semiconductor quality engineers

4

Label Generation

Data enrichment creates labeled dataset for new defect types

Enhanced Detection

Discover how WatchDog™ transforms defect detection with dual-layer architecture

Customer Case Study

Real-world results from leading semiconductor manufacturers

Subtle Process Variation Defects

The Challenge:

A leading photomask manufacturer began encountering unknown defect patterns sporadically across masks used in advanced node production. Traditional inspection systems failed to capture these anomalies, as they represented entirely new patterns absent from historical defect libraries.

Averroes WatchDog Solution:

We implemented a two-layer modeling approach combining supervised learning and our WatchDog anomaly detection system. The supervised model was trained on known defect classes, while WatchDog continuously monitored for any patterns outside of this known defect space, flagging anomalies that traditional inspection systems could not capture.

91.5%
Detection Rate
10%
False Positives
Zero
Defect Escapes

Key Benefits

Reduces Inspection Overhead

Automated anomaly detection reduces manual review time while maintaining high accuracy standards.

Minimizes Quality Complaints

Catch defects before they reach customers, protecting your reputation and reducing costly returns.

Improves OEE

Increases Overall Equipment Effectiveness by detecting issues early and preventing production losses.

Fast Deployment

Production-ready in hours with minimal data requirements. No lengthy training or data collection needed.

Unknown Defect Detection

Discovers defects that traditional systems miss, adapting automatically to new patterns and variations.

Continuous Learning

System improves over time through feedback loop, constantly enhancing detection capabilities.

Deployment & Integration

Flexible deployment options designed for semiconductor manufacturing environments

Real-time Processing

Immediate detection and alerting of anomalies as they occur in production.

API Integration

Seamless connection to existing MES/SCADA systems via standard APIs.

On-premise Deployment

Complete control over data and processing in your secure fab environment.

Edge Deployment

Real-time processing at fab production line for immediate defect detection.

Containerized Solutions

On-premise deployment for sensitive fab environments using Docker/Kubernetes. Maintain complete data security while leveraging advanced AI capabilities.

WatchDog™ Across Your Fab

From lithography to BEOL—comprehensive defect detection at every critical process step

Post-Lithography

  • Stochastic printing failures
  • Resist residue patterns
  • Micro-bridging detection
  • Reticle contamination

Post-Etch

  • Grass/fencing textures
  • Polymer redeposition
  • Chamber drift prediction
  • Profile anomalies

Post-CMP

  • Novel scratch patterns
  • Dishing/erosion trends
  • Slurry contamination
  • Pad wear monitoring

Post-Deposition

  • Film stress defects
  • Particle generation
  • Thickness non-uniformity
  • Delamination detection

BEOL Interconnect

  • Void detection
  • Via misalignment
  • Copper protrusions
  • Barrier breakthrough

WatchDog™ vs Traditional Solutions

CapabilityTraditional AOIWatchDog™
Detects unknown defect types
Real-time production speed
Requires 100+ defect examples
Adapts to process changes automatically
Model retraining timeWeeksHours
Zero defect escapees guarantee

Stop Million-Dollar Yield Losses Today

Join leading semiconductor manufacturers using WatchDog™ to achieve zero defect escapees and protect their bottom line.