‍

Introduction

In modern ASIC and FPGA verification, the hardest bugs to find are those triggered by rare data combinations or unusual event sequences—patterns that appear only once in millions of transactions. Traditional debug approaches require manual log inspection, making root cause analysis time-consuming and error-prone. Cogita-PRO addresses this challenge with a comprehensive suite of AI-powered anomaly detection algorithms specifically tailored for verification data.

‍

Understanding Verification Anomalies

Verification failures typically fall into two categories:

1. Data Anomalies: A unique combination of transaction field values that has never occurred before. The key question: What makes this transaction different from all others in the test?

2. Sequence Pattern Anomalies: Errors caused not by unusual data, but by a unique ordering of events that deviates from expected protocol flows. The key question: What sequence of interactions led to the failure?

Cogita-PRO provides specialized detection methods for both anomaly types, enabling automatic identification of root causes without manual scripting or exhaustive log analysis.

‍

Data Anomaly Detection

Cogita-PRO employs four complementary methods for detecting unusual data patterns:

Machine Learning Approaches (Large-Scale Data)

Neural Networks train on passing tests to learn complex, non-linear feature interactions, then identify anomalous transactions in failing tests. Ensemble Methods combine multiple algorithms to detect rare multi-dimensional field combinations through consensus voting, reducing false positives while highlighting specific problematic field combinations.

Key Benefits:

• Scales to high-dimensional data (15+ fields, millions of transactions)
• Generalizes across test scenarios without manual feature engineering
• Identifies complex interactions that single methods miss

Use Cases: SoC-level integration, interconnect fabrics, protocols where bugs manifest as specific combinations of addresses, transaction types, and state values.

Statistical and Explainability Methods 

Statistical Outlier Detection provides fast, training-free analysis to identify extreme values, timing violations, and frequency anomalies—ideal for interactive debug sessions and timing-related bugs.

Describe Analysis serves as the critical explainability layer, translating anomaly scores into actionable insights by identifying which specific fields deviate from baseline and quantifying how suspect transactions differ from normal behavior.

Key Benefits:

• No training required—works on single failing tests
• Eliminates manual comparison of thousands of field values
• Provides root cause attribution and feature importance ranking
• Explains anomalies detected by any upstream method

‍

Sequence Pattern Anomaly Detection

Cogita-PRO provides three methods for detecting temporal and ordering violations:

Transaction Path Analysis

Automatically extracts all routing paths from logs, clusters them into unique types, and compares against golden reference models from passing tests. Ideal for NoC fabrics and multi-path subsystems with high routing diversity.

Workflow Example:

1. Passing tests → Extract/validate paths → Save golden model
2. Failing test → Extract paths → Compare to golden
3. Identify novel anomalous path → Debug focused investigation

‍
Discrete State Machine Analysis

Two complementary methods validate protocol behavior:

Transition Extraction builds complete state graphs from logs, detecting illegal transitions and tracking multi-FSM correlations—perfect for cache coherency (MESI/MOESI) and flow control validation.

Sequence Extraction analyzes temporal event ordering across multiple fields, detecting illegal sequences and revealing unexpected cross-FSM dependencies—essential for deadlock debugging and causality violations.

Combined Benefits:

• Automatic discovery—no manual specification required
• Differential analysis highlights deviations causing failures
• Multi-dimensional comparison across concurrent state machines
• Coverage metrics quantify protocol exercise completeness

Concrete Results:

• Data anomalies: Neural networks and ensembles catch rare multi-field combinations, while describe analysis explains exactly why transactions were flagged
• Sequence anomalies: Path extraction validates routing correctness, FSM analysis catches protocol violations automatically
• Debug acceleration: From log file to root cause in minutes instead of hours

‍

Conclusion

Verification anomalies represent the hardest class of bugs because they occur exactly once—unique combinations that standard checkers and assertions cannot anticipate. Cogita-PRO's integrated approach combines:

1. Detection diversity - Multiple algorithms (ML, statistical, sequence-based) ensure comprehensive coverage
2. Automatic learning - Golden models extracted from passing tests without manual specification
3. Explainability - Every anomaly comes with root cause attribution, not just a score
4. Scale flexibility - Works from small block-level tests to full SoC regressions

By automating both the detection and explanation of data and sequence anomalies, Cogita-PRO enables verification teams to focus on fixing bugs rather than hunting for them in massive log files—accelerating time-to-tapeout while improving coverage confidence.

Ready to transform your debug workflow? Contact us to see how Cogita-PRO integrates into your verification environment and reduces debug time by orders of magnitude.