Circuit Fingerprinting for Tamper Detection

ISEF Category: Embedded Systems

Subcategory: Circuits  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A tiny chip swap can change how a board behaves long before it fully fails. Your device can act like an electrical fingerprint scanner for a PCB. It sends in a signal, reads the board’s response, and looks for signs that a part does not belong. That makes this project part detective work, part signal processing.

What Is It?

Circuit fingerprinting means you test a circuit by sending in a known electrical signal and measuring how the board responds. Every board has small quirks from trace length, solder joints, component tolerances, and layout. Think of it like hearing a person speak. You know the words, but each voice still sounds a little different.

Reflectometry is one way to do that. In simple terms, you send a signal into a path and watch for reflections, like an echo. A clean board and a tampered board can produce different echoes because the electrical path has changed. A CNN, or convolutional neural network, is a machine learning model that can learn patterns in those response curves and sort normal boards from suspicious ones.

Why This Is a Good Topic

This is a strong science fair topic because you can test a real security problem with measurable signals. You can vary component type, solder changes, board age, or hidden modifications, then see whether your system still finds the difference. The project connects to supply-chain security, anti-counterfeit testing, and hardware trust. You can also learn useful skills in sensing, feature extraction, machine learning, and experimental design.

Research Questions

• How does pseudo-random current injection strength affect the separability of board fingerprints?
• What is the effect of counterfeit components on reflectometry signatures from a powered-off PCB?
• Does a CNN classify tampered and untampered boards more accurately than a simple threshold method?
• To what extent does component type change the stability of a board’s electrical fingerprint over repeated trials?
• Which frequency bands carry the most useful information for detecting board tampering?
• How does adding solder rework to one component alter the reflectometry response of the whole board?

Basic Materials

• Development board or custom test PCB with exposed test points.
• MCU or single-board computer with ADC input.
• Signal generator or DAC-capable MCU output.
• Oscilloscope or logic analyzer with analog capture.
• Breadboard, jumper wires, and probe clips.
• Resistors, capacitors, and common PCB components for test variants.
• Multimeter with continuity and resistance modes.
• Computer for data logging and model training.

Advanced Materials

• Custom PCB test fixtures with controlled trace geometry.
• High-speed oscilloscope or mixed-signal oscilloscope.
• Precision current source or waveform generator.
• High-resolution ADC or front-end conditioning board.
• Soldering and rework tools for controlled tamper scenarios.
• Environmental chamber or thermal control setup for stability testing.
• Reference counterfeit, aged, or reworked components.
• SEM, X-ray, or microinspection access for validation if available.

Software & Tools

• Python: Cleans waveform data, builds features, and trains classifiers.
• TensorFlow Lite: Runs a compact CNN on a microcontroller or embedded target.
• scikit-learn: Compares the CNN against simpler baseline models.
• ImageJ: Measures inspection images if you pair electrical data with board photos.
• Jupyter Notebook: Organizes analysis, plots signals, and records model tests.

Experiment Steps

1. Define the tamper types you will test, such as component swap, solder rework, or hidden damage.
2. Choose one measurement path on the PCB and map where you can inject and read the signal safely.
3. Plan a baseline dataset from unmodified boards so you can compare every later fingerprint against a normal reference.
4. Design a feature pipeline that turns raw reflectometry traces into inputs a classifier can learn from.
5. Build comparison models, starting with a simple baseline and then a CNN, so you can test whether the extra complexity helps.
6. Plan validation with held-out boards, not just repeated scans of the same board, so your results reflect real detection performance.

Common Pitfalls

• Training and testing on scans from the same physical board, which makes the model memorize board-specific noise instead of tamper features.
• Changing probe pressure or contact position between trials, which can alter the trace more than the tamper does.
• Mixing powered and unpowered measurements, which creates signals that are not comparable.
• Using only one kind of tamper, which makes the classifier fail on new counterfeit or rework patterns.
• Skipping baseline calibration, which leaves you unable to tell whether a signal change came from the board or the measurement setup.

What Makes This Competitive

A stronger version of this project does more than say, “the model worked.” It tests whether your method generalizes across board layouts, component families, and tamper styles. Good entries compare multiple signal representations, use clean train-test splits by board, and report false positives as well as accuracy. If you can show which parts of the fingerprint carry the most information, you add real engineering insight.

Project Variations

• Test whether the same fingerprint method works on consumer boards from different device classes, such as chargers, routers, or toys.
• Replace the CNN with a smaller embedded classifier and compare detection speed, memory use, and accuracy.
• Add thermal aging or moisture exposure as a second stressor and see whether tamper detection still works after board wear.

Learn More

• NIH PubMed: Search for review articles on counterfeit electronics, hardware security, and PCB fault detection.
• NOAA Science On a Sphere? No, use NOAA and NASA data only if you expand toward environmental sensing, not needed for this topic.
• NASA NTRS: Search the Technical Reports Server for papers on reflectometry, signal processing, and embedded classification.
• MIT OpenCourseWare: Find free courses on digital signal processing, machine learning, and embedded systems.
• IEEE Xplore Abstracts: Read abstracts and open-access papers on circuit fingerprinting, reflectometry, and hardware Trojan detection.
• USGS Publications Warehouse: Useful if you later compare this project’s sensing methods to other inspection and anomaly-detection systems.