Unintentional Electromagnetic Emission

Unintentional emissions originate from the non-ideal behavior of analog components—transistors, capacitors, and interconnects—that deviate from their designed specifications. During semiconductor fabrication, manufacturing process variation introduces atomic-scale differences in doping concentrations, oxide thickness, and channel dimensions. These variances create unique non-linear transfer functions in amplifiers, mixers, and oscillators, causing each device to radiate a subtly distinct electromagnetic signature even when executing identical digital operations. The emissions are deterministic, repeatable, and intrinsic to the physical hardware, making them impossible to clone or spoof without access to the exact same silicon die.

The radiated energy spans a broad frequency spectrum, often extending far beyond the intentional transmission band. Key components include:

• Harmonic emissions: Integer multiples of the clock frequency generated by non-linear switching
• Intermodulation products: Sum and difference frequencies created when multiple signals mix in non-linear junctions
• Broadband noise: Random thermal and shot noise shaped by the device's impedance profile
• Spurious tones: Unintended oscillations from parasitic resonances in the PCB layout and package parasitics

The spectral envelope and relative power distribution across these components form a highly discriminative fingerprint that can be captured using a spectrum analyzer or software-defined radio.

The unintentional emissions are not static; they are modulated by the instantaneous digital state of the device. As different logic blocks activate, the current draw through the power distribution network changes, altering the amplitude and phase of the radiated field. This creates a state-dependent emission profile where specific cryptographic operations, memory accesses, or instruction sequences produce identifiable modulation patterns. Advanced authentication systems exploit this by challenging the device with a known stimulus and verifying that the resulting emission pattern matches the expected golden reference signature for that specific hardware instance.

The measurement regime significantly impacts the captured signature:

• Near-field probing: Uses magnetic (H-field) or electric (E-field) probes placed millimeters from the device. Captures localized emissions from specific ICs or PCB traces with high signal-to-noise ratio. Ideal for in-situ verification of individual components on an assembled board.
• Far-field capture: Uses antennas at a distance exceeding one wavelength. Captures the composite radiated signature of the entire device, including antenna effects and enclosure resonances. More representative of operational scenarios but subject to multipath channel distortion. Both methods require careful calibration and channel-robust feature learning to ensure consistent authentication across varying measurement setups.

Unintentional emissions are sensitive to environmental and operational conditions, which must be compensated for in any practical authentication system:

• Temperature drift: Semiconductor carrier mobility and threshold voltages shift with temperature, altering amplifier gain and oscillator frequency. Temperature-drift compensation algorithms normalize features against a thermal model.
• Supply voltage variation: Fluctuations in the power rail directly modulate emission amplitude and phase noise.
• Aging effects: Hot carrier injection and electromigration cause slow, long-term changes in transistor characteristics. Drift compensation techniques track these gradual shifts to prevent false rejections.
• Electromagnetic interference: Ambient RF noise can mask weak emissions, requiring shielded measurement environments or differential measurement techniques.

In supply chain authentication, the unintentional emission profile serves as a physical unclonable function (PUF) that cannot be replicated by counterfeiters. The process works as follows:

• A golden reference signature is captured from a verified-authentic component at enrollment
• Incoming components are subjected to identical stimulus and measurement conditions
• Higher-order statistical features—including bispectrum, kurtosis, and cyclostationary moments—are extracted to quantify the emission's non-Gaussian characteristics
• A one-class classifier or anomaly detection model compares the candidate signature against the golden reference
• Components with out-of-family emissions are flagged as suspect for further forensic analysis This technique detects remarked, recycled, and cloned ICs that would pass traditional visual and electrical testing.