DAC Integral Non-Linearity

INL quantifies the maximum deviation of the real analog output from the ideal output, measured at each digital code after correcting for gain and offset errors. It is typically expressed in Least Significant Bits (LSBs) .

• Endpoint Method: The straight line is drawn between the first and last measured points.
• Best-Fit Method: A linear regression minimizes the maximum deviation, isolating the non-linear shape.
• A high-quality DAC might exhibit INL of ±0.5 LSB, while a low-cost unit could be ±4 LSB.

INL describes the smooth, low-order polynomial curvature of the DAC's response, distinct from random noise or quantization error. This curvature is a direct consequence of transistor mismatch in the current-steering cells or resistor ladders.

• S-Curve: A common INL profile caused by systematic gradients across the die.
• Bow: A parabolic shape resulting from finite output impedance in current sources.
• This deterministic shape is highly stable over time and temperature, making it an excellent physical-layer identifier.

While INL is the cumulative error, Differential Non-Linearity (DNL) is the step-to-step error between adjacent codes. INL is mathematically the running integral of DNL.

• A single large DNL spike (e.g., a missing code) creates a permanent offset in the INL curve from that point onward.
• A DAC with excellent DNL can still have poor INL if small errors accumulate systematically.
• Fingerprinting systems often use both metrics: DNL for local granularity, INL for the global structural signature.

The INL profile is heavily influenced by the DAC's internal architecture, which dictates how fabrication variances manifest.

• Binary-Weighted DACs: Exhibit major-carry transitions (e.g., 0111 to 1000) where many switches toggle simultaneously, causing large glitch-induced INL spikes.
• Thermometer-Coded DACs: Use unary-weighted elements, producing a smooth, monotonic INL curve with no major-carry discontinuities.
• Segmented DACs: Combine both architectures, creating a hybrid INL signature with distinct coarse and fine structure.

INL directly degrades the Error Vector Magnitude (EVM) of a transmitted signal by warping the ideal constellation points.

• In a 64-QAM signal, INL causes a non-linear displacement of symbols that varies with signal amplitude.
• This creates a unique, amplitude-dependent distortion pattern in the I/Q plane.
• Unlike random noise, this distortion is deterministic and can be extracted as a robust fingerprint even in moderate channel conditions.

The INL signature is primarily determined by static process variation during semiconductor fabrication, making it a permanent, unclonable hardware identifier.

• Temperature Drift: INL shape remains largely invariant, though the absolute gain and offset may shift slightly.
• Aging: The underlying transistor matching is stable over decades; INL-based fingerprints do not require frequent recalibration.
• This permanence makes INL a superior feature for supply chain authentication and long-lifespan IoT device identification.

The deviation in the step size between two adjacent digital codes from the ideal value of 1 Least Significant Bit (LSB). While INL measures the cumulative deviation from the ideal transfer function, DNL measures the local error at each code transition.

• Relationship to INL: INL is the running sum of DNL errors. A single large DNL spike directly contributes to a sharp change in the INL curve.
• Missing Codes: A DNL of -1 LSB indicates a code that is never output, creating a gap in the transfer function.
• Fingerprinting Value: The specific pattern of DNL errors across the code range is a highly unique, static signature of the DAC's internal resistor ladder or current-steering network mismatches.

The fundamental, irreducible difference between the original continuous analog value and its discrete digital representation, bounded by ±0.5 LSB for an ideal converter. This error is inherent to the digitization process itself.

• Contribution to INL: Quantization error is distinct from INL. INL describes the inaccuracy of the output levels themselves, while quantization error is the granularity limitation.
• Noise Floor: In an ideal DAC, quantization error manifests as white noise. INL, however, introduces signal-dependent harmonic distortion that rises above this noise floor.
• RF Fingerprinting: The statistical distribution of quantization error can be subtly shaped by the DAC's non-ideal step sizes, linking it back to the DNL pattern.

A critical DAC property guaranteeing that the analog output always increases (or stays the same) for an increasing digital input code. A non-monotonic DAC will exhibit a dip or reversal in its transfer function.

• INL Correlation: Non-monotonicity occurs when the INL error exceeds ±1 LSB, causing the transfer function to fold back on itself.
• System Impact: In control systems, non-monotonicity can cause instability. In communications, it creates severe, non-linear distortion that is easily identifiable as a hardware defect.
• Signature Stability: A non-monotonic glitch is a catastrophic, highly distinctive feature that makes a specific DAC unit trivially identifiable from its monotonic counterparts.

The ratio of the RMS amplitude of the desired output signal to the RMS amplitude of the largest single spurious tone (spur) in the output spectrum, excluding DC. INL is a primary generator of these spurs.

• INL-Driven Spurs: The specific pattern of INL errors directly determines the frequencies and amplitudes of harmonic and intermodulation spurs. A bowed INL curve creates dominant 2nd-order harmonics, while an S-shaped curve creates 3rd-order harmonics.
• Fingerprinting Metric: SFDR is a single-number aggregate of distortion. The specific spectral location of the worst spur, however, is a more granular, device-unique fingerprint.
• Degradation: Poor INL performance directly degrades SFDR, limiting the DAC's ability to generate clean signals for sensitive RF applications.

The deviation of the actual slope of the DAC's transfer function from the ideal slope, measured after correcting for offset error. It represents a linear scaling mismatch across the entire output range.

• Distinction from INL: Gain error is a linear, first-order error (a rotation of the transfer line). INL is the non-linear, residual deviation that remains after gain and offset errors are removed.
• Compensation: Gain error is trivially calibrated out in software or analog front-ends. INL, being non-linear, is much harder to correct, making it a more persistent and reliable fingerprint.
• End-Point vs. Best-Fit: The method used to define the "ideal line" for INL calculation (end-point vs. least-squares best-fit) determines whether gain error is included in or excluded from the INL measurement.

The analog output voltage when the digital input code is zero. It represents a constant vertical shift of the entire transfer function from the origin.

• Zero-Code Output: In an ideal DAC, code 0 produces 0V. Offset error causes a non-zero voltage, shifting the entire constellation in the I/Q plane.
• Origin Offset in RF: In direct-conversion transmitters, DAC offset error directly contributes to I/Q DC Offset, causing an unwanted carrier feedthrough spike at the center frequency.
• Fingerprinting Stability: Like gain error, offset error is a linear impairment. However, its specific magnitude is a stable, device-unique value that contributes to the overall hardware signature, particularly in the LO Leakage metric.