High-Speed ADC Architectures
1. Key Performance Metrics in High-Speed ADCs
Key Performance Metrics in High-Speed ADCs
Signal-to-Noise Ratio (SNR)
The signal-to-noise ratio (SNR) quantifies the ratio of the desired signal power to the noise power within the ADC's bandwidth. For an ideal N-bit ADC, the theoretical maximum SNR is given by:
In practice, thermal noise, quantization noise, and jitter reduce the achievable SNR. High-speed ADCs often operate close to their thermal noise floor, making SNR a critical metric for dynamic range.
Effective Number of Bits (ENOB)
ENOB represents the ADC's actual resolution when accounting for all noise and distortion sources. It relates directly to SNR:
For example, a 12-bit ADC with 65 dB SNR has an ENOB of 10.5 bits. This metric is particularly important in communications systems where dynamic range directly impacts bit error rates.
Spurious-Free Dynamic Range (SFDR)
SFDR measures the difference between the fundamental signal amplitude and the largest spurious tone in the frequency domain, regardless of whether it's harmonic or non-harmonic. It's expressed as:
In radar and spectrum analysis applications, SFDR determines the system's ability to detect weak signals in the presence of strong interferers.
Total Harmonic Distortion (THD)
THD quantifies the power sum of harmonic distortion components relative to the fundamental:
Where M is typically 5-10 harmonics. In audio and precision measurement systems, THD below -80 dBc is often required.
Noise Power Ratio (NPR)
NPR evaluates ADC performance in multi-carrier systems by measuring noise and intermodulation distortion in a notched bandwidth:
This metric is critical for cable modem and cellular base station designs where multiple channels must coexist without interference.
Jitter Sensitivity
The SNR degradation due to aperture jitter (tj) in a high-speed ADC is:
At 1 GHz input frequency, just 100 fs of jitter limits SNR to 56 dB, demonstrating why low-jitter clock distribution is essential in RF sampling ADCs.
Power Efficiency (FOM)
The Walden figure of merit (FOM) compares ADC power efficiency across architectures:
State-of-the-art high-speed ADCs achieve FOM values below 100 fJ/conversion-step, with pipeline and time-interleaved architectures typically outperforming flash designs at resolutions above 8 bits.
1.2 Sampling Theory and Nyquist Criterion
Fundamentals of Sampling
Sampling converts a continuous-time signal x(t) into a discrete-time sequence x[n] by capturing its amplitude at uniformly spaced intervals Ts, where Ts is the sampling period. Mathematically, this is represented as:
The sampling frequency fs is the reciprocal of Ts:
Aliasing and the Nyquist-Shannon Theorem
When a signal is sampled at a rate insufficient to capture its highest frequency components, aliasing occurs—a phenomenon where higher frequencies masquerade as lower ones. The Nyquist-Shannon sampling theorem provides the critical condition to avoid aliasing:
Here, fmax is the highest frequency present in the signal. The term 2fmax is called the Nyquist rate. Sampling below this rate results in spectral overlap, corrupting the signal irreversibly.
Practical Implications in ADC Design
In high-speed ADCs, meeting the Nyquist criterion requires careful consideration of:
- Anti-aliasing filters: Analog low-pass filters must attenuate frequencies above fs/2 before sampling.
- Jitter tolerance: Clock timing uncertainties must be minimized to prevent deviations from uniform sampling intervals.
- Oversampling: Sampling above the Nyquist rate (e.g., 4×) relaxes filter requirements and improves SNR.
Mathematical Derivation of the Sampling Theorem
The Fourier transform of a sampled signal xs(t) (using impulse train modulation) reveals its spectral structure:
For perfect reconstruction, the shifted spectra X(f − kfs) must not overlap. This condition is satisfied only if fs > 2fmax, isolating the baseband spectrum (k = 0).
Beyond Nyquist: Bandpass Sampling
For signals with energy concentrated in a band [fL, fH], the sampling frequency can be reduced below 2fH if:
where n is an integer satisfying 1 ≤ n ≤ ⌊fH/B⌋ and B = fH − fL. This technique is pivotal in software-defined radios.
Real-World Case Study: Undersampling in RF ADCs
Modern RF ADCs (e.g., TI’s ADC12DJ5200RF) exploit bandpass sampling to digitize GHz signals directly. By sampling at 5.2 GSPS, a 6 GHz carrier (with 100 MHz bandwidth) can be aliased to 200 MHz—demonstrating Nyquist’s flexibility in practical systems.
1.3 Quantization Noise and Signal-to-Noise Ratio (SNR)
Quantization noise arises from the fundamental discretization error introduced when an analog signal is converted into a digital representation by an analog-to-digital converter (ADC). The process maps continuous amplitude values to a finite set of discrete levels, introducing an inherent uncertainty bounded by ±½ least significant bit (LSB).
Quantization Error and Noise Power
For a uniformly distributed quantization error e over the interval [−Δ/2, Δ/2], where Δ is the step size (LSB voltage), the probability density function (PDF) is rectangular. The mean square quantization error is derived as:
For an N-bit ADC with a full-scale range VFSR, the step size Δ equals VFSR/(2N). Substituting Δ into the noise power expression yields:
Signal-to-Noise Ratio (SNR)
The SNR for an ADC is defined as the ratio of the power of a full-scale sinusoidal input signal to the quantization noise power. A full-scale sine wave has a peak-to-peak amplitude equal to VFSR and an RMS value of VFSR/(2√2). Its power is:
Combining this with the quantization noise power gives the SNR:
Expressed logarithmically in decibels (dB), this simplifies to the well-known formula:
Practical Considerations
In real-world ADCs, additional noise sources (thermal noise, clock jitter, nonlinearity) degrade the SNR below the theoretical limit. Oversampling and dithering techniques can mitigate quantization noise by spreading its power over a wider bandwidth, allowing subsequent digital filtering to improve effective resolution.
For high-speed ADCs, dynamic performance metrics like spurious-free dynamic range (SFDR) and effective number of bits (ENOB) become critical. ENOB accounts for all noise and distortion, providing a more accurate measure of usable resolution than the nominal bit depth.

2. Basic Structure and Operation
2.1 Basic Structure and Operation
The fundamental architecture of a high-speed analog-to-digital converter (ADC) consists of three primary functional blocks: the sample-and-hold (S/H) circuit, the quantization engine, and the encoding logic. These components work in concert to convert continuous-time analog signals into discrete digital codes with minimal latency and distortion.
Sample-and-Hold Stage
The S/H circuit captures the input voltage at precise intervals defined by the sampling clock. Its operation is governed by:
where Ts is the sampling period. The aperture jitter of this stage directly impacts the ADC's signal-to-noise ratio (SNR), with the relationship:
where σj is the RMS jitter and fin is the input frequency.
Quantization Process
The held voltage is compared against a reference ladder in flash architectures or processed through successive approximation in SAR designs. For an N-bit converter, the least significant bit (LSB) voltage is:
Quantization introduces an irreducible error bounded by ±½LSB, resulting in a theoretical maximum signal-to-quantization-noise ratio (SQNR):
Encoding and Output
The comparator outputs are converted to binary codes through thermometer-to-binary encoders in flash ADCs or shift registers in pipeline designs. Modern high-speed implementations employ Gray coding to minimize metastability errors during high-frequency operation.
Critical timing parameters include:
- Acquisition time: S/H settling duration
- Conversion latency: Pipeline stages × clock period
- Data valid delay: Clock-to-output timing
In time-interleaved architectures, multiple sub-ADCs sample in phase-staggered fashion to achieve aggregate rates exceeding individual converter limits, though mismatches between channels introduce spurious tones requiring careful calibration.

2.2 Advantages and Limitations
Key Advantages of High-Speed ADCs
High-speed analog-to-digital converters (ADCs) enable real-time signal processing in applications requiring wide bandwidths and fast sampling rates. Their primary benefits include:
- Wide Bandwidth Capture: Sampling rates exceeding 1 GSPS allow direct RF sampling, eliminating mixers in software-defined radios and radar systems.
- Time-Interleaved Architectures: Parallel sub-ADCs achieve aggregate sample rates >10 GSPS while maintaining moderate power consumption per channel.
- Improved Resolution at High Frequencies: Modern pipelined ADCs maintain 10-12 ENOB at input frequencies >1 GHz through calibration techniques.
Fundamental Limitations
Despite their capabilities, high-speed ADCs face inherent constraints:
- Aperture Uncertainty: Clock jitter (σjitter) directly limits SNR at high input frequencies:
$$ SNR_{max} = -20 \log_{10}(2\pi f_{in} \sigma_{jitter}) $$For a 5 GHz signal with 100 fs RMS jitter, the theoretical SNR ceiling is 48 dB.
- Power-Speed Tradeoff: Power dissipation scales superlinearly with sampling rate (P ∝ fsα, where α ≈ 1.5-2.0 for CMOS implementations).
- Interleaving Artifacts: Time mismatches between sub-ADCs create spurious tones at fs/M ± fin, requiring calibration circuits that increase die area by 15-30%.
Architecture-Specific Tradeoffs
Flash ADCs
Provide the highest sampling rates (>20 GSPS) but face exponential growth in comparators (2N-1 for N bits), limiting practical resolution to 6-8 bits. Input capacitance scales with comparator count, creating bandwidth bottlenecks.
Pipelined ADCs
Balance speed (1-5 GSPS) and resolution (10-14 bits) through multi-stage quantization, but require precision amplifier settling within half a clock cycle. Digital error correction relaxes comparator requirements at the cost of latency (4-10 cycles).
SAR ADCs
Recent charge-redistribution designs achieve 1-2 GSPS with 8-10 bits while maintaining µW/MHz efficiency. However, their sequential operation makes them sensitive to comparator metastability at ultra-high speeds.
Practical Implementation Challenges
Board-level considerations become critical at multi-GHz sampling rates:
- Skin effect increases conductor loss (α ∝ √f), requiring specialized PCB materials (Rogers 4350, Megtron 6) for interconnects.
- Clock distribution networks demand <λ/10 path length matching to maintain aperture coherence across parallel ADCs.
- Switching noise from digital outputs couples into analog inputs through substrate/supply lines, necessitating isolation techniques like deep n-wells and separate LDO regulators.

2.3 Comparator Design for Flash ADCs
Key Requirements for High-Speed Comparators
In Flash ADCs, comparators must resolve input differences within a single clock cycle, necessitating high gain, low offset, and minimal propagation delay. The metastability error probability Pe scales exponentially with the comparator's time constant τ and the available decision time Td:
where τ is determined by the comparator's small-signal transconductance gm and load capacitance CL:
Latched Comparator Topologies
Regenerative latch comparators dominate high-speed designs due to their positive feedback mechanism. A differential pair with cross-coupled inverters achieves rapid decision-making through exponential voltage separation:
Kickback Noise Mitigation
The switching action injects charge through parasitic capacitances, perturbing reference ladder voltages. Techniques include:
- Pre-amplification: Reduces latch input swing
- Switched-capacitor isolation: Samples inputs before latching
- Guard transistors: Absorb injected charge
Offset Voltage Compensation
Random mismatches in threshold voltages (ΔVTH) and current mirror ratios create input-referred offsets. For an N-bit Flash ADC, the maximum tolerable offset must be less than ½ LSB:
Auto-zeroing techniques store offset on capacitors during reset phases. Dynamic offset cancellation (DOC) achieves <1mV precision in modern CMOS processes.
Propagation Delay Analysis
The total delay tpd comprises preamp settling (tlin) and regeneration time (treg):
where ΔV0 is the initial voltage difference at latch activation. In 28nm CMOS, sub-100ps delays are achievable with optimized transistor sizing.
Power-Speed Tradeoffs
The comparator's power dissipation scales with both bandwidth and resolution. The figure of merit (FoM) combines these factors:
State-of-the-art designs achieve <10fJ/conversion-step by employing:
- StrongARM latches: Combines dynamic precharging with regeneration
- Time-interleaved comparators: Relaxes timing constraints
- Sub-threshold biasing: Reduces gm/ID at marginal speed penalty

3. Stage-by-Stage Conversion Process
3.1 Stage-by-Stage Conversion Process
High-speed analog-to-digital converters (ADCs) often employ a pipeline or multi-stage architecture to achieve both high resolution and fast conversion rates. The stage-by-stage conversion process breaks down the quantization task into smaller, manageable steps, each handled by a sub-ADC and residue amplifier. This approach mitigates the trade-off between speed and accuracy inherent in flash ADCs.
Pipeline Stage Operation
Each stage in a pipeline ADC performs a coarse quantization of the input signal, amplifies the residue, and passes it to the next stage. The process for the i-th stage can be mathematically described as follows:
where:
- \( V_{in,i} \) is the input voltage to the stage,
- \( D_i \) is the digital output from the sub-ADC,
- \( V_{ref} \) is the reference voltage,
- \( G_i \) is the gain of the residue amplifier.
Sub-ADC and Residue Generation
The sub-ADC in each stage typically uses a low-resolution flash architecture (e.g., 1.5–4 bits) for minimal latency. The residue voltage, representing the quantization error, is amplified to improve the signal-to-noise ratio (SNR) for subsequent stages. The gain \( G_i \) is usually set to \( 2^{N_i} \), where \( N_i \) is the number of bits resolved by the stage.
Timing and Synchronization
Pipeline stages operate in a time-interleaved fashion, with each stage processing the residue from the previous stage after a clock cycle. The synchronization of stage outputs is critical to avoid timing skew. Digital correction logic aligns the outputs, compensating for comparator offsets and gain mismatches.
Error Sources and Calibration
Key non-idealities include:
- Gain errors due to capacitor mismatch in switched-capacitor amplifiers,
- Comparator offsets in sub-ADCs,
- Thermal noise from sampling switches and amplifiers.
Modern designs employ foreground or background calibration techniques, such as LMS-based algorithms, to correct these errors dynamically.
Practical Implementation Example
A 12-bit, 500 MS/s pipeline ADC might use eight 1.5-bit stages followed by a 4-bit flash ADC. Each stage resolves 1.5 bits (three comparators), with a gain of 2. The residue amplification allows later stages to refine the LSBs without requiring excessive precision in early stages.
where ENOB (Effective Number of Bits) depends on the signal-to-noise-and-distortion ratio (SNDR). Pipeline ADCs typically achieve ENOBs within 0.5–1.5 bits of their nominal resolution.

3.2 Digital Error Correction Techniques
Digital error correction (DEC) is essential in high-speed ADCs to mitigate nonlinearities, offset errors, and gain mismatches introduced by analog imperfections. By leveraging digital post-processing, DEC enhances the effective resolution and linearity of the converter without requiring prohibitively precise analog components.
Redundancy-Based Correction
Redundancy is a fundamental technique where extra bits or stages are incorporated to detect and correct errors. In pipeline ADCs, for instance, digital correction logic compensates for comparator offsets and capacitor mismatches by overlapping the quantization ranges of adjacent stages. The redundancy allows a margin for error in the analog domain, which is later resolved digitally.
Here, Vres is the residual voltage passed to the next stage, Dout is the coarse digital estimate, and Vref is the reference voltage. The digital backend reconstructs the final output by combining the coarse and fine conversions while correcting for overlap errors.
Calibration Techniques
Background and foreground calibration methods dynamically adjust ADC parameters to correct for drift and process variations. Least-mean squares (LMS) and lookup table (LUT)-based approaches are commonly employed:
- Foreground Calibration: Performed during startup or idle periods, this method injects known test signals to measure and correct static errors.
- Background Calibration: Continuously adapts to changing conditions by comparing ADC outputs with statistical or redundant reference measurements.
In LMS calibration, Δwk represents the weight update, μ is the step size, ek is the error signal, and xk is the input. This iterative process minimizes the mean-squared error between the actual and ideal ADC output.
Error Averaging and Noise Shaping
Time-interleaved ADCs benefit from error averaging across multiple channels to reduce timing skew and gain mismatches. Additionally, noise-shaping techniques, such as delta-sigma modulation, push quantization noise out of the band of interest, which is later filtered digitally.
The figure illustrates noise shaping in a time-interleaved ADC, where the combined output averages out uncorrelated errors from individual channels.
Practical Implementation Considerations
Implementing DEC requires careful trade-offs between correction accuracy, latency, and hardware complexity. For instance, redundant signed digit (RSD) coding in pipeline ADCs simplifies error detection but increases digital logic overhead. Similarly, adaptive calibration algorithms must converge rapidly without introducing instability.
Modern high-speed ADCs often combine multiple DEC techniques, such as redundancy with background calibration, to achieve >14-bit effective resolution at sampling rates exceeding 1 GS/s. Case studies in SerDes interfaces and radar systems demonstrate DEC's critical role in maintaining signal integrity under real-world conditions.

3.3 Power and Speed Trade-offs
The relationship between power consumption and conversion speed in high-speed ADCs is governed by fundamental physical limits and architectural choices. As sampling rates increase, power dissipation grows due to increased switching activity, higher bandwidth requirements, and reduced settling times. The power-speed trade-off can be analyzed through several key metrics.
Thermodynamic Limits and kT/C Noise
At the circuit level, the minimum energy required for a single conversion step is constrained by thermal noise. For an N-bit ADC, the theoretical lower bound on power consumption is derived from the kT/C noise limit:
where k is Boltzmann's constant, T is temperature, and fs is the sampling frequency. Practical implementations typically consume 10-100× this theoretical minimum due to non-ideal circuit behavior.
Architectural Dependencies
Different ADC architectures exhibit distinct power-speed scaling characteristics:
- Flash ADCs: Power scales exponentially with resolution (P ∝ 2N) due to the comparator bank, but achieves the highest speeds
- Pipeline ADCs: Power scales linearly with resolution and sampling rate (P ∝ N·fs)
- SAR ADCs: Power scales favorably (P ∝ log2N·fs) but with inherent speed limitations
Figure of Merit Analysis
The Walden FoM (Figure of Merit) quantifies the energy efficiency of ADCs:
where ENOB is the effective number of bits. State-of-the-art designs achieve FoMW values below 10 fJ/conversion-step at multi-GS/s speeds through techniques like:
- Time-interleaved architectures with optimized clock distribution
- Advanced CMOS processes with reduced parasitic capacitance
- Dynamic comparator designs with reduced kickback noise
Practical Design Considerations
In real-world implementations, additional factors influence the power-speed trade-off:
- Clock jitter requirements become more stringent at higher speeds (σjitter < 1/(2N+1πfin))
- Interconnect RC delays limit maximum sampling rates in large-array designs
- Supply voltage scaling reduces dynamic power but increases sensitivity to threshold variations
Modern high-speed ADCs often employ hybrid architectures to balance these constraints. For example, a 10-bit 5 GS/s ADC might combine:
- 4-way time-interleaving to reduce per-channel speed requirements
- Pipeline stages with optimized stage resolutions
- Background calibration to maintain linearity at reduced power
Emerging technologies like superconducting ADCs and photonic sampling promise to push these limits further, with demonstrated FoM improvements of 10-100× over conventional CMOS approaches in specialized applications.

4. Binary Search Algorithm
4.1 Binary Search Algorithm
The binary search algorithm is fundamental to successive approximation register (SAR) analog-to-digital converters (ADCs), enabling high-speed conversion with minimal hardware complexity. The method operates by iteratively narrowing down the input voltage range through a series of comparisons against a digitally generated reference.
Mathematical Basis of Binary Search
For an N-bit ADC, the binary search requires exactly N steps to resolve the input voltage Vin to within one least significant bit (LSB). At each step k, the algorithm compares Vin against a threshold VDAC(k), generated by a digital-to-analog converter (DAC). The decision rule is:
where bk is the k-th bit of the output code. The DAC voltage is updated recursively:
Hardware Implementation
A SAR ADC implementing this algorithm consists of:
- Sample-and-hold (S/H) circuit to freeze the input voltage during conversion.
- Comparator to perform the binary decisions.
- SAR logic to control the iterative search. N-bit DAC to generate the reference voltages.
Performance Considerations
The conversion time Tconv of a binary search ADC is deterministic:
where TDAC, Tcomp, and Tlogic are the settling time of the DAC, comparator decision time, and SAR logic delay, respectively. This makes SAR ADCs particularly suitable for applications requiring predictable latency.
Practical Limitations
While theoretically elegant, real-world implementations face challenges:
- Comparator metastability can cause bit errors when Vin ≈ VDAC.
- DAC nonlinearity introduces integral nonlinearity (INL) errors.
- Clock jitter becomes critical at high speeds (>10 MSPS).
Modern SAR ADCs mitigate these issues through techniques like redundant signed digit (RSD) correction and background calibration, pushing resolutions to 16 bits at multi-megasample rates.

4.2 Capacitor DAC Design
The capacitor-based digital-to-analog converter (DAC) is a fundamental building block in high-speed successive approximation register (SAR) ADCs. Its performance directly impacts linearity, settling time, and power efficiency. This section explores key design considerations, including capacitor matching, charge redistribution, and noise analysis.
Charge Redistribution Principle
The operation of a capacitor DAC relies on charge conservation during switching events. Consider a binary-weighted capacitor array with N bits, where the unit capacitance is Cu. The total capacitance is:
During conversion, capacitors are switched between Vref and ground, redistributing charge to generate an analog output voltage. The output for a digital code D is:
Capacitor Mismatch and Linearity
Non-ideal capacitor matching introduces integral nonlinearity (INL) and differential nonlinearity (DNL). The standard deviation of capacitance mismatch in modern CMOS processes is typically 0.1% to 0.5%. For an N-bit DAC, the worst-case INL due to mismatch is:
Common layout techniques to mitigate mismatch include:
- Common-centroid routing to minimize gradient effects
- Dummy capacitors at array boundaries
- Unit capacitor replication with interdigitated placement
kT/C Noise and Power Scaling
The thermal noise power of a capacitor DAC is determined by the sampling operation:
For a target signal-to-noise ratio (SNR), the minimum unit capacitance is:
This leads to a fundamental trade-off between noise, resolution, and power consumption. Advanced designs employ noise-shaping techniques or split-capacitor arrays to relax this constraint.
Switching Schemes and Energy Efficiency
Traditional capacitor DACs consume significant energy during bit trials. Modern switching schemes reduce power dissipation:
- Monotonic switching: Eliminates redundant transitions
- Vcm-based switching: Reuses charge from previous comparisons
- Split-capacitor arrays: Reduces effective capacitance for MSBs
The energy per conversion for a monotonic switching scheme is:
compared to the conventional scheme's CtotalVref2.
Advanced Architectures
Recent research has demonstrated hybrid DAC architectures combining capacitor arrays with:
- Resistor ladders for improved linearity in MSBs
- Time-domain interpolation to reduce capacitor count
- Dynamic element matching to average out mismatch errors
These techniques enable 12-16 bit resolution at sampling rates exceeding 100 MS/s in advanced CMOS nodes.

4.3 High-Speed SAR ADC Techniques
Successive Approximation Register (SAR) ADCs are widely used for medium-to-high-resolution applications, but achieving high-speed operation requires overcoming inherent trade-offs between resolution, power consumption, and settling time. Recent advances in SAR ADC architectures have pushed sampling rates into the hundreds of MS/s while maintaining 10–12-bit resolution.
Time-Interleaved SAR ADCs
Time-interleaving multiple SAR ADCs is a common technique to increase effective sampling rates. For M interleaved channels, the aggregate sampling rate becomes:
However, mismatches in offset, gain, and timing skew between channels introduce spurious tones. Calibration techniques such as:
- Background digital correction
- Pseudo-random dithering
- Time-skew estimation via correlation
are essential for maintaining SFDR > 70 dB in interleaved designs.
Asynchronous SAR Control Logic
Traditional synchronous SAR ADCs waste time with worst-case bit-cycling. Asynchronous control eliminates clock cycle redundancy by:
- Using comparator ready signals to trigger next bit trial
- Dynamic logic that propagates completion signals
- Self-timed capacitor switching networks
This approach reduces conversion time by 30–50% compared to synchronous implementations. The maximum sampling rate becomes limited by comparator regeneration time and DAC settling:
where N is resolution in bits, tcomp is comparator delay, and tDAC is capacitive DAC settling time.
Split-Capacitor DAC Techniques
The DAC settling time dominates high-speed SAR ADC performance. Split-capacitor architectures improve speed by:
- Dividing MSB and LSB capacitor arrays
- Using charge redistribution between sub-arrays
- Reducing effective RC time constants
The energy per conversion step in a split-capacitor DAC scales as:
where Cunit is the unit capacitance and N is the ADC resolution. This achieves 40–60% power reduction compared to conventional binary-weighted DACs.
Comparator Metastability Mitigation
At high speeds, comparator metastability causes catastrophic decision errors. Advanced techniques include:
- Pre-amplifiers with built-in hysteresis
- Parallel comparators with voting logic
- Time-domain metastability detection and correction
The metastability error rate follows:
where τ is the available decision time and τ0 is the comparator time constant. Modern designs achieve error rates < 10-12 at 500 MS/s.
Hybrid SAR-Pipeline Architectures
Combining SAR and pipeline stages leverages the best of both architectures:
- SAR for high-accuracy residue generation
- Pipeline for high-speed coarse conversion
The optimal stage resolution balance minimizes total power:
where Cpar is parasitic capacitance. Recent 1 GS/s implementations achieve 10 ENOB with 15 mW power consumption in 28 nm CMOS.

5. Principle of Time-Interleaving
5.1 Principle of Time-Interleaving
Time-interleaving is a technique used to achieve sampling rates beyond the limits of a single analog-to-digital converter (ADC) by combining multiple ADCs operating in parallel. The fundamental concept relies on phase-shifting the sampling clocks of M identical ADCs such that each converter samples the input signal at staggered time intervals.
Mathematical Foundation
For M interleaved ADCs with a sampling period Ts, the effective sampling period becomes:
Each ADC samples the input signal at time-shifted intervals of Teff. The sampling instants for the k-th ADC (where k = 0,1,...,M-1) are given by:
Clock Phase Relationships
The clocks for an M-way time-interleaved ADC must maintain precise phase alignment. For ideal operation, the phase shift between adjacent ADCs must be:
Any deviation from this ideal phase relationship introduces timing skew errors, which manifest as spurious tones in the output spectrum.
Practical Implementation Challenges
Real-world implementations must address several critical non-idealities:
- Timing skew: Mismatches in clock distribution paths cause sampling time errors
- Gain mismatch: Variations in ADC front-end gain produce amplitude discontinuities
- Offset mismatch: DC offsets between channels create fixed pattern noise
- Bandwidth mismatch: Differences in analog front-end bandwidths cause frequency-dependent errors
Error Correction Techniques
Modern time-interleaved ADCs employ several calibration methods:
- Background calibration using pseudo-random dither signals
- Digital post-processing with adaptive filters
- Clock deskew circuits with sub-picosecond resolution
- Mixed-signal correction loops for gain and offset
Performance Metrics
The effective resolution bandwidth (ERBW) of a time-interleaved ADC system is given by:
where BWanalog is the bandwidth of the individual ADC channels. The spurious-free dynamic range (SFDR) is typically 10-20 dB lower than for a single ADC due to interleaving artifacts.
Applications in High-Speed Systems
Time-interleaving enables ADC solutions for:
- 5G mmWave receivers (sample rates > 10 GS/s)
- Ultra-wideband radar systems
- High-speed oscilloscopes and test equipment
- Direct RF sampling architectures
State-of-the-art implementations achieve >100 GS/s sampling rates using 16-32 way interleaving in advanced CMOS and SiGe processes.

5.2 Channel Mismatch Calibration
In time-interleaved ADCs, channel mismatches introduce nonlinear distortions that degrade signal integrity. These mismatches arise from three primary sources: offset, gain, and timing skew. Calibration techniques mitigate these errors by either foreground or background correction methods.
Offset Mismatch Correction
Offset mismatch occurs when DC levels differ between channels. A straightforward calibration involves measuring the average output of each channel with a zero input and subtracting the offset:
where Oraw[k] is the raw offset of the k-th channel and N is the number of channels. This correction is typically performed during a dedicated calibration phase.
Gain Mismatch Correction
Gain variations result in amplitude mismatches between channels. Calibration involves applying a known reference signal and adjusting each channel’s gain to match a reference channel. The corrected output is:
where Aref is the reference amplitude and Ak is the measured amplitude of the k-th channel.
Timing Skew Calibration
Timing skew, the most challenging mismatch to correct, arises from sampling phase misalignment. Background calibration techniques often leverage statistical properties of the input signal. A common approach uses a least-mean-square (LMS) adaptive filter to estimate and compensate for skew:
where μ is the step size, e[n] is the error signal, and ∂y[n]/∂t is the signal derivative.
Practical Implementation Considerations
- Foreground Calibration: Requires interruption of normal operation but offers high precision.
- Background Calibration: Operates continuously but may introduce latency.
- Mixed-Signal Techniques: Combine analog tuning with digital correction for optimal performance.
Advanced ADCs, such as those used in 5G and radar systems, often integrate on-chip calibration engines to dynamically adjust for environmental variations.

5.3 Applications in Ultra-High-Speed Systems
High-speed analog-to-digital converters (ADCs) are critical in systems where signal bandwidths exceed several gigahertz. These applications demand architectures that balance resolution, sampling rate, and power efficiency while mitigating high-frequency distortion mechanisms such as jitter-induced noise and aperture uncertainty.
Photonics and Optical Communication
Coherent optical receivers leverage high-speed ADCs with sampling rates beyond 50 GS/s to demodulate quadrature amplitude-modulated (QAM) signals. The time-interleaved pipeline architecture is dominant here, as it achieves the required effective number of bits (ENOB) while compensating for skew errors through digital calibration. For instance, a 64-QAM system operating at 32 GBaud requires an ADC with at least 8-bit resolution and a signal-to-noise ratio (SNR) exceeding 40 dB.
where N is the ADC resolution in bits, fs the sampling rate, and B the signal bandwidth.
Radar and Electronic Warfare
Phased-array radar systems utilize high-speed ADCs for digital beamforming, where direct RF sampling at X-band (8–12 GHz) eliminates mixers and IF stages. Flash ADCs with 6–8 bits and >10 GS/s rates are common, employing time-interleaved architectures to maintain spurious-free dynamic range (SFDR) above 60 dBc. Digital post-processing corrects for interleaving artifacts like gain mismatches.
Particle Physics and Synchrotron Facilities
Ultra-high-speed ADCs digitize detector outputs in particle colliders, where event rates exceed 100 MHz. Successive approximation register (SAR) ADCs with redundancy-aided calibration achieve 12-bit resolution at 1 GS/s, critical for energy and timing measurements. Jitter requirements are stringent, with aperture jitter < 100 fs to prevent timing smearing:
5G mmWave and Beyond
Millimeter-wave 5G systems (24–100 GHz) employ high-speed ADCs in hybrid beamforming architectures. RF-sampling ADCs with 10–12 bits and 5–10 GS/s enable direct conversion, reducing phase noise compared to heterodyne systems. Delta-sigma modulators with noise shaping are favored for their inherent anti-aliasing in crowded spectra.
Test and Measurement Equipment
Oscilloscopes with >100 GHz bandwidths rely on interleaved ADCs to achieve real-time sampling rates exceeding 200 GS/s. Calibration techniques like background gradient-error correction mitigate nonlinearities in the folding stages of subranging architectures. The figure of merit (FoM) for such ADCs is typically below 50 fJ/conversion-step:
Emerging applications like quantum computing readout and terahertz imaging continue to push ADC performance boundaries, driving research into photonic ADCs and superconducting quantizers.
6. Key Research Papers and Books
6.1 Key Research Papers and Books
- PDF Design of High-Speed Analog-to-Digital Converters using Low-Accuracy ... — New architectures and circuits for high-speed ADCs are explored in Paper II to Paper VI with test-chips based on the Flash and Pipelined ADC architectures. Two Flash ADCs were developed, both based on a new comparator that suppresses common-mode kick-back by a factor of 6x compared to conventional topologies, which is presented in paper II.
- PDF High-Speed ADC Architectures - Springer — This book focus on CMOS high-speed ADC architectures - two-step flash and folding and interpolation - which have the characteristic of possessing sub-blocks with no special linearity ments, but whose offset voltages are of the upmost importance they determine the overall ADC linearity. Those architectures described in detail in the present chapter. The techniques improve the linearity of ...
- Design of Sample and Hold for High-Speed Analog to Digital ... - Springer — ADC is the main analog block in space application such as video processing, data acquisition system, bias generator. Sample and hold circuit is crucial component of high-speed ADC. It is very important to design CMOS implementation of ADC with reduced size and cost for space application. In sample and hold design, Op-Amp is the key component.
- Design and Comparative Analysis of Low-Power, High-Speed, 3 ... - Springer — A biomedical signal is mostly in the range of mV [4]. In this paper, flash or parallel ADC is considered since their speed of converting an analog signal to digital signal is faster than other ADCs. This ADC is the most suitable for several applications and hardware implementation of low-resolution, high-speed, modern-day electronic circuit design.
- PDF Comparator Design for High Speed ADC - Lu — Abstract High-speed Analog-to-Digital Converters (ADCs) play an important role in mod-ern electronic systems, especially those systems that require high data rates and low power consumption. These converters are essential components in various applications, including telecommunications, medical imaging, radar systems, and wireless communication.
- PDF Design of High-speed, High-resolution Sar A/D — A modified pipelined-SAR architecture is pro-posed, which uses two switched-capacitor digital-to-analog converters (DACs) at the ADC frontend. This technique separates the high-speed SAR operation from the low noise residue computation and improves the conversion speed to over 150 MS/s while maintaining an SNDR > 65 dB with good power efficiency.
- PDF Design of High Speed, Low Power, to Digital Converters - DiVA — This licentiate thesis presents my research during the period March 2006 to July 2009 at the Electronic Devices group, Department of Electrical Engineering, Linköping University, Sweden. The following papers are included in the thesis: •Paper I - Timmy Sundström, Boris Murmann and Christer Svensson, "Power Dissipation Bounds for High‐Speed Nyquist Analog‐to‐Digital Converters ...
- Full text of "Circuit Analysis Theory And Practice" - Archive.org — the situation in the front stages of a high resolution ADC [76, 77, 73], Since 1 .5 bits/stage architecture does not permit extensive scaling, a higher speed can
- High-speed ADC systems with HBTs for measuring ... - ScienceDirect — This paper presents very high-speed Analog-to-Digital Converter (ADC) systems for measuring instrument applications, and also related theoretical results.
- PDF Comparator Design for High-Speed ADCs - DiVA portal — The ADC's architecture can vary, but in this report, a Successive-Approximation-Register (SAR) ADC will be used as an example, which consists of a SAR block, a comparator, and a Digital-to-Analog converter (DAC) block.
6.2 Online Resources and Tutorials
- PDF Section 6.1: Digital-to-analog Converter Architectures — high resolution measurement sigma-delta adcs 6.102 band-pass sigma-delta converters 6.107 sigma-delta dacs 6.108 summary 6.110 references 6.111 section 6.4: defining the specifications 6.115 section 6.5: dac and adc static transfer functions and dc errors 6.117 section 6.6: data converter ac errors 6.129 noise in practical adcs 6.131
- Understanding Serial LVDS Capture in High-Speed ADCs - Texas Instruments — A typical multichannel device has one (or two) LVDS pairs per ADC channel, one common bit clock output, and one frame clock output. Refer to Figure 1-2for the LVDS output interface of an 8-channel ADC device. Figure 1-2.Typical Multichannel ADC with a Serial LVDS Interface 6 Understanding Serial LVDS Capture in High-SpeedADCs SBAA205- July 2013
- PDF Chapter 6 Interfacing to Data Converters F - Analog — 3. Data Converter Architectures 4. Data Converter Process Technology 5. Testing Data Converters 6. Interfacing to Data Converters 6.1 Driving ADC Analog Inputs 6.2 ADC and DAC Digital Interfaces 6.3 Buffering DAC Analog Outputs 6.4 Driving ADC and DAC Reference Inputs 6.5 Sampling Clock Generation 7. Data Converter Support Circuits 8.
- PDF High-speed Pipelined Adc Using a Bucket Brigade Front-end — electronic format. An original signed hard copy of the signature page is on file in ... Archives. iii. iv Abstract Advanced wireless technologies, such as LTE and LTE advanced, require low-power, high-speed, and high-resolution analog-to-digital converters (ADCs). At present, ... 5.1.1 ADC High Level Architecture ..... 56 5.1.2 Capacitor Sizing ...
- PDF Dynamic Amplifiers for High-speed Pipelined A/D Conversion — HIGH-SPEED PIPELINED A/D CONVERSION . A DISSERTATION . SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ... which take 35 times more energy than other architectures that do not - emphasize high speed. Furthermore, existing non-pipelined solutions for this ... 6 2.1. REVIEW OF BASIC OPERATION AND ARCHITECTURE.....6 2.2. CONVENTIONAL RESIDUE AMPLIFIER ...
- PDF Design of High-Performance Pipeline Analog-to-Digital Converters in Low ... — High-performance applications such as broadband communication systems require high-performance analog-to-digital converters (ADCs) with high-resolution and band-width (over 14 bits and several MHz). Such applications are often the domain of pipeline ADCs, due to their highly e-cient and conversion-speed-centric architecture. The in-
- Pipelined ADC Design and Enhancement Techniques - Academia.edu — The Analog-to-Digital Converter (ADC) is the main link between the analog input and DSP part. However, for applications like hand-held or wireless devices, ADC should be featured with low power and high speed. The pipeline ADC architecture is best suitable for medium resolution, low power and high-speed applications.
- PDF AN1189: Incremental Analog to Digital Converter (IADC) - Silicon Labs — The EFR32 Wireless Gecko Series 2 IADC is an intermediate architecture combining techniques from both Successive Approximation Register (SAR) and Delta-Sigma style converters. The maximum resolution for normal and high speed1 modes is 12 bits without over-sampling, which can achieve up to one million samples per second (1 Msps) in normal mode ...
- PDF Comparator Design for High-Speed ADCs - DiVA portal — the architecture in older technology, their performance was limited. Today, there are many different architectures, where the most promising will be evaluated in this report. To reduce the problems with different trade-offs, architectures like the double-tail latch became popular in high-speed circuits. This architecture has a first stage acting
- PDF SAR Cyclic and Integrating ADCs - University of Delaware — SAR ADC Limitations - 14 - •Conversion rate typically limited by finite bandwidth of RC network during sampling and bit-tests •For high resolution, the binary weighted capacitor array can become quite large •E.g. 16-bit resolution, C total~100pF for reasonable kT/C noise contribution
6.3 Industry Standards and Datasheets
- High-Speed Data Interface for Precision High-Speed ADC in Semiconductor ... — as the ADC can rapidly switch between signals to capture data. All of these benefits of a high-speed ADC can help improve the accuracy and efficiency of testing electronic devices. A high-speed ADC requires a high-speed data interface with the controller of the system for transmission of digital data.
- PDF Chapter 6 Interfacing to Data Converters F - Analog — 3. Data Converter Architectures 4. Data Converter Process Technology 5. Testing Data Converters 6. Interfacing to Data Converters 6.1 Driving ADC Analog Inputs 6.2 ADC and DAC Digital Interfaces 6.3 Buffering DAC Analog Outputs 6.4 Driving ADC and DAC Reference Inputs 6.5 Sampling Clock Generation 7. Data Converter Support Circuits 8.
- Data converter architectures | High Speed Data Converters — In this chapter, we discuss some high speed ADC architectures, which include the flash, pipelined, and time-interleaved ADCs. In addition, architectures that were historically used in low speed applications, such as successive approximation (SAR) and delta-sigma converters, are covered because of their recent resurgence in the high speed space.
- PDF Data Converters for High Speed CMOS Links A PhD Thesis — high speed data converter performance. The transceiver chip provides a high bandwidth signal path and precision clocks, despite the large parasitic capacitances and transistor matching errors of CMOS technology. Small, high bandwidth sample-and-hold amplifiers are used in the ADC, and
- PDF Session 22 Overview: High-Speed Data Converters — resolution. We present an interleaved CMOS ADC architecture based on an asynchronous redundant SAR ADC core element. It was measured up to a sampling rate of 100GS/s and can be operated from a single supply voltage. At 90GS/s, the measured SNDR stays above 36.0dB SNDR up to 6.1GHz and 33.0dB up to 19.9GHz input frequency while consuming 667mW ...
- PDF High Speed Analog to Digital Converter Basics - Texas Instruments — High-Speed,Analog-to-DigitalConverter Basics ... The goal of this document is to introduce a wide range of theories and topics that are relevant to high-speed, analog-to-digitalconverters (ADC). This document provides details on sampling theory, data-sheetspecifications, ADC selection criteria and evaluation methods, clock jitter, and other ...
- ADADC85 Datasheet and Product Info | Analog Devices — chip count—high reliability; Industry-standard pinout; Z models for ±12 V operation available; ... Data Sheet 1. ADADC84/ADADC85: Fast, Complete 12-Bit A/D Converters Data ... Designing Power Supplies for High Speed ADC 02/01/2012; Cross Reference Guide ...
- GitHub - bmurmann/ADC-survey: ADC Performance Survey 1997-2024 (ISSCC ... — ADC Performance Survey 1997-2024 (ISSCC & VLSI Circuit Symposium) - bmurmann/ADC-survey. ... By industry. Healthcare Financial services Manufacturing Government View all industries ... a versatile building block for ultra-low-power to ultra-high-speed applications," in IEEE Communications Magazine, vol. 54, no. 4, pp. 78-83, Apr. 2016. ...
- High-Speed ADC Architectures - SpringerLink — High-Speed ADC Architectures Download book PDF. Part of the book series: Analog Circuits and Signal Processing Series ((ACSP)) 1424 Accesses. The rapid technological evolution leads to CMOS technologies with lower feature sizes, allowing the integration of more complex systems in an single integrated circuit. The digital circuits, which are ...
- Which ADC Architecture Is Right for Your Application? — Figure 1. ADC architectures, applications, resolution, and sampling rates. The classification in Figure 1 shows in a general way how these application segments and the associated typical architectures relate to ADC resolution (vertical axis) and sampling rate (horizontal axis). The dashed lines represent the approximate state of the art in mid ...








