Fact-checked by Grok 2 weeks ago

Switched capacitor

A switched-capacitor is an that employs switches and capacitors to realize discrete-time operations, such as , , and filtering, by periodically sampling and transferring charge between capacitors under clock control. These circuits emulate the behavior of resistors through the equivalent resistance of a switched capacitor, given by R_{eq} = \frac{1}{f_c C}, where f_c is the clock and C is the , enabling precise control without physical resistors. Developed in the late 1970s amid the rise of integrated circuit technology, switched-capacitor circuits addressed the challenges of fabricating accurate resistors in silicon processes, with foundational contributions from researchers like James C. Candy and Gabor C. Temes in 1977, who established key principles for their use in sampled-data systems. Over the subsequent decades, particularly through the , they evolved into a cornerstone of analog design, benefiting from high , low power consumption, and tunability via clock frequency adjustments, which allow dynamic reconfiguration without altering component values. Switched-capacitor circuits operate in two non-overlapping clock : a sampling where the input signal charges a through a switch, and a transfer where that charge is redistributed to another , often via an , to perform functions like or . This charge-transfer mechanism, rooted in early concepts of from James Clerk Maxwell's work, ensures high and in discrete-time , though it introduces challenges like clock feedthrough and that require careful design mitigation. Key applications span and , including active filters for audio and video systems, analog-to-digital converters (ADCs) and digital-to-analog converters (DACs) in , switched-capacitor integrators in sigma-delta modulators, and DC-DC converters for efficient without inductors, particularly in low- to medium-power integrated circuits. In biomedical and RF domains, they enable compact sensor interfaces and low-voltage operations, while their compatibility with deep-submicron processes supports ongoing advancements in mixed-signal systems.

Overview

Definition and Basic Operation

A switched capacitor is a fundamental building block in analog integrated circuits, consisting of a capacitor connected to one or more switches that are controlled by a periodic to enable discrete-time charge transfer between nodes, rather than continuous current flow as in traditional resistor-based circuits. This configuration allows the switched capacitor to simulate the behavior of resistors, inductors, or other continuous-time elements in a sampled-data system, making it particularly suitable for discrete-time tasks such as filtering and . In its basic operation, the switches alternate between two non-overlapping clock phases to facilitate charge transfer. During the first phase (typically denoted as φ1), the capacitor is connected to an input (), charging to a voltage proportional to and storing charge Q = C × , where C is the . In the second phase (φ2), the switches reconfigure to connect the capacitor to an output , transferring the stored charge to that , which effectively dumps the charge and creates an average flow over the clock cycle. This process repeats with each clock period T, mimicking a resistive I_eq = ( - Vout) / R_eq, where the equivalent is given by R_{eq} = \frac{T}{C} with T as the clock period (or equivalently, R_eq = 1 / (f_clk × C), where f_clk is the clock ). The timing of switch states can be visualized as follows: φ1 closes the input switch while φ2 opens it, and vice versa for the output switch, ensuring no overlap to prevent charge leakage or short circuits. Switched capacitors offer significant advantages in integrated circuit design, particularly in CMOS processes, where fabricating precise resistors is challenging due to process variations and area constraints. They provide high precision through accurate capacitance ratios (often better than 0.1% matching), tunability by adjusting the clock frequency to alter the effective resistance without changing hardware, and inherent compatibility with digital clock signals, reducing the need for continuous analog components. These attributes make switched capacitors ideal for on-chip implementations in applications like filters and data converters.

Historical Development

The roots of switched-capacitor technology lie in early sampled-data systems developed during the , building on foundational work in signal sampling and discrete-time processing following Claude Shannon's sampling theorem in 1949. Concepts for charge-transfer devices, which enabled the manipulation of discrete charge packets on capacitors, emerged in the late 1960s, with inventions like the and laying groundwork for inductorless filters to replace bulky passive components in analog circuits. These concepts addressed the need for compact, active realizations of filters amid the transition from vacuum tubes to . A key milestone occurred in the 1970s, when advancements in MOSFET technology enabled reliable analog switches, spurring the development of practical switched-capacitor circuits by companies like and . This era saw the realization of resistor emulation through switched capacitors, allowing fully integrated analog filters without precision s. The first commercial switched-capacitor filters appeared in the late 1970s, with Intel's 2912 PCM filter introduced in 1979, revolutionizing monolithic filter design for applications like and audio processing. Influential researchers, including James C. Candy, Gabor C. Temes, George S. Moschytz, and R. , contributed significantly in the late 1970s; Candy and Temes established key principles for their use in sampled-data systems, while Moschytz advanced analysis and synthesis methods for MOS switched-capacitor filters, and developed key integrator topologies insensitive to parasitics. During the and , switched-capacitor techniques proliferated with scaling, enabling integration into analog-to-digital converters (ADCs) and digital-to-analog converters (DACs) for higher resolution and lower power. This period marked the rise of delta-sigma modulators, where switched-capacitor integrators facilitated and noise shaping, achieving resolutions beyond 16 bits in processes down to 1 μm. By the mid-, these circuits were standard in mixed-signal ICs, benefiting from submicron for denser, more efficient implementations. In the post-2000 era, switched-capacitor circuits have played a pivotal role in low-power designs for () devices, leveraging nanoscale nodes below 65 nm to minimize area and energy consumption. Ongoing research focuses on adaptive switched-capacitor architectures for applications, such as reconfigurable receivers that use stacked capacitor networks to reject while operating at sub-1 V supplies and under 1 mW. Recent advancements up to 2025 incorporate AI-optimized tuning for power converters and equalizers, enhancing efficiency in battery-constrained edge devices through automated parameter optimization.

Fundamental Principles

Resistance Simulation

Switched capacitors simulate resistive behavior through periodic charge and discharge cycles driven by a , effectively transferring charge between two nodes to mimic flow. In this configuration, a is alternately connected to an input voltage V_1 and an output voltage V_2 using non-overlapping clock phases, resulting in an average I = C (V_1 - V_2) / T, where C is the and T is the clock period. This mechanism allows the switched capacitor to emulate a without the physical implementation of resistive elements, which is particularly advantageous in integrated circuits where precise resistors are challenging to fabricate. The derivation of the equivalent resistance begins with the charge transferred per switching cycle. Assuming ideal switches with zero and instantaneous switching, the charges to a voltage difference \Delta V = V_1 - V_2 during one , storing charge Q = C \Delta V = C (V_1 - V_2). In the subsequent , this charge is transferred to the output node. The average current over the clock period T is then I_\text{avg} = Q / T = C (V_1 - V_2) / T. By definition, for a , I = (V_1 - V_2) / R_\text{eq}, so equating the expressions yields R_\text{eq} = T / C. Since the clock f_\text{clk} = 1 / T, this simplifies to R_\text{eq} = 1 / (C f_\text{clk}). This equivalence holds under the assumptions of ideal components and a signal much lower than the clock . A typical for this parallel switched-capacitor consists of a C connected between two nodes via four switches: two controlled by clock \phi_1 (connecting C to V_1 and or during charging) and two by complementary \phi_2 (connecting to V_2 and during discharging). The switches ensure non-overlapping operation to prevent short circuits, with the capacitor effectively bridging the nodes in a time-averaged sense. The equivalent is inversely proportional to the clock , enabling by adjusting f_\text{clk}; for instance, doubling the halves R_\text{eq}. This dependence is valid in the low-frequency regime where the signal is significantly below f_\text{clk}, ensuring the continuous-time applies. Practical considerations include a minimum to maintain the resistor approximation, as excessively low frequencies lead to discrete-time effects dominating the behavior. Additionally, non-zero switch on-resistance introduces a parallel path that slightly reduces the effective resistance, though this is minimal with low-resistance switches in modern processes.

Capacitor Charge Transfer

In switched capacitor circuits, the core mechanism involves the precise transfer of charge stored on a , governed by principles of . During the charging phase (phase 1), the C is connected to the input voltage V_{in} via a switch, allowing it to accumulate a charge Q = C \cdot V_{in}. This process ensures that the 's voltage equilibrates with V_{in}, storing energy proportional to the input signal. In the subsequent transfer phase (phase 2), the switch reconfigures the capacitor to connect to the output at voltage V_{out}, enabling charge redistribution. The amount of charge transferred to the output is \Delta Q = C \cdot (V_{in} - V_{out}), reflecting the difference in voltages across the capacitor before and after switching. This transfer is inherently voltage-dependent: if V_{out} \neq 0, the process is incomplete, yielding an effective gain of less than unity due to partial charge retention on the capacitor. Over successive clock cycles, the cumulative of these \Delta Q transfers determines the average circuit behavior, such as in discrete-time . The accuracy of charge transfer relies critically on switch timing, implemented using non-overlapping two-phase clock signals (\phi_1 and \phi_2) to avoid erroneous charge sharing between phases. Non-overlap ensures that switches do not conduct simultaneously, preventing unintended paths for charge flow; for instance, \phi_2 turns off before \phi_1 activates, minimizing transient errors. A standard two-phase clock features \phi_1 high during charging and \phi_2 high during transfer, with a brief dead time between them to maintain isolation. To achieve precise voltage scaling or in switched capacitor operations, circuits often employ multiple capacitors with carefully matched ratios, as the transfer efficiency and overall function depend on relative capacitances rather than absolute sizes. For example, connecting capacitors in series or configurations allows ratios like C_1 / C_2 to define the scaling factor, enabling high accuracy (typically 0.1% matching in integrated implementations). This ratio-based approach underpins the equivalence to resistive elements in discrete-time domains, though the atomic charge transfer remains the foundational physics.

Core Circuits

Parasitic-Sensitive Integrators

The parasitic-sensitive switched-capacitor integrator is a fundamental circuit in discrete-time , employing an (op-amp) with an input sampling C_1 and a feedback integrating C_2. The topology features C_1 connected such that one terminal is switched between the input voltage V_{in} and a (typically ), while the other terminal links to the op-amp's inverting summing junction (); C_2 connects between the op-amp output V_{out} and the same summing junction. Switches, controlled by non-overlapping clock phases \phi_1 and \phi_2, facilitate charge transfer without continuous elements, enabling monolithic implementation. Operation occurs in two phases. During \phi_1 (sampling), C_1 charges to the input voltage V_{in}(nT), storing charge Q_1 = C_1 V_{in}(nT), while the summing junction remains at and C_2 holds the prior integrated value. In \phi_2 (integration), the switch redirects C_1's reference terminal to , dumping Q_1 onto the summing node; the op-amp adjusts V_{out} to maintain , accumulating charge on C_2 such that the output updates as V_{out}((n+1)T) = V_{out}(nT) - (C_1 / C_2) V_{in}(nT). This process simulates continuous-time in discrete steps, with the z-transform yielding the ideal H(z) = -\frac{C_1}{C_2} \frac{z^{-1}}{1 - z^{-1}}, derived from and summation over clock cycles. However, finite op-amp open-loop gain A limits virtual ground precision, causing incomplete charge transfer and leakage; the effective transfer function approximates H(z) \approx -\frac{C_1}{C_2} \frac{z^{-1}}{1 - (1 - \beta) z^{-1}}, where \beta \approx 1/(1 + A) introduces a pole shift and gain error proportional to $1/A. This non-ideality reduces integration accuracy, particularly at high frequencies or with low-gain op-amps. A primary limitation arises from parasitic capacitance C_p at the summing junction, which parallels C_2 during both phases, effectively increasing the integrating capacitance to C_2 + C_p; the gain thus alters to (C_1 / C_2) / (1 + C_p / C_2), introducing a fixed error that degrades precision unless C_p \ll C_2. This sensitivity stems from the summing node's exposure, where junction or interconnect parasitics (typically 0.1–1 pF in early ) can shift ratios by 1–10%, limiting applications to low-precision scenarios. This design, notably developed by Bedrich Hosticka in 1977, emerged in the mid-1970s as one of the first practical switched-capacitor building blocks, offering simplicity for analog filters but constrained by parasitic effects in integrated realizations. Seminal work in 1978 demonstrated its use in filters, highlighting the trade-off between ease of implementation and precision limitations.

Parasitic-Insensitive Integrators

The parasitic-insensitive switched-capacitor addresses the limitations of earlier designs by employing a that isolates parasitic capacitances through strategic switching and bottom-plate sampling. In this circuit, an input C_1 is connected in series with the signal source via a switch to the of an , while a feedback C_2 is connected in parallel across the amplifier's input and output. Bottom-plate sampling is achieved by first connecting the bottom plate of C_1 to the (sampling the input voltage on the top plate) before switching the top plate to the input, which dumps parasitic charges associated with the bottom plate without affecting the stored signal charge. Operation occurs over two non-overlapping clock phases using four switches controlled by complementary clocks \phi_1 and \phi_2. During \phi_1, switches connect C_1's bottom plate to and top plate to the input, charging C_1 to the input voltage V_{in}; simultaneously, C_2 is reset if needed. In the \phi_2 phase, C_1's bottom plate connects to the output and top plate to , transferring charge to C_2 and updating the output voltage. This configuration supports both non-inverting (direct charge transfer) and inverting (reversing plate connections during transfer) phases, which cancel the influence of parasitics by ensuring they are sampled and discharged at the virtual ground node where voltage is fixed. The ideal discrete-time transfer function remains H(z) = \frac{C_1}{C_2} \frac{z^{-1}}{1 - z^{-1}}, equivalent to a lossless , but with gain accuracy independent of parasitic capacitances at the due to their negligible voltage swing. This provides higher precision in monolithic integrated circuits by minimizing errors from and wiring parasitics, rendering it essential for constructing biquadratic filter sections (biquads) in switched-capacitor s; it was introduced in the late through concurrent developments by multiple research groups. Unlike the preceding parasitic-sensitive , it ensures robust performance without requiring parasitic cancellation techniques. However, precise clock non-overlap is required to avoid direct charge paths that could cause injection errors and degrade .

Switched-Capacitor Multipliers

Switched-capacitor multipliers scale an analog input signal by a coefficient through precise control of charge transfer between capacitors using clocked switches. These circuits underpin multiplying digital-to-analog converters (MDACs), which integrate -to-analog with signal scaling via charge redistribution, enabling efficient implementation in integrated circuits. The technique leverages the inherent matching of on-chip capacitors to achieve high accuracy without relying on precise resistors, making it ideal for processes. A fundamental switched-capacitor multiplier employs two capacitors switched according to digital inputs to realize binary scaling of the analog signal. For instance, with equal capacitors C_s (sampling) and C_f (feedback), the circuit can multiply the input by 0 or 1: in the sampling phase, both capacitors charge to V_in; in the transfer phase, C_s connects fully to the feedback node for multiplication by 1, or is isolated for 0. More generally, a weighted capacitor array extends this to multi-bit scaling, where capacitors of values C, 2C, 4C, ..., 2^{n-1}C sample V_in collectively. Switches then selectively connect array capacitors to transfer charge proportional to the digital bits b_i, with the total transferred charge Q = V_in \sum b_i \cdot 2^{i-1} C determining the output across the feedback capacitor C_f. The output voltage derives from charge balance at the of an : V_{out} = -\left( \sum_{i=1}^{n} b_i 2^{-i} \right) V_{in} assuming C_f equals the total array scaled for a maximum of -1, with b_i as the binary bits (MSB i=1). This arises from conservation of charge: the net charge injected into C_f equals the sum of sampled charges from activated capacitors, divided by C_f to yield the voltage. The negative sign reflects the inverting configuration, common in -based realizations. Operation occurs in two non-overlapping clock phases to ensure complete and minimize errors from switch charge injection. In phase φ_1 (sampling), the array's bottom plates connect to V_in while top plates connect to the , charging each capacitor to V_in. The feedback capacitor resets or initializes. In phase φ_2 (transfer/hold), top plates remain at the inverting input (), and bottom plates route based on bits: to ground for b_i = 1 (transferring full C_i V_in charge) or to V_in for b_i = 0 (zero net transfer). This redistributes charge selectively, producing the scaled output. The approach parallels the R-2R ladder in resistive DACs but operates in the charge domain, offering lower power and better integrability. A representative 4-bit MDAC illustrates the structure: the comprises capacitors C_1 = C, C_2 = 2C, C_3 = 4C, C_4 = 8C (total 15C), paired with C_f = 15C for a full-scale of -15/16 ≈ -1. During φ_1, all bottom plates sample V_in (0 to V_fs), top plates at via feedback. Switches (controlled by non-overlapping clocks and digital bits) ensure bottom-plate sampling to reduce nonlinearity. In φ_2, bits b_4 (MSB) to b_1 (LSB) dictate: for b_i = 1, switch C_i bottom to ground; for 0, to V_in. yields V_out = -V_in (b_4/2 + b_3/4 + b_2/8 + b_1/16), approximating the desired fractional scaling. Timing diagrams show φ_1 high for ~1/2 clock period, φ_2 delayed slightly to avoid overlap, with dictated by amplifier bandwidth. This configuration achieves precise multiplication, with capacitor matching ensuring <1 LSB error for n=4.

Applications

Filters and Signal Processing

Switched-capacitor (SC) circuits facilitate the implementation of discrete-time filters for analog signal processing by employing cascaded integrators to realize various filter types, including low-pass, high-pass, and bandpass configurations. These filters often utilize biquad sections, which consist of two integrators combined with feedback and feedforward paths to achieve second-order responses, allowing for precise control over pole and zero placement in the z-domain. The design of SC filters typically begins with established s-domain prototypes, such as Butterworth or Chebyshev active-RC filters, which are transformed to the z-domain using the bilinear transform to preserve frequency response characteristics while accounting for the discrete-time nature of the system. This method maps the continuous-time frequency axis to the discrete one via s = \frac{2}{T} \frac{1 - z^{-1}}{1 + z^{-1}}, where T is the sampling period determined by the clock frequency f_{\text{clk}} = 1/T. To avoid frequency warping, prewarping is applied to critical frequencies like the cutoff \omega_c, ensuring the desired response; the prewarped analog cutoff frequency is given by f_{c,\text{analog}} = \frac{f_{\text{clk}}}{\pi} \tan\left( \pi \frac{f_c}{f_{\text{clk}}} \right), where f_c is the desired digital cutoff frequency. A representative example is a second-order low-pass SC filter realized with a biquad structure, having the z-domain transfer function H(z) = \frac{(C_1 / C_2)^2}{1 - 2 \cos(\theta) z^{-1} + z^{-2}}, where C_1 and C_2 are feedback and input capacitors, respectively, setting the DC gain as (C_1 / C_2)^2. The coefficients are calculated based on the desired cutoff frequency f_c and clock rate f_{\text{clk}}, with the pole angle \theta determined approximately as \theta \approx 2\pi f_c / f_{\text{clk}} for low f_c / f_{\text{clk}} ratios, or more precisely using prewarping: \theta = 2 \arctan(\pi f_c / f_{\text{clk}}) to align the response with the analog prototype. This structure leverages parasitic-insensitive integrators for robust performance. SC filters offer key advantages, including tunability of the cutoff frequency by adjusting the clock rate without altering passive components, and immunity to resistor drift or tolerance issues since no physical resistors are used—capacitor ratios provide high precision (typically 0.1%). These attributes make SC filters prevalent in integrated circuits for audio processing, such as anti-aliasing in codecs, and telecommunications, including channel selection in modems. For higher-order filters, stray-insensitive ladder implementations are employed, simulating LC ladder prototypes with SC integrators and bilinear resistors to minimize sensitivity to parasitic capacitances, enabling sharp roll-off and low dynamic range requirements in cascaded designs.

Data Converters

Switched-capacitor techniques play a pivotal role in data converters by enabling precise charge-based sampling, quantization, and reconstruction of analog signals in integrated circuits, particularly for applications demanding high linearity and low power. These circuits use capacitor arrays switched by clocked transistors to simulate resistors and perform discrete-time operations, converting between analog and digital domains with minimal analog components. In analog-to-digital converters (ADCs), switched capacitors facilitate charge redistribution for bit trials, while in digital-to-analog converters (DACs), they control charge steering for accurate output generation. Successive approximation register (SAR) ADCs rely on a switched-capacitor array to implement comparator-based bit decisions via charge redistribution. During the sampling phase, the analog input charges a binary-weighted capacitor array, typically totaling 2C for an N-bit converter where capacitors are sized as C, C/2, ..., C/2^{N-1}. In the conversion phase, the SAR logic starts with the most significant bit (MSB), switching the largest capacitor to the reference voltage V_REF if the comparator indicates the need, effectively redistributing charge across the array to test the bit value; this process repeats sequentially for each bit in one cycle per bit, completing the quantization in N clock cycles. This architecture achieves resolutions up to 18 bits with low power, as the charge sharing minimizes the need for continuous reference current. Delta-sigma modulators incorporate switched-capacitor integrators in an oversampled feedback loop with an internal DAC to perform high-resolution conversion through noise shaping. The integrator, often a parasitic-insensitive type, accumulates the error between the input signal and the quantized feedback, shifting quantization noise to higher frequencies outside the signal band for subsequent digital filtering. For a first-order modulator, the peak signal-to-noise ratio (SNR) is approximated by \text{SNR} = 6.02N + 1.76 + 10 \log_{10}(2L + 1) where N is the quantizer bits and L is the oversampling ratio, enabling effective resolutions exceeding 16 bits at moderate clock rates. Higher-order topologies extend this by further suppressing in-band noise, commonly used in audio and sensor interfaces. Pipeline ADCs divide the conversion into cascaded stages, each employing a switched-capacitor multiplying DAC (MDAC) for sub-quantization and residue amplification. In a typical 1.5-bit-per-stage design, the MDAC samples the input, subtracts the coarse DAC output based on a flash quantizer decision, and amplifies the residue by a factor of 2 using a closed-loop op-amp configuration with sampling capacitor C_s and feedback capacitor C_f. The inter-stage gain is given by A = 2 \times \frac{C_s}{C_f}, ensuring the residue remains within the next stage's range while providing speed advantages through pipelining, with overall resolutions up to 14 bits at sampling rates beyond 1 GS/s in deep-submicron processes. Redundancy in bit decisions tolerates comparator offsets up to 0.5 LSB. Switched-capacitor DAC implementations offer advantages over traditional current-steering designs by achieving high linearity through charge-based operation and integrated calibration. In these DACs, digital codes activate switches to connect binary- or segmented-capacitor arrays to a reference, steering discrete charge packets to the output for voltage or current generation; predistortion and mismatch correction via auxiliary ADCs mitigate capacitor variations, yielding integral nonlinearity (INL) below 1 LSB. For high-speed applications, pipelined switched-capacitor cores with open-loop drivers maintain spurious-free dynamic range (SFDR) above 53 dB up to 400 MHz bandwidth in 90-nm CMOS, with switched-capacitor multipliers occasionally integrated for scaled reference generation in segmented architectures. Modern switched-capacitor data converters in advanced CMOS nodes exhibit strong performance in effective number of bits (ENOB) and power efficiency, driven by optimized switching schemes and reduced parasitics. A 65-nm SAR ADC delivers 7.76 ENOB at 1 MS/s with 5.75 µW power consumption, achieving a Walden figure-of-merit (FoM) of 26.52 fJ/conversion-step suitable for biomedical sensing. Similarly, 65-nm SAR variants reach 9.2 ENOB at 90 MS/s with sub-1 mW power. Recent delta-sigma designs in advanced nodes continue to improve ENOB and FoM, underscoring scalability for 5G and IoT deployments.

Discrete-Time Systems

Switched-capacitor circuits play a crucial role in sampled-data systems by enabling the implementation of discrete-time operations that mimic continuous-time behaviors through charge transfer mechanisms timed by a clock signal. In these systems, Z-domain modeling is employed to analyze stability and performance, where the z-transform represents the discrete-time response of the circuit. This approach allows designers to map continuous-time transfer functions to their discrete equivalents using techniques such as the bilinear transform, which preserves key frequency characteristics while accounting for the sampling process. The bilinear z-transform facilitates the derivation of difference equations from continuous-time prototypes, ensuring that the discrete-time system approximates the desired analog response with minimal warping at low frequencies relative to the sampling rate. For instance, a continuous-time integrator can be realized as a discrete accumulator in the z-domain, with the transfer function H(z) = \frac{1}{1 - z^{-1}}, enabling stability analysis via pole placement in the z-plane. These models are essential for predicting system behavior in feedback loops, where the location of poles outside the unit circle indicates instability. A representative example of switched-capacitor application in discrete-time control is the realization of proportional-integral-derivative (PID) controllers, where the analog PID transfer function is discretized to form a difference equation implemented via charge-sharing networks. In such designs, the proportional term is achieved through direct voltage scaling with capacitors, the integral via accumulating , and the derivative approximated by differencing samples over clock periods, all synchronized to the system clock for real-time control. This approach has been demonstrated in power electronics, such as closed-loop , where the PID ensures stable voltage regulation under varying loads. Integration of switched-capacitor circuits with digital logic in mixed-signal systems-on-chips (SoCs) requires precise clock synchronization to align analog sampling phases with digital processing cycles, preventing timing mismatches that could degrade signal integrity. Anti-aliasing filters, often implemented as switched-capacitor low-pass stages ahead of sampling, are critical to suppress frequencies above the Nyquist rate, typically set at half the clock frequency, ensuring faithful representation of the input signal in the discrete domain. In applications like sensor interfaces, switched-capacitor discrete-time systems provide high-resolution digitization of low-level signals from transducers, such as in capacitive or piezoelectric sensors, by employing sampled-data amplification and conditioning. Similarly, in communications, adaptive equalizers utilize switched-capacitor structures to implement discrete-time finite impulse response (FIR) or infinite impulse response (IIR) filters that dynamically adjust to channel impairments, enhancing data recovery in receivers. These systems leverage the clock-driven nature of switched capacitors for seamless interfacing with digital signal processors. A key challenge in these discrete-time systems is clock jitter, which introduces timing uncertainty in sampling instants and manifests as noise, limiting the signal-to-noise ratio (SNR). The impact is quantified by the formula \text{SNR}_\text{jitter} = -20 \log_{10} (2 \pi f_\text{in} \sigma_\text{jitter}), where f_\text{in} is the input signal frequency and \sigma_\text{jitter} is the root-mean-square jitter in seconds; for a full-scale sine wave, this degradation becomes dominant at high f_\text{in}, often constraining SNR to below 80 dB for \sigma_\text{jitter} > 1 ps at GHz inputs. Mitigation involves low-jitter clock sources and jitter-shaping techniques in oversampled architectures.

Power Management

Switched-capacitor (SC) circuits are widely used in DC-DC converters for efficient without inductors, making them ideal for integration in low- to medium-power applications. These converters operate by charging and discharging capacitors in specific topologies, such as series-parallel, Dickson, or , to achieve step-up or step-down conversion ratios. The equivalent resistance of the switched capacitors determines efficiency, with charge transfer controlled by clock phases to minimize losses. Key advantages include small size, low electromagnetic interference due to the absence of inductors, and tunability via clock frequency or capacitance ratios. They are particularly suited for portable electronics, biomedical implants, and RF power amplifiers, where space constraints and low power (typically <1 W) are critical. Efficiencies can exceed 90% at optimal loads, though ripple and regulation require careful design with multi-phase operation or hybrid inductive assistance in advanced implementations. As of 2025, SC DC-DC converters in deep-submicron CMOS support voltages from sub-1 V to multi-volt outputs, enabling efficient power delivery in IoT devices and system-on-chips.

Analysis and Design Considerations

Equivalent Circuit Modeling

Switched-capacitor (SC) networks, being discrete-time systems, are often modeled as equivalent continuous-time circuits to leverage familiar analog design tools and intuition. This modeling approach facilitates the analysis and synthesis of SC circuits by transforming their z-domain descriptions into s-domain equivalents, allowing designers to approximate the behavior of resistors, integrators, and filters using continuous-time RC networks. The primary methods involve domain transformations and graph-based or matrix formulations for charge conservation. A key technique for emulating continuous-time filters in SC designs is the bilinear transform, which maps the z-domain transfer function to the s-domain via the substitution s = \frac{2}{T} \frac{1 - z^{-1}}{1 + z^{-1}}, where T is the sampling period. This transformation preserves the stability of the discrete-time system and warps the frequency axis to match the continuous-time response, enabling direct synthesis from active-RC prototypes. For instance, in filter design, the bilinear method ensures that the SC circuit's frequency response closely approximates the desired continuous-time characteristic up to the Nyquist frequency, with minimal aliasing when prewarping critical frequencies. Network analysis of SC circuits employs charge-flow graphs, which are analogous to but track charge transfers between capacitors during switching phases. These graphs represent nodes as capacitor voltages and branches as charge injections or transfers, facilitating the derivation of discrete-time transfer functions through or similar reduction techniques. For more complex topologies, matrix methods such as nodal charge analysis formulate the system as a set of linear equations \mathbf{Q} \mathbf{v}(n) = \mathbf{C} \mathbf{v}(n-1) + \mathbf{b} u(n), where \mathbf{Q} and \mathbf{C} are capacitance matrices, \mathbf{v} are node voltages, and u is the input; solving via yields the overall transfer function. This approach is particularly efficient for multi-stage filters, as it systematically accounts for inter-stage charge redistribution without explicit time-domain simulation. A representative example is modeling a first-order SC low-pass filter as an equivalent RC circuit. Consider a basic integrator-based topology where a sampling capacitor C_s transfers charge to an integrating capacitor C_i at clock period T; the equivalent resistance is R_{eq} = \frac{T}{C_s}, yielding a continuous-time pole at \frac{1}{R_{eq} C_i} = \frac{C_s}{T C_i}. The discrete transfer function H(z) = \frac{k}{1 - z^{-1}} (with k = \frac{C_s}{C_i}) maps via bilinear transform to an s-domain low-pass response H(s) \approx \frac{\omega_c}{s + \omega_c}, where the cutoff \omega_c \approx \frac{2}{T} \tan\left(\frac{\pi f_c}{f_s}\right) and f_s = 1/T is the sampling frequency, confirming the RC equivalence for design purposes. For practical verification, simulation tools like SPICE support SC modeling through behavioral sources, such as voltage-controlled current sources (VCCS) to emulate charge transfers: a switch is represented by a time-dependent resistor or multiplier, while capacitors use piecewise-linear models alternating between open and short states per clock phase. This allows transient and small-signal AC analyses of the equivalent continuous-time circuit, capturing frequency-dependent behaviors without full switch-level detail. The frequency response of SC networks is derived from their discrete transfer functions and plotted as magnitude and phase versus normalized frequency f/f_s. Using the bilinear mapping, the s-domain equivalent provides Bode plots that align with continuous-time expectations, highlighting passband ripple and stopband attenuation; for the first-order example, the magnitude rolls off at -20 /decade beyond the warped cutoff, with phase shifting from 0° to -90°. These plots guide capacitor sizing to meet specifications like unity gain at DC and attenuation at higher frequencies.

Non-Ideal Effects and Parasitics

In switched-capacitor circuits, parasitic capacitances arise primarily from the fabrication process, connecting the top and bottom plates of capacitors to the substrate or adjacent structures. The bottom-plate parasitic capacitance is typically larger (around 10-20% of the main capacitor value in standard CMOS processes) than the top-plate one (1-2%), leading to incomplete charge transfer during switching phases. This results in signal-dependent voltage errors and reduced efficiency, particularly in integrators and multipliers, where the error can be modeled as a fraction of the input voltage proportional to the parasitic-to-main capacitance ratio. For instance, in DC-DC converters, energy loss due to these parasitics is given by \frac{1}{2} C_{par} (V_{in} - V_{out})^2 per cycle, degrading overall power efficiency by up to 10-15% without mitigation. To mitigate these effects, bottom-plate sampling techniques connect the bottom plate to ground before switching the top plate to the input signal, minimizing the impact of larger bottom-plate parasitics on the sampled voltage since the parasitic path is isolated from the signal. Top-plate parasitics, being smaller and more uniform, have less influence but can still cause offset in differential circuits. Parasitic-insensitive topologies, such as those using fully differential structures or charge recycling schemes, further reduce losses by redirecting trapped charge back into the circuit; for example, a general charge recycling method in series-parallel converters achieves up to 52% reduction in parasitic energy loss, improving efficiency by 7% at 240 mW load power. These approaches prioritize layout strategies like common-centroid capacitor arrays to balance parasitic influences across matched capacitors. Thermal noise, specifically kT/C noise, is another key non-ideal effect in SC circuits, originating from the random thermal motion of charge carriers during the sampling phase when a capacitor is connected to a resistive switch. The root-mean-square (RMS) noise voltage is given by v_n = \sqrt{\frac{kT}{C}}, where k is Boltzmann's constant, T is the absolute temperature, and C is the sampling capacitance; this noise is independent of the switch resistance and is sampled onto the capacitor, contributing to the overall noise floor. In precision applications, such as ADCs, kT/C noise limits the maximum signal amplitude and thus the SNR, often requiring capacitors larger than 1 pF to achieve >70 dB SNR at . Mitigation strategies include using larger capacitors (at the cost of area and power), correlated double sampling to cancel low-frequency noise components, or noise-shaping techniques in oversampled systems like sigma-delta modulators. Switch non-idealities, particularly in MOS transistors, introduce errors through finite on-resistance and charge injection during turn-off. The on-resistance R_{on} forms a time constant \tau = R_{on} \cdot C with the sampling capacitor C, causing incomplete settling within the clock half-period and resulting in exponential voltage errors, typically 0.1-1% for high-speed circuits if \tau exceeds 10% of the phase duration. This settling error is more pronounced in high-frequency applications, limiting bandwidth and introducing distortion. Charge injection occurs as channel charge redistributes to the source/drain and gate capacitances upon switch opening, producing an output voltage offset \Delta V = \frac{C_{gate}}{C} \cdot V_{GS}, where C_{gate} is the gate capacitance and V_{GS} is the gate-source voltage; this can contribute up to 5-10 mV offset in 10-bit precision circuits, with signal dependency arising from channel charge variation. Mitigation for switch errors includes using transmission gates with complementary NMOS/PMOS pairs to average charge injection and reduce signal dependency, or bottom-plate sampling to decouple injection from the input. Bootstrapped switches maintain constant V_{GS} to minimize variation, achieving sub-1 mV errors in analog-to-digital converters. Dummy switches can also cancel injection by symmetric , though they increase area by 20-30%. Operational amplifier limitations, including finite DC gain A_{op} and , cause integrator leakage by allowing charge to "leak" through the , introducing an term \varepsilon = 1 / A_{op} that scales the integrated output by $1 - \varepsilon per cycle. For typical op-amps with A_{op} = 60-80 dB, this leakage accumulates in recursive filters, reducing Q-factor by 10-20% and causing deviations up to 5% in high-order designs. Finite further exacerbates , with the effective at GBW / (2\pi), where GBW is the gain-bandwidth product, limiting to clock frequencies below 1/10 of GBW for <0.1% . In low-voltage processes, these effects worsen due to reduced A_{op} (often <60 dB), impacting dynamic range by 3-6 dB in sigma-delta modulators. Compensation involves stabilization to boost effective or using higher-GBW amplifiers, though at cost; alternatively, leakage-compensating switched-capacitor structures add auxiliary paths to restore charge, doubling effective in some integrators. Clock feedthrough refers to capacitive coupling from clock signals through switch gate-drain overlap capacitances, injecting voltage spikes \Delta V_{ft} \approx \frac{C_{gd}}{C + C_{gd}} V_{clk} onto the , where C_{gd} is the gate-drain and V_{clk} is the clock ; this adds and nonlinearity, degrading signal-to-noise-and-distortion ratio (SNDR) by 2-5 dB in 12-bit systems. Clock introduces , modulating sampling instants and causing noise floor rise proportional to \sigma_t^2 f_{in}^2 / 12, where \sigma_t is jitter standard deviation and f_{in} is input frequency, limiting to 70-80 dB for 1 ps jitter at 100 MHz sampling. These effects are critical in high-speed converters, where feedthrough spurs can fold into the band, reducing spurious-free (SFDR) by up to 10 dB. Mitigation strategies include low-jitter clock generation using phase-locked loops and signaling to reject common-mode feedthrough; switch sizing trades off C_{gd} against R_{on}, optimizing for minimal total error. In pipelined ADCs, dithering randomizes effects, preserving above 75 dB. Design trade-offs in switched- circuits center on capacitor array matching, requiring <0.1% accuracy for 12-bit resolution, achieved via unit-cell layouts like common-centroid to minimize gradient errors, though increasing area by 20-50%. Mismatch causes (INL) up to 2-3 LSB without correction. Modern employ digital calibration techniques, such as background least-mean-square algorithms measuring capacitor ratios via embedded test structures, correcting mismatches to <0.05% and extending by 6-10 dB in SAR ADCs. These methods balance power (adding 10-20% overhead) and speed, essential for sub-1 V processes where parasitics scale unfavorably.