RF Capacitor Selection | SRF, ESR & Package Guide

What Determines RF Capacitor Selection?

RF capacitor selection is driven by three electrical parameters that barely matter at low frequencies but dominate above a few hundred MHz: self-resonant frequency (SRF), equivalent series resistance (ESR), and equivalent series inductance (ESL). A capacitor that measures 100 pF on an LCR bridge at 1 MHz may look purely inductive at 2.4 GHz if its SRF sits at 900 MHz. Choosing the right capacitor means matching these parasitic-limited parameters to your operating frequency, circuit function (DC block, bypass, matching, or coupling), and PCB pad geometry.

The Equivalent Circuit: Why a Capacitor Stops Being a Capacitor

Every real capacitor is a series RLC network. The nominal capacitance C sits in series with an equivalent series inductance (ESL, typically 0.2–0.8 nH for an 0402 MLCC) and an equivalent series resistance (ESR, from a few milliohms to several ohms depending on dielectric). In parallel with this series branch sits an insulation resistance that matters only at DC. At the self-resonant frequency the series L and C cancel, leaving only ESR—this is the frequency where impedance is lowest and the part transitions from capacitive to inductive behavior.

The SRF is given by:

SRF = 1 / (2π √(C × ESL))

For a 100 pF 0402 capacitor with 0.4 nH ESL, SRF ≈ 796 MHz. Above that frequency the part looks like an inductor whose reactance rises at +20 dB/decade. This is why blindly specifying “100 pF” without checking the SRF against your operating band is the single most common RF capacitor mistake.

Key Parameters for RF Capacitor Selection

Self-Resonant Frequency (SRF)

As a rule of thumb, a capacitor used for DC blocking or coupling should have an SRF at or above the operating frequency, so the part still presents low impedance. A capacitor used in a matching network, by contrast, must operate well below its SRF so its impedance is predictably capacitive and the value tracks what the simulation expects. A safe guideline: keep the operating frequency below roughly 0.7 × SRF when you need a predictable capacitance value, or right at or above SRF when you simply need low series impedance (bypass and DC-block roles).

Equivalent Series Resistance (ESR)

ESR determines the minimum achievable impedance at SRF. For a bypass capacitor feeding an LNA, lower ESR means a cleaner AC ground. Murata GJM-series 0402 MLCCs aimed at RF achieve ESR below 0.15 Ω at SRF, while general-purpose X7R parts of the same size may exceed 1 Ω. ESR also sets the capacitor’s quality factor:

Q = 1 / (2π f C × ESR)

At 2.4 GHz, a 1 pF NP0/C0G cap with 0.3 Ω ESR yields Q ≈ 221—adequate for most matching networks. An X7R cap of the same value might show Q below 30, which adds loss to filters and degrades noise figure.

Equivalent Series Inductance (ESL)

ESL is dominated by the internal electrode geometry and the external pad connection. Smaller packages have shorter current paths and lower ESL: an 0201 MLCC typically shows 0.15–0.3 nH, while an 0805 part may reach 0.7–1.0 nH. Reverse-geometry (low-inductance) packages like the Murata LLA or AVX LICC series flip the terminations to the long side, cutting ESL roughly in half for the same footprint area. When designing RF bypass capacitor placement, ESL is as important as capacitance value because it determines how high in frequency the bypass remains effective.

Dielectric Type

For any circuit where the capacitance value must be predictable—matching networks, resonators, filters—use NP0/C0G or thin-film. X7R is acceptable only for bypass and bulk decoupling where the exact value matters less than having “enough” capacitance at the frequency of interest. This distinction between high-frequency and high-speed design disciplines carries into every passive selection decision.

Choosing Capacitor Value and Package by Frequency Band

The table below gives practical starting points. All SRF values are approximate and vary by vendor; always verify against the manufacturer’s S-parameter data.

DC Blocking Capacitors: Selection Logic

A DC block must pass RF with minimal insertion loss while stopping DC bias. The key trade-off: too large a capacitance lowers the SRF and may push the part into its inductive region at your signal frequency; too small a value raises the capacitive reactance and reflects signal energy.

For a 50 Ω system at frequency f, the insertion loss from a series capacitor (ignoring ESR) is approximately:

IL ≈ 20 log₁₀ (1 + (1 / (4π f C × 50)²))^0.5

At 2.4 GHz, a 10 pF cap contributes about 0.004 dB of mismatch loss—negligible. A 0.5 pF cap at the same frequency contributes roughly 0.8 dB. But the 10 pF part in an 0402 package may have an SRF of only 2.8 GHz, putting the operating point dangerously close to resonance where the impedance is unpredictable and dominated by ESR. The 0.5 pF part might have an SRF above 15 GHz, but its reactance at 2.4 GHz is about 133 Ω—too high for a clean pass-through.

The sweet spot for a 2.4 GHz DC block in 0402 NP0 is typically 3.3–6.8 pF: low enough reactance for negligible mismatch, with SRF safely above 5 GHz.

Bypass and Decoupling: Multi-Cap Strategy

RF amplifier supply pins need bypassing from near-DC up to well above the operating frequency. A single capacitor cannot cover that range because every cap becomes inductive above its SRF. The classic approach stacks three to four capacitors in parallel, each covering a different frequency decade:

• Bulk: 10 µF X5R 0805—effective from DC to ~1 MHz.
• Mid-range: 100 nF X7R 0402—covers 1 MHz to ~100 MHz.
• RF: 100 pF NP0 0402—covers 100 MHz to ~2 GHz.
• Microwave: 10 pF NP0 0201—covers 1 GHz to ~8 GHz.

Place the smallest-value cap closest to the pin. Each additional cap in the chain should be physically farther from the IC, because the trace inductance to a distant bulk cap is irrelevant at GHz frequencies where only the nearest small cap matters. The anti-resonance dip between adjacent caps (where one is inductive and the next is capacitive and they form a parallel resonant tank) can create a high-impedance notch. Adding a small series resistor (1–10 Ω) in the path of the mid-range cap damps this resonance at the cost of slightly worse low-frequency bypassing—a worthwhile trade in most RF front-end circuits.

Matching Network Capacitors: Q, Tolerance and Stability

In an L-section or pi-section matching network, the capacitor’s actual impedance at the design frequency must match the simulation. That requires:

1. Operating well below SRF—at least 0.7 × SRF, preferably 0.5 × SRF—so the impedance is dominated by capacitance, not parasitics.
2. Tight tolerance: ±0.1 pF or ±5 %, whichever is tighter. At 1 pF, a ±0.25 pF tolerance means the actual value could be anywhere from 0.75 pF to 1.25 pF—a 25 % swing that shifts the match point significantly.
3. High Q: Every ohm of ESR the signal sees adds insertion loss. For a power amplifier output match at 3.5 GHz handling 2 W, even 0.3 dB of extra loss in the matching network wastes 14 % of the output power as heat and degrades PAE.
4. Stable dielectric: NP0/C0G is mandatory. X7R capacitance shifts with applied voltage, temperature, and aging, making it useless for repeatable matching.

For simulations, never rely on an ideal capacitor model above 500 MHz. Download the vendor’s S-parameter file (Murata provides these through SimSurfing, and TDK through their SEAT tool) and import it into your RF EDA environment. The difference between an ideal 1.5 pF cap and the real S2P model at 5 GHz can easily be 0.3 pF equivalent—enough to shift a match by several dB of return loss.

PCB Pad and Layout Effects on Capacitor Performance

The capacitor itself is only part of the story. The PCB landing pad, via to ground (for shunt caps), and trace transition all add parasitic inductance and capacitance that shift the effective SRF and impedance.

• Pad size: Oversized pads add fringing capacitance (0.02–0.1 pF on typical FR-4, more on high-Dk laminates). Use pads matched to the vendor’s recommended footprint.
• Via inductance: A single 0.3 mm (12 mil) via through 1.6 mm FR-4 adds roughly 0.7–1.0 nH. For a shunt bypass cap, this inductance is in series with the cap to ground and directly raises the effective ESL. Use multiple vias, or a via-in-pad design with backfill, to cut this below 0.3 nH.
• Ground plane proximity: The RF PCB thickness and dielectric constant of the substrate between the pad and the nearest ground plane affect the parasitic capacitance to ground. On a 5 mil (0.127 mm) Rogers RO4350B substrate (Dk 3.48), a 0402 pad pair adds roughly 0.08 pF of stray capacitance—small, but comparable to the component value when you are working with sub-pF caps at X-band.
• Orientation: Align series DC-block caps so the current path is collinear with the transmission line. A 90° jog at the cap pad creates a discontinuity that radiates and reflects, especially above 10 GHz.

When your design reaches the prototype stage and measured S-parameters diverge from simulation, these layout parasitics are usually the first place to investigate. A structured RF prototype debug workflow will catch pad-related shifts quickly.

Vendor Selection and Data You Need

Three vendors dominate RF MLCC supply: Murata (GJM, GRM series), TDK (CGA, CGJ series), and Johanson Technology (R-series, S-series). AVX (Kyocera) and Presidio Components serve niche high-reliability and space-grade markets.

When evaluating a part, demand these data items from the vendor:

• S-parameter files (.s2p): Measured on a known reference board, de-embedded to the component plane. This is non-negotiable above 1 GHz.
• SRF specification: Listed per capacitance value and package, not just a generic “typical” number for the series.
• ESR and Q vs. frequency curves: Ideally measured, not just modeled.
• Voltage and temperature coefficients: Especially for Class II dielectrics, verify the actual capacitance at your DC bias and operating temperature, not just the zero-bias, 25 °C nominal.

The IPC passive-component standards (IPC-FC-234 and related documents) provide land-pattern guidance that ensures consistent parasitic behavior across assembly lots.

Common Mistakes in RF Capacitor Selection

1. Using the LCR-bridge value at GHz frequencies. A 100 pF cap measured at 1 MHz is not 100 pF at 5 GHz. Use S-parameter data.
2. Ignoring voltage derating of X7R. A “100 nF” X7R cap under 70 % of rated voltage may actually present only 30–50 nF. For RF bypass this is usually acceptable; for a filter or matching network it is catastrophic.
3. Specifying 0201 or 01005 without checking assembly capability. Smaller packages improve RF performance but demand tighter pick-and-place accuracy, solder-paste aperture control, and reflow profiling. Factor this into your RF PCB cost optimization analysis.
4. Neglecting the anti-resonance between parallel bypass caps. The impedance peak between two adjacent-value caps can be 10–100× higher than either cap alone at that frequency. Simulate the bypass network, not just individual caps.
5. Assuming all NP0 caps are equal. Different vendors use different internal electrode counts and geometries. Two “1 pF NP0 0402” parts from different manufacturers can differ by 0.15 nH in ESL and 1 GHz in SRF.

Quick-Reference Selection Checklist

• Define the circuit function: DC block, bypass, matching, or filter.
• Identify the operating frequency and required impedance behavior (capacitive, resonant, or “just low”).
• Pick dielectric: NP0/C0G for signal path; X7R/X5R only for bulk bypass.
• Choose the smallest practical package for your assembly process to minimize ESL.
• Verify SRF > f_operating for DC block/bypass; SRF > 1.4 × f_operating for matching.
• Download and simulate with the vendor’s S2P model, not an ideal cap.
• Account for PCB pad parasitics, via inductance, and substrate material properties.
• Prototype, measure, and compare to simulation.

Frequently Asked Questions

What capacitor value should I use for a DC block at 5.8 GHz?

A 0.5–1.5 pF NP0 capacitor in an 0201 package typically works well. At 1 pF the reactance is about 27 Ω at 5.8 GHz—enough to pass signal in a 50 Ω system with under 0.3 dB mismatch loss—and the SRF for an 0201 1 pF NP0 is usually above 12 GHz, keeping the part safely capacitive.

Can I use X7R capacitors in RF matching networks?

No. X7R capacitance varies with applied DC voltage (up to –80 % at rated voltage), temperature (±15 %), and aging (–2.5 % per decade-hour). These shifts make the match unpredictable. Use NP0/C0G or thin-film capacitors for any impedance-sensitive application.

How do I find the SRF of a specific capacitor?

Check the vendor’s online simulation tool (Murata SimSurfing, TDK SEAT, or Johanson Technology S-parameter library). Look for the frequency where |Z| reaches its minimum in the impedance-vs-frequency plot, or where the phase crosses zero. Do not rely on a single “typical SRF” in a datasheet—measure or simulate with the actual S2P file.

Does the PCB substrate affect capacitor performance?

Yes. A higher-Dk substrate increases pad-to-ground parasitic capacitance, effectively adding capacitance in parallel with the component. On thin, high-Dk substrates (e.g., 5 mil RO3010, Dk 11.2), this stray capacitance can exceed 0.1 pF per pad—significant when the component value is 0.5 pF. Always include substrate effects in your EM simulation.

When should I switch from lumped capacitors to distributed elements?

As a practical boundary, lumped MLCCs become difficult to use predictably above about 20–25 GHz. Beyond that, distributed elements (open stubs, coupled lines, interdigital capacitors etched directly on the PCB) offer more repeatable performance because they eliminate solder-joint and package parasitics entirely. Thin-film integrated passive devices (IPDs) extend the lumped range to roughly 40–60 GHz for specialized applications.