Bulk vs Decoupling Capacitor: The Three-Tier Capacitor Strategy

Introduction

One of the most persistent sources of confusion in PCB design is the relationship between bulk capacitors and decoupling capacitors. Engineers new to hardware design often treat these as two alternative names for the same component, or assume that placing more capacitance in any form near an IC is always better. Neither assumption is correct.

Bulk capacitors and decoupling capacitors serve different purposes, operate in different frequency ranges, use different component types, and must be placed in different locations relative to the IC they serve. Understanding why each tier exists—and what physical limitations make the other tiers necessary—is the foundation of effective PCB power integrity design. This article explains the three-tier capacitor strategy from first principles, covering what each tier does, why it cannot be replaced by the others, and how to implement all three correctly in a real design.

1. Table of Contents

• Introduction
• The Core Problem: Why One Capacitor Cannot Do Everything
• The Three-Tier Capacitor Strategy
• Tier 1: Bulk Capacitance
• Tier 2: Mid-Frequency Decoupling
• Tier 3: High-Frequency Bypass
• Comparison Table: All Three Tiers at a Glance
• Why You Cannot Replace All Three Tiers with One Large Capacitor
• Anti-Resonance: The Hidden Risk of Mixing Tiers
• Real-World Examples by IC Type
• Implementation Checklist
• FAQ

---

The Core Problem: Why One Capacitor Cannot Do Everything

Every capacitor has a self-resonant frequency (SRF) above which it stops behaving as a capacitor and starts behaving as an inductor. This happens because every real capacitor has parasitic series inductance (ESL—Equivalent Series Inductance) from its internal electrodes, termination connections, and PCB traces. At the SRF, the capacitive reactance exactly cancels the inductive reactance, and the component presents its minimum impedance (equal to the ESR, Equivalent Series Resistance). Above the SRF, the inductive reactance dominates and impedance rises with frequency.

This means that a 100 μF electrolytic capacitor, which is excellent at handling slow transients (milliseconds), provides almost no useful decoupling at 100 MHz because its ESL (typically 10–30 nH) pushes its SRF well below 10 MHz. Conversely, a 100 pF ceramic capacitor is effective at 1 GHz but stores far too little charge to handle a 1 A load step lasting 1 μs.

No single capacitor type can provide low impedance across the full frequency range from DC to 1 GHz. The three-tier strategy solves this by using different capacitor types at different locations, each optimized for its own frequency range, with their contributions cascading to cover the complete spectrum of transient current demand. The fundamental ESR and ESL parameters that determine each tier's effective frequency range are explained in depth in the ESR and ESL in Capacitors guide.

---

The Three-Tier Capacitor Strategy

The three-tier strategy places capacitors at three distinct frequency ranges, using the right component type at the right location for each range:

• Tier 1 (Bulk): Large capacitors near the voltage regulator; handle slow, large transients; 10 Hz – 1 MHz
• Tier 2 (Mid-frequency decoupling): Medium MLCCs distributed near each IC; handle intermediate transients; 1 MHz – 100 MHz
• Tier 3 (High-frequency bypass): Small, low-ESL MLCCs immediately at IC power pins; handle fast transients; 100 MHz – 1 GHz+

Each tier hands off responsibility to the next as frequency increases. The voltage regulator handles DC and very-low-frequency variations. When the load changes faster than the regulator can respond, Tier 1 bulk capacitors supply the charge. When the transient is faster than the bulk capacitors can respond (limited by their ESL), Tier 2 MLCCs take over. When the transient is faster still, Tier 3 capacitors—placed within fractions of a millimeter of the power pin—provide the instantaneous charge.

---

Tier 1: Bulk Capacitance

Bulk capacitors store the large reservoirs of charge needed to sustain a load during transitions that are faster than the power supply regulator can respond to, but slower than MLCCs can handle. Their defining characteristics are large capacitance (10–1,000 μF or more) and relatively high ESL (5–30 nH), which limits their effectiveness to frequencies below approximately 1–5 MHz.

Component types used at Tier 1:

• Aluminum electrolytic capacitors: The traditional bulk capacitor; very high capacitance per dollar; polarized (observe polarity); ESR relatively high (0.05–1 Ω); limited operating life at high temperature; not suitable for very high ripple current applications without derating
• Solid polymer aluminum capacitors: Lower ESR than electrolytic (0.005–0.05 Ω); better high-frequency performance; longer life; higher cost; increasingly used in power electronics and server power delivery
• Large MLCCs (0805–1210, 10–100 μF): Non-polarized; excellent ESR; suitable for bulk decoupling in space-constrained designs; significant DC bias derating must be accounted for (a 100 μF / 6.3 V X5R MLCC may deliver only 30–50 μF at 3.3 V operating voltage). The DC bias effect is explained in the MLCC guide
• Tantalum capacitors: High capacitance density; lower ESR than aluminum electrolytic; polarized; failure mode (short-circuit) is more destructive than aluminum electrolytic; generally avoided in new designs where polymer or MLCC alternatives exist

Placement: Tier 1 capacitors are placed near the voltage regulator output and at each major power entry point on the board. They do not need to be immediately adjacent to individual ICs. A typical layout places 2–4 bulk capacitors within 20–40 mm of each power connector or regulator, with one or two additional bulk capacitors near large IC loads (FPGAs, processors) that draw large, slow load steps.

Sizing: A simplified formula for bulk capacitance sizing is C_bulk ≥ I_max × t_response / ΔV, where I_max is the maximum load current step, t_response is the regulator response time, and ΔV is the maximum allowable voltage droop. For a 5 A load step, a 100 μs regulator response time, and 50 mV allowable droop: C_bulk ≥ 5 × 100 × 10⁻⁶ / 0.05 = 10,000 μF. In practice this drives toward multiple paralleled large capacitors near the regulator output.

---

Tier 2: Mid-Frequency Decoupling

Mid-frequency decoupling capacitors handle the transient current demands from digital switching events that are too fast for bulk capacitors to supply (limited by their ESL) but not so fast that they require Tier 3's smallest packages. This is the most familiar tier to most PCB engineers: the 100 nF 0402 MLCC placed near each IC is the archetypal Tier 2 component.

Component types: X7R or X5R MLCC in 0402 or 0603 package; 100 nF to 10 μF; working voltage rated at 2×–3× the supply voltage to account for DC bias derating. For applications where piezoelectric noise is a concern (PLL supply, ADC reference), C0G/NP0 is used at this tier despite its lower capacitance density. The dielectric selection trade-offs are detailed in the X7R vs C0G vs X5R guide.

Placement: Within 1–5 mm of each IC power pin. One 100 nF capacitor per power supply pin is the typical starting specification; critical ICs may warrant 2–4 per power pin. The capacitors should be on the same board side as the IC, between the power plane via and the IC power pin, oriented with the long axis parallel to the current flow direction to minimize trace inductance.

Effective frequency range: An 0402 100 nF X7R MLCC has a typical SRF of 30–80 MHz. Below this frequency, it provides capacitive decoupling. Above this frequency, its impedance rises due to ESL dominance, and Tier 3 must take over.

---

Tier 3: High-Frequency Bypass

Tier 3 capacitors handle the fastest transients in the design: the nanosecond-scale switching of SerDes transmitters, DDR data bus switching, and high-speed logic transitions. At these speeds, the ESL of any capacitor placed more than 0.5 mm from the IC power pin dominates the impedance path. Tier 3 is about minimizing physical inductance more than maximizing capacitance.

Component types: C0G/NP0 or low-ESL X7R MLCC in 0201 package; 1–100 nF; voltage rating appropriate to the supply but otherwise as small (low-ESL) as possible. Some designs use specialized ultra-low-ESL MLCC variants (reverse geometry capacitors, where the terminations are on the long sides rather than the short ends, reducing the internal electrode loop inductance by 50–70%).

Placement: Immediately adjacent to or directly beneath the IC power pin. The ideal is via-in-pad: the Tier 3 capacitor sits on the back side of the PCB directly beneath the IC's power delivery via array, with the shortest possible current path. For ICs in BGA packages, Tier 3 capacitors are often placed in the escape routing area between BGA rows on the back side of the board.

Effective frequency range: A 0201 10 nF C0G MLCC has an SRF of 150–300 MHz, providing effective decoupling to several hundred MHz. For GHz-range applications, the power and ground plane pair's distributed capacitance (discussed below) continues the decoupling function above the SRF of any discrete component.

---

Comparison Table: All Three Tiers at a Glance

Parameter	Tier 1: Bulk	Tier 2: Mid-Frequency	Tier 3: High-Frequency
Target frequency range	DC – 1 MHz	1 MHz – 100 MHz	100 MHz – 1 GHz+
Capacitance value	10 μF – 1,000 μF+	100 nF – 10 μF	1 nF – 100 nF
Component type	Aluminum electrolytic, polymer, large MLCC	X7R / X5R MLCC	C0G or low-ESL X7R MLCC
Typical package	Radial can, D2PAK polymer, 1206–2220 MLCC	0402 / 0603	0201 / 0402
Typical ESL	10–30 nH	0.5–1.5 nH	0.1–0.5 nH
Typical SRF	0.1–5 MHz	30–80 MHz	150–500 MHz
Placement distance from IC	10–50 mm (near regulator)	1–5 mm (near IC power pin)	< 0.5 mm or via-in-pad under IC
Primary function	Sustain voltage during slow load steps; reservoir for regulator response gap	Suppress mid-frequency switching noise; per-IC charge reservoir	Suppress fast transients; maintain voltage at IC core during nanosecond switching
Polarity	Polarized (electrolytic); non-polarized (polymer, MLCC)	Non-polarized	Non-polarized

---

Why You Cannot Replace All Three Tiers with One Large Capacitor

This is the most common misunderstanding in capacitor selection. The reasoning goes: “If I place a 1,000 μF capacitor right next to the IC, it has enormous energy storage and should handle all transients.” There are two fundamental reasons this fails:

1. ESL makes large capacitors ineffective at high frequencies. A 1,000 μF aluminum electrolytic capacitor has ESL of approximately 10–20 nH. Its SRF is around 0.1–0.5 MHz. At 100 MHz, it presents an inductive impedance of approximately X_L = 2π × 10⁸ × 15 × 10⁻⁹ ≈ 9.4 Ω—far higher than a 100 nF 0402 MLCC at the same frequency (which presents approximately 0.05–0.1 Ω near its SRF). The large capacitor physically cannot supply charge fast enough to follow a nanosecond transient.

2. Physical placement is constrained by package size. A large aluminum electrolytic capacitor is physically too large to place within 1 mm of an IC power pin. Even a large MLCC (1210 package) cannot be placed close enough to suppress GHz-range transients. The Tier 3 function requires a package small enough (0201) to fit within the BGA escape zone or immediately adjacent to the IC power pad.

The correct mental model is that each tier hands off responsibility to the next as the transient speed increases beyond what the lower tier can handle. The three tiers are not redundant—they are complementary, each covering a frequency range that the others physically cannot reach.

---

Anti-Resonance: The Hidden Risk of Mixing Tiers

When two capacitors of very different values are placed in parallel on the same power rail without any series impedance between them, an anti-resonance peak can appear in the combined PDN impedance. This happens because the larger capacitor (above its SRF) acts as an inductor, and the smaller capacitor acts as a capacitor; together they form an LC parallel resonant circuit that presents a very high impedance at the resonant frequency.

For example, a 10 μF MLCC (SRF ~5 MHz) in parallel with a 100 nF MLCC (SRF ~50 MHz) may create an anti-resonance peak at approximately 15–20 MHz where the combined PDN impedance rises above both individual capacitors' impedance curves. If this peak falls within the IC's sensitive frequency range, it can cause supply noise at exactly the frequency where the IC is most vulnerable.

Mitigation strategies:

• Add a small series resistance or ferrite bead between tiers: A 0.5–2 Ω series resistance (either a small value resistor or the DC resistance of a ferrite bead) damps the anti-resonance peak without significantly affecting the intended decoupling function of either tier
• Space the capacitor values more carefully: Anti-resonance peaks are strongest when the SRF values of adjacent tiers are separated by more than a decade; choosing Tier 1 and Tier 2 values whose SRF ratio is less than 10:1 reduces peak amplitude
• Use PDN simulation: The only reliable way to identify anti-resonance peaks and verify their frequency and amplitude is through simulation with the actual layout geometry

---

Real-World Examples by IC Type

IC Type	Tier 1 (Bulk)	Tier 2 (Mid-freq)	Tier 3 (High-freq)	Special Considerations
STM32 microcontroller	10 μF polymer near regulator	100 nF X7R 0402 per VCC/VDDIO pin	10 nF C0G 0402 near VDDA pin	Separate VDDA (analog) decoupling from VDDIO (digital); ferrite bead between VCC and VDDA recommended
Intel/AMD CPU (laptop/desktop)	4–8 × 100 μF polymer near VRM	10–50 × 100 nF 0402 X7R around socket	20–100 × 10 nF 0201 under socket area	CPU vendors provide detailed PDN design guides with specific capacitor counts and placement zones
Xilinx FPGA (mid-range)	2 × 47 μF polymer per VCCINT domain	1 × 100 nF per VCCINT bank pin	1 × 10 nF C0G per VCCAUX pin	Each VCCO bank needs its own Tier 2 decoupling; MGT supply needs ferrite bead isolation from main VCCO
DDR5 memory (LPDDR5)	22 μF per VDD/VDDQ domain near supply	100 nF per VDD/VDDQ pin (0402)	10 nF C0G (0201) adjacent to each DRAM ball	JEDEC requires tight VDDQ regulation; reference plane continuity under memory bus traces is critical
AI GPU (VCORE, simplified)	8–16 × 470 μF polymer per phase	Thousands of 100 nF 0402 X7R in array	Thousands of 10 nF 0201 via-in-pad under package	PDN impedance target < 0.1 mΩ; simulation mandatory; see decoupling placement guide for GPU PDN details

---

Implementation Checklist

• Tier 1 checklist:
  • ☐ Bulk capacitors placed within 20–40 mm of voltage regulator output
  • ☐ Capacitance value sized for worst-case load step and regulator response time
  • ☐ Voltage rating ≥ 1.5× supply voltage (electrolytic) or ≥ 2× (MLCC, for bias derating)
  • ☐ Temperature rating adequate for PCB thermal environment
• Tier 2 checklist:
  • ☐ 100 nF X7R 0402 capacitor within 3–5 mm of each IC power supply pin
  • ☐ Voltage rating at least 2× supply rail voltage to account for X7R DC bias derating
  • ☐ Power via connects from plane directly to capacitor pad, then short trace to IC power pin
  • ☐ C0G/NP0 used instead of X7R for PLL AVDD, ADC VREF, and other noise-sensitive supplies
• Tier 3 checklist:
  • ☐ 10 nF or smaller 0201 capacitor placed within 0.5 mm of high-speed IC power pins
  • ☐ Via-in-pad considered for back-side placement directly beneath BGA
  • ☐ SRF of chosen capacitor verified to be above the highest target decoupling frequency
  • ☐ Ground via and power via on separate pads; no shared vias between capacitors
• PDN verification:
  • ☐ Impedance vs frequency simulation completed for all critical supply rails on high-speed ICs
  • ☐ Anti-resonance peaks identified and either damped or moved outside the IC's sensitive frequency band
  • ☐ No plane splits beneath any IC or its decoupling capacitors

---

FAQ

Is there a difference between a bulk capacitor and a reservoir capacitor?
The terms are functionally synonymous in most contexts. “Bulk capacitor” is the most common term in digital PCB design, referring to the large capacitors near the voltage regulator that store the bulk of the charge reservoir. “Reservoir capacitor” is used more commonly in analog power supply design and audio electronics. Both refer to the same Tier 1 function: providing a large local charge store to sustain supply voltage during load transients that are faster than the regulator can respond.

Can I use the same capacitor value for all three tiers?
No. Each tier requires a different capacitance value and package size to be effective in its frequency range. Using 100 nF capacitors for bulk capacitance leaves the supply vulnerable to large slow load steps that require hundreds of microfarads to sustain the rail voltage. Using 100 μF capacitors for high-frequency bypass fails because their high ESL prevents effective decoupling above 1–5 MHz. The three tiers must use different values, different component types, and must be placed at different distances from the IC to function correctly together.

How do I know which tier is responsible for a specific noise problem?
Measure the frequency of the noise on the supply rail using an oscilloscope with a high-frequency probe. Noise below 1 MHz indicates inadequate Tier 1 bulk capacitance or insufficient regulator bandwidth. Noise in the 1–100 MHz range indicates inadequate Tier 2 mid-frequency decoupling—check capacitor placement distance and value. Noise above 100 MHz indicates inadequate Tier 3 high-frequency bypass—check capacitor package size, ESL, and physical distance from the power pin. You can also identify the responsible tier by temporarily adding capacitors of different values at different locations and observing which change reduces the noise.

Do I need all three tiers for every IC on the board?
No. The need for all three tiers scales with the switching speed and power delivery demands of the IC. A simple 3.3 V logic IC switching at 20 MHz needs only Tier 1 (shared with other ICs on the rail) and Tier 2 (100 nF per power pin). It does not need dedicated Tier 3 high-frequency bypass. A DDR5 memory controller, a high-speed FPGA, or an AI accelerator GPU core running at hundreds of MHz–GHz absolutely requires all three tiers, with Tier 3 capacitors via-in-pad directly beneath the package. Use the IC datasheet's power supply recommendations as the starting point; the manufacturer knows which tiers are necessary for their device.

What happens if I skip Tier 2 and use only bulk and high-frequency capacitors?
Skipping Tier 2 creates a coverage gap in the 1–100 MHz range where neither the bulk capacitor (limited by its ESL) nor the high-frequency capacitor (limited by its small capacitance at lower frequencies) provides adequate decoupling. This gap manifests as mid-frequency noise on the supply rail, which can cause logic errors in digital circuits, increased jitter in clock circuits, and elevated EMI emissions. The three-tier strategy works because each tier covers the blind spot of the adjacent tier; removing any tier leaves a frequency range unprotected.

---

Need Help with PCB Assembly and Component Placement?

Implementing a correct three-tier decoupling strategy requires both the right component selection and precise PCB assembly—particularly for Tier 3 components placed at sub-millimeter distances from IC power pins or as via-in-pad beneath BGA packages. NextPCB provides high-density SMT assembly with 0201 capability, via-in-pad processing for HDI boards, and automated optical inspection to verify passive component placement accuracy.

Start with a free DFM check of your layout using the online Gerber viewer, or submit a BOM to the BOM service for component sourcing alongside your PCB assembly.

Get a PCBA quote from NextPCB →