PDN Decoupling Planner
Set the rail's ripple budget and current step to get the target impedance. Then build the capacitor network, with each part's ESR, ESL and mounting inductance, and see where |Z| rises above the target. Or let the planner pick the parts for you.
1 · Rail and target
2 · Frequency band
3 · Regulator (VRM)
4 · Plane pair and load path
5 · Mounting
Pads, traces and vias add loop inductance in series with every capacitor. The via geometry and how deep the planes sit matter more than the package. Changing this sets every row.
Result
Impedance vs frequency
The shaded bands are where |Z| exceeds Z_t inside the analysis band. Dots mark anti-resonance peaks. Tick "Plot" in the table to overlay up to four capacitor groups.
Capacitor network
Each row is a group of identical parts in parallel. Values take SI suffixes (100n, 4.7u). ESR at resonance, body ESL and mounting inductance come from the library defaults; edit them to your part's vendor model.
| Plot | On | Qty | Value (F) | Package | ESR (mΩ) | ESL (nH) | Mount (nH) | Derate (%) | Mounted SRF |
|---|
Where it misses the target
| From | To | Peak |Z| | At | Peak / Z_t |
|---|
Anti-resonance peaks
| Frequency | |Z| | vs Z_t | Status |
|---|
Auto-suggest a BOM
Adds one capacitor at a time from the library, each time choosing the part that cuts the most excess impedance across the band, until |Z| ≤ Z_t from the lowest to the highest frequency. It reuses parts already chosen when they do nearly as well, to keep the BOM short. Uses the VRM, plane, path and mounting settings above.
Capacitor library and sources
Typical values for general-purpose 2-terminal parts. ESR is at self-resonance. Real parts vary by ±50 % or more, so use the vendor model for the part you will buy.
| Value | Package | Type | ESR | Body ESL | Note |
|---|
Target impedance, real capacitors, and why peaks appear
A load that changes its current by ΔI pulls the rail by ΔV = |Z(f)|·ΔI at every frequency the step contains. Keep |Z| under a target across that band and the ripple stays inside its budget. This is the target-impedance method of Smith et al.:
A real capacitor is a series RLC. Its inductance is the part's own ESL plus the mounting loop: pads, any trace, the vias and the plane spreading under it. That mounting term is usually as large as the ESL or larger, so it gets its own column.
The rail node is everything in parallel: the regulator, the plane pair and each capacitor group. Admittances add as complex numbers, so phase matters. Above its SRF a big capacitor looks like an inductor. Below its SRF a small one still looks like a capacitor. Between the two they form a parallel LC tank, and |Z| peaks there. This is the anti-resonance. Its height is set by the ESRs, so very-low-ESR parts make it worse.
The auto-suggest
The planner evaluates |Z| on 240 log-spaced points across your band. At each step it tries adding one more of every allowed library part, then keeps the one that most reduces the total excess Σ max(0, ln(|Z|/Z_t)). If a part already on the BOM gets within 80 % of the best improvement, it takes that part instead, which keeps the line count down. It stops when the network meets the target everywhere, when the part limit is reached, or when no part helps. It is greedy, so it finds a good answer, not the provably smallest one.
Limits
- A single lumped node. Plane cavity resonances, the distance from each capacitor to the load, and the IC package and die capacitance are not modelled. Above roughly 50–100 MHz the board's capacitors stop mattering and the package does the work. Use the path inductance to see where that ceiling sits for your layout.
- ESR and ESL are taken as constant. Real MLCC ESR varies with frequency, and Class II ceramics (X5R/X7R) lose capacitance with DC bias, temperature and age. A small, high-value part at its rated voltage can lose more than half its value. Enter the derating from the vendor's DC-bias curve at your rail voltage.
- The target-impedance method assumes a flat target and a linear, time-invariant network. A load with a known current spectrum can be checked more precisely in the time domain.
- This is an engineering estimate for planning. Before layout sign-off, verify with a PDN simulation that uses the vendor S-parameter or SPICE models and your real stackup.
Where the library numbers come from
- Body ESL: 0.3 / 0.4 / 0.5 / 0.6 / 0.8 / 1.0 nH for 0201 / 0402 / 0603 / 0805 / 1206 / 1210. UG483 Table 2-5 specifies a maximum of 0.5 nH for its 0603 and 0805 parts and 1 nH for 1210. AVX's fixture measurements read higher, at 0.87, 1.05 and 1.25 nH for 0.1 µF in 0603, 0805 and 1206 [3], because what counts as "the part" depends on the test fixture.
- ESR: typical values at resonance, falling with capacitance from 150 mΩ at 1 nF to 2.5 mΩ at 100 µF. They fall inside UG483's specified ranges, for example 1–20 mΩ for the 0.47 µF and 4.7 µF parts. The two tantalum entries are the UG483 parts, with ESR taken from their part numbers.
- Mounting: 0.4 nH for via-in-pad close to the planes (measured 0.44 nH for an 0402 [5]); 0.6 nH for an optimised top-side 0402 (0.57 nH [4]); 0.8 nH, UG483's 0402 example; 1.5 nH, the top of UG483's 300–1,500 pH range; 2.3 nH for the same 0402 on the far side of the board [4].
- Plane pair: εr 4 at 102 µm gives 224 pF/in² and μ0·d = 128 pH per square. That matches UG483 Table 2-10 (225 pF/in², 130 pH/sq).
References
- L. D. Smith, R. E. Anderson, D. W. Forehand, T. J. Pelc, T. Roy, "Power distribution system design methodology and capacitor selection for modern CMOS technology," IEEE Trans. Advanced Packaging, vol. 22, no. 3, pp. 284–291, Aug. 1999.
- Xilinx (AMD), 7 Series FPGAs PCB Design Guide, UG483 v1.13 (2017): Table 2-5 PCB capacitor specifications, capacitor mounting inductance (300–1,500 pH; 0402 example L_SELF 0.9 nH + L_MOUNT 0.8 nH, 53 → 38 MHz), Table 2-10 plane capacitance and spreading inductance.
- J. Cain, "Parasitic Inductance of Multilayer Ceramic Capacitors," KYOCERA AVX technical paper.
- Intel (Altera), AN 574, Printed Circuit Board (PCB) Power Delivery Network (PDN) Design Methodology: 0402 mounting inductance 0.57 nH top side, 2.3 nH bottom side. Also the Intel PDN Tool user guide.
- Axotron, "Inductance of some decoupling layouts" (2014): measured 0402 via-in-pad loop inductance, 0.44 nH top, 1.04 nH bottom.
- H. Johnson, "Parasitic Inductance of Bypass Capacitor II," High-Speed Digital Design newsletter, sigcon.com: mounting geometry dominates (0603: 0.51 nH side vias vs 2.13 nH long skinny traces).
- E. Bogatin, Signal and Power Integrity — Simplified, 3rd ed., Prentice Hall, 2018, ch. on the PDN and target impedance.
Rail misbehaving under load?
Kyros designs and brings up the power stages and decoupling networks behind FPGAs, MCUs and mixed-signal boards, and measures the result on the bench.
