by John Lauer
An independently designed 12 V to 5 V, 1 A synchronous buck molecule (TI TPS54202), with calculations, ngspice and PSpice simulation, KiCad 10 schematic and board, and the board laid out and analysed with adom-aiflow.
| Name | Last updated |
|---|---|
| fields | |
| img | |
| research | |
| results | |
| tools | |
| README.md |
ECAS vs MLCC: should this buck move to polymer capacitors?
This study asks one question of the 12 V to 5 V buck: would swapping its ceramic capacitors (MLCCs) for Murata ECAS polymer aluminium capacitors give a smaller board and a better converter? We tested every candidate against the board's own simulation models, using each part's measured data in place of the value on its label.
The short answer: not as a straight swap. Moving every capacitor to ECAS makes this board worse:
- it puts 25 V on a 25 V part when the board is plugged into a live rail;
- it slows the control loop roughly sixfold;
- it nearly doubles the capacitor cost.
ECAS does earn one place, though: in place of the 47 µF aluminium can at the input, with a 0.22 Ω resistor in series. There it cuts the tallest capacitor from 5.8 mm to 2.0 mm and frees a quarter of the input bulk's area. It also damps the input as well as the can does, and keeps doing so in the cold, where the can does not.
The study also turned up something about the board as built. Its output MLCCs hold less than half their label value at 5 V. Once that is accounted for, the 68 pF feed-forward capacitor runs the loop too fast at the worst-case corner, and 47 pF is the right value.
Contents
- What a capacitor really is on a rail
- The candidates
- Ten capacitor sets, one set of tests
- ECAS on the output
- ECAS on the input
- On the real copper
- Does the board get smaller?
- What it costs
- A finding about the board as built
- What we would do
- How it was done
What a capacitor really is on a rail
A class II ceramic loses capacitance as DC voltage is applied across it. The loss depends on the size, the dielectric and the voltage rating. A polymer capacitor barely changes. Murata publishes the measured curve for every part in SimSurfing, so we pulled the curves for the parts on this board and for each candidate.

| part | on this board | label | at its rail | what the design assumed |
|---|---|---|---|---|
| 22 µF 10 V X5R 0805 (C6 to C8) | output, 5 V | 22 µF | 10.2 µF (46 %) | 12.1 µF (55 %) |
| 10 µF 25 V X7R 1206 (C2, C3) | input, 12 V | 10 µF | 3.5 µF (35 %) | 6 µF (60 %) |
| 47 µF 10 V X5R 0805 | candidate, 5 V | 47 µF | 14.5 µF (31 %) | |
| 10 µF 25 V X5R 0603 | candidate, 12 V | 10 µF | 1.5 µF (15 %) | |
| 33 µF 25 V ECAS | candidate, 12 V | 33 µF | 33.4 µF (101 %) |
The board's two 1206 input capacitors together hold 6.9 µF at 12 V, not the 12 µF in the design calculation. Its three output capacitors hold 30.6 µF, not 36.3 µF. A bigger MLCC in the same package does not buy much either: a 47 µF 0805 at 5 V holds less than half of what its label suggests over a 22 µF one.
The 0805 and 1206 curves are Murata's own parts (GRM21BR61A226ME44, GRM31CR71E106KA12) of the same size, voltage and dielectric as the Samsung and Yageo parts on the board.
The candidates
Adom Parts Search checked Mouser, DigiKey and JLCPCB in one query per family. The prices and stock below are from 11 October 2026.
| technology | part | value | ESR | size (mm) | height | @1 | @100 | Mouser stock |
|---|---|---|---|---|---|---|---|---|
| aluminium electrolytic (as built, C1) | Panasonic EEE-FK1V470P | 47 µF 35 V | 0.36 Ω max | 6.3 can | 5.8 | $0.64 | $0.26 | 99,272 |
| X7R MLCC (as built, C2, C3) | Yageo CC1206KKX7R8BB106 | 10 µF 25 V | a few mΩ | 3.2 x 1.6 | 1.6 | $1.25 | $0.51 | 18,133 |
| X5R MLCC (as built, C6 to C8) | Samsung CL21A226MPQNNNE | 22 µF 10 V | a few mΩ | 2.0 x 1.25 | 1.25 | $0.20 | $0.08 | JLCPCB 8,361 |
| polymer aluminium (ECAS) | Murata ECASD41E336M040KA0 | 33 µF 25 V | 40 mΩ max | 7.3 x 4.3 | 2.0 | $2.16 | $1.06 | 21,685 |
| polymer aluminium (ECAS) | Murata ECASD61A157M010KA0 | 150 µF 10 V | 10 mΩ max | 7.3 x 4.3 | 3.1 | $2.54 | $1.29 | 4,605 |
| polymer tantalum (POSCAP) | Panasonic 6TPE100MAZB | 100 µF 6.3 V | 35 mΩ | 3.5 x 2.8 | 2.0 | $1.15 | $0.58 | 20,000 |
| polymer tantalum | KEMET T520B107M006ATE070 | 100 µF 6.3 V | 70 mΩ | 3.5 x 2.8 | 2.0 | $0.98 | $0.44 | 59,885 |
| hybrid polymer | Panasonic EEH-ZA1V470P | 47 µF 35 V | low | 6.3 can | 5.8 | $1.44 | $0.71 | 27,508 |
| X5R MLCC | Murata GRM21BR61E226ME44K | 22 µF 25 V | a few mΩ | 2.0 x 1.25 | 1.25 | $0.36 | $0.14 | 3,457,126 |
| X5R MLCC | Murata GRM21BR61A476ME15 | 47 µF 10 V | a few mΩ | 2.0 x 1.25 | 1.25 | $0.59 | $0.24 | 0 (JLCPCB 8,092) |
Three things the search turned up:
- ECAS has been pruned once already. Murata's product withdrawal notice D07322 (1 March 2024) ended the older ECAS part numbers, with a last order date of 31 March 2025. It pointed buyers to newer ECAS numbers ending in
KA0. DigiKey lists theKA0parts above as Active, with tens of thousands in stock, but it is a series a careful designer second-sources. The 7343 (EIA D case) footprint is shared with other makers' polymer capacitors. - The hybrid polymer can (EEH-ZA) saves nothing: same can, same height.
- The polymer tantalums are 6.3 V parts. On a 5 V rail that leaves less margin than KEMET's 10 % derating guideline for polymer tantalum asks for, so on this board they would want a 10 V part.
The measured ESR is far below the datasheet maximum: 7.3 mΩ for the 33 µF ECAS at 500 kHz against its 40 mΩ rating. That matters below. ECAS also has about five times the inductance of an 0805 (1.5 nH against 0.3 nH), so it can never replace the small ceramic in the switching loop.
Ten capacitor sets, one set of tests
Every set ran through the same four tests. The power stage and control loop are the board's own ngspice models from the simulation guide: the averaged peak-current-mode loop, calibrated to the TPS54202 datasheet's crossover equation, and the cycle-by-cycle switching model. Each capacitor is a resistor, inductor and capacitor in series, with C at its real DC bias and ESR and ESL from Murata's measured impedance curves.
- Loop: crossover and phase margin at 9, 12 and 16 V, 0.5 and 1 A, with the feed-forward capacitor (Cff) swept from 0 to 220 pF.
- Switching: 12 V in, 1 A out, output and input ripple through a 20 MHz bandwidth limit, then a 0.5 A load step at 1 A/µs.
- Hot plug: the board touching a live 16 V rail through 0.5 µH and 1 µH of wiring.
- Input filter: the impedance the converter sees looking back into its input, against its own negative input resistance (Middlebrook).
| set | input | output | Cff | crossover | load step | settles in | start-up overshoot | hot plug, 16 V | cap cost @100 | cap area | tallest cap |
|---|---|---|---|---|---|---|---|---|---|---|---|
| A, as built | 47 µF can + 2 x 10 µF 1206 | 3 x 22 µF 0805 | 68 pF | 40.5 kHz | 81 mV | 71 µs | 11 mV | 17.0 V | $1.53 | 112 mm² | 5.8 mm |
| A2, as built, Cff 47 pF | as A | as A | 47 pF | 30.3 kHz | 84 mV | 70 µs | 11 mV | 17.0 V | $1.53 | 112 mm² | 5.8 mm |
| B, everything ECAS | 33 µF ECAS + 1 x 10 µF 1206 | 150 µF ECAS | 220 pF | 6.4 kHz | 60 mV | 102 µs | 52 mV | 25.3 V | $2.87 | 100 mm² | 3.1 mm |
| B2, ECAS + one ceramic each side | 33 µF ECAS + 10 µF 1206 | 150 µF ECAS + 22 µF 0805 | 220 pF | 5.9 kHz | 59 mV | 104 µs | 55 mV | 25.3 V | $2.95 | 106 mm² | 3.1 mm |
| C2, ECAS for the can, bare | 33 µF ECAS + 2 x 10 µF 1206 | 3 x 22 µF 0805 | 68 pF | 40.5 kHz | 81 mV | 71 µs | 12 mV | 25.2 V | $2.33 | 88 mm² | 2.0 mm |
| C, ECAS + 0.22 Ω for the can | 33 µF ECAS + 0.22 Ω + 2 x 10 µF 1206 | 3 x 22 µF 0805 | 68 pF | 40.5 kHz | 81 mV | 71 µs | 11 mV | 18.5 V | $2.34 | 92 mm² | 2.0 mm |
| D, denser MLCC only | 47 µF can + 2 x 22 µF/25 V 0805 | 2 x 47 µF 0805 | 68 pF | 46.2 kHz | 81 mV | 71 µs | 10 mV | 17.2 V | $1.04 | 98 mm² | 5.8 mm |
| E, polymer tantalum out | as built | 100 µF POSCAP + 22 µF 0805 | 220 pF | 10.8 kHz | 63 mV | 100 µs | 39 mV | 17.0 V | $1.95 | 115 mm² | 5.8 mm |
| F, smallest | 33 µF ECAS + 0.22 Ω + 2 x 22 µF/25 V 0805 | 2 x 47 µF 0805 | 68 pF | 46.2 kHz | 81 mV | 71 µs | 10 mV | 18.6 V | $1.85 | 78 mm² | 2.0 mm |
| F2, recommended | as F | as F | 47 pF | 32.6 kHz | 84 mV | 69 µs | 11 mV | 18.6 V | $1.85 | 78 mm² | 2.0 mm |
Every set holds output ripple between 2.3 and 4 mV at 12 V (7.7 mV for the POSCAP, from its 35 mΩ ESR). Every set has at least 100 times margin between the input filter's peak impedance and the converter's 27 Ω negative input resistance. Cff for the polymer outputs was chosen by the sweep: the best worst-case phase margin, then the fastest crossover. Cap area is the sum of the KiCad library courtyards.
ECAS on the output

The 150 µF ECAS really is 150 µF at 5 V, five times what the three 0805s hold. The step dip falls from 81 to 60 mV and output ripple from 3.8 to 3.0 mV. But the TPS54202 is internally compensated, and its crossover falls as output capacitance rises: 3.95 / (VOUT x COUT), datasheet equation 14. Five times the capacitance gives roughly a sixth of the crossover, 6.4 kHz instead of 40 kHz. Recovery from the step stretches from 71 to 102 µs, and start-up overshoot rises from 11 to 52 mV.
The load step budget on this board is 150 mV (3 %), and the as-built ceramics already meet it with room to spare. ECAS on the output buys a better number this board does not need. In return it costs a slower loop, an extra $1.29 and a 7.3 x 4.3 mm footprint where three 0805s sit today. It would be the right call for a pulsed load with a tight droop budget and room on the board, but not here.
ECAS on the input

The 47 µF can on the input is not there for its capacitance. TI's datasheet (section 7.3) adds it for a supply more than a few inches away: the bench lead's inductance rings against the low-ESR ceramics. The can's 0.3 Ω of ESR is what damps that ring. Swap in an ECAS with 7 mΩ and the damping is gone. A hot plug onto 16 V through a metre of wire then rings to 25.2 V, on a 25 V capacitor, 5 V short of the TPS54202's 30 V absolute maximum.
The fix is the classic one: put the damping back as a resistor in series with the polymer capacitor. A sweep found the best value.
| series R | hot plug, 16 V, 1 µH | hot plug, 16 V, 0.5 µH | input filter peak |
|---|---|---|---|
| 0.10 Ω | 20.7 V | 18.6 V | 0.144 Ω |
| 0.15 Ω | 19.5 V | 17.7 V | 0.156 Ω |
| 0.22 Ω | 18.5 V | 17.1 V | 0.201 Ω |
| 0.33 Ω | 17.9 V | 17.9 V | 0.274 Ω |
| 0.47 Ω | 18.7 V | 19.3 V | 0.357 Ω |
| 1.0 Ω | 22.1 V | 22.7 V | 0.586 Ω |
| as built: the can | 17.0 V | 16.9 V | 0.248 Ω |
With 0.22 Ω, the ECAS damps nearly as well as the can at room temperature, with a lower filter impedance. It carries 90 mA RMS of the input ripple, so the resistor dissipates under 2 mW. It also holds its damping in the cold. An aluminium electrolytic's ESR rises steeply as it cools, while polymer and a thick-film resistor barely move.
| the can's ESR | hot plug, 16 V, 1 µH | input filter peak |
|---|---|---|
| 0.30 Ω (typical, 25 °C) | 17.0 V | 0.248 Ω |
| 0.36 Ω (Panasonic's 100 kHz maximum) | 17.2 V | 0.288 Ω |
| about 1 Ω (cold; an estimate, about three times) | 22.0 V | 0.584 Ω |
On the real copper
Adom Fields solves the board's real copper and checks each rail's impedance from its load pin. Every capacitor is a branch, with its own mounting inductance: the path from its pads to the load pin over the ground plane. Pointed at a folder of measured curves (capCurves in the spec), it uses each part's real capacitance at the rail's bias, and its ESR and ESL. It reads the dielectric from the MPN and footprint, so the can is an electrolytic and the ECAS a polymer.
We solved two copies of the board: as built, and with C1's MPN set to the 33 µF ECAS. A third curve adds the 0.22 Ω resistor to the ECAS solve's own branches.

| VIN at U1 pin 3 | supply-lead resonance | at 500 kHz | Middlebrook margin |
|---|---|---|---|
| as built (47 µF can) | 277 mΩ at 67 kHz | 42 mΩ | 39.4 dB |
| C1 = 33 µF ECAS, bare | 522 mΩ at 30 kHz | 11 mΩ | 33.9 dB |
| C1 = 33 µF ECAS + 0.22 Ω | 230 mΩ at 56 kHz | 41 mΩ |
Adom Fields sees the same thing the hot-plug simulation does: without the can's ESR, the supply leads' resonance nearly doubles. It stays under the 0.6 Ω target and Middlebrook's limit, so no fix is listed, but the margin to the target falls from about 2x to 1.15x. With the 0.22 Ω resistor, the resonance is the lowest of the three.
Adom Fields also checks a hot plug on its own: it switches the supply leads onto every capacitor at its rail bias and solves the step in time. With the same leads as our simulation (50 mΩ, 1 µH) on the 12 V rail:
| VIN hot plug, 12 V, 1 µH | Adom Fields | ngspice (this study) | Adom Fields' review |
|---|---|---|---|
| as built (47 µF can) | 13.1 V (9 % overshoot) | 13.2 V | no issue |
| C1 = 33 µF ECAS, bare | 18.8 V (56 % overshoot) | 18.9 V | hot-plug-overshoot: "add at least 220 µF of bulk capacitance with about 159 mΩ of ESR (an electrolytic, or a polymer part plus a series resistor)" |
Two independent tools agree within 1 %, and the layout tool reaches the same fix this study did: a polymer part plus a series resistor.
The output rail has the same story in reverse. With the measured parts, Adom Fields counts C6 to C8 at 10.2 µF each at 5 V, and VOUT's worst point is a quarter of its target.
Does the board get smaller?
Not by itself. The molecule's size is set by its pins, not its parts: four corner machine pins on a 24 x 16 mm span, sized in 4 mm steps so they land on the scaffold grid. The board is 28 x 20 mm (560 mm²). The next size down is 24 x 20 mm, 80 mm² smaller.
| set | capacitor courtyards | saved |
|---|---|---|
| A, as built | 112 mm² | |
| C, ECAS + 0.22 Ω for the can | 92 mm² | 20 mm² |
| D, denser MLCC only | 98 mm² | 14 mm² |
| B, everything ECAS | 100 mm² | 12 mm² |
| F, smallest | 78 mm² | 34 mm² |
Even the smallest set saves 34 mm², less than half of one 4 mm step. The can is still what stands in the way of the step, though. It is 7.1 mm wide across the board's long axis, where the left-hand contacts, the feedback divider, U1, L1 and the right-hand contacts already sit in a row. An ECAS turned lengthwise is 4.8 mm wide, so a 24 x 20 mm molecule becomes worth attempting once the can is gone. Proving it takes a fresh placement and route, not arithmetic.
What does change at once is height. The can is 5.8 mm tall; with it gone, the tallest capacitor is the 2.0 mm ECAS and the tallest part on the board is the 2.8 mm inductor. That leaves more clearance over the board for probes and fixtures, and makes for a flatter molecule.
What it costs
At 100 boards, the capacitors on this buck cost $1.53 as built. The two Yageo 1206s are the expensive part of that ($0.51 each); the can is $0.26.
- Everything ECAS (B): $2.87, +$1.34 per board.
- ECAS + 0.22 Ω for the can, the rest as built (C): $2.34, +$0.81.
- The smallest set (F, F2): $1.85, +$0.31, because 22 µF 25 V Murata 0805s replace the 1206s for less.
- Denser MLCCs alone (D): $1.04, $0.49 cheaper than as built, but the can stays.
At reel quantities DigiKey lists the 33 µF ECAS at $0.72 (3,000). One supply note: Murata's 47 µF 0805 showed no stock at Mouser or DigiKey today (JLCPCB had 8,092), so it wants a second source, such as Samsung's CL21A476MQYNNNE.
A finding about the board as built
With label values, the design put the crossover at 28.6 kHz with the 68 pF Cff. With the measured 30.6 µF it is 40.5 kHz, already at TI's 40 kHz guideline. Then the corners:
| output capacitance | Cff 68 pF (as built) | Cff 47 pF |
|---|---|---|
| measured, 25 °C (30.6 µF) | 40.9 kHz | 30.4 kHz |
| 20 % low (part tolerance) | 96 kHz | 43 kHz |
| 32 % low (tolerance and X5R at 85 °C) | 206 kHz, 30° phase margin at 9 V | 62 kHz, over 100° |
The cycle-by-cycle switching model confirms what the averaged loop model predicts at that last corner. With 68 pF, the output ripple at 9 V grows from 4.1 to 9.8 mV, and most of it is a 3.4 mV tone at 250 kHz: half the switching frequency, the start of sub-harmonic oscillation in a peak-current-mode loop. With 47 pF that tone is 0.13 mV.
So the board should carry 47 pF for C9, a one-part change. With 47 pF the as-built board crosses over at 30.3 kHz, and its load step (84 mV), recovery (70 µs) and ripple (3.6 mV) are unchanged. This is also the one real argument for a polymer output on this board: polymer capacitance does not move with voltage, temperature or age, so the loop does not either. Across the same tolerance, the ECAS output's crossover moves from 6.4 to 8.7 kHz.
These loop numbers come from our models, not from a bench. The next step is a Bode plot on a built board.
What we would do
- Change C9 to 47 pF now. It makes the as-built board robust to its own capacitors.
- Keep MLCCs on the output. They meet the step budget, keep the loop at 30 to 40 kHz, and cost cents. Two 47 µF 0805s do the work of three 22 µF ones.
- Replace the can with a 33 µF 25 V ECAS plus a 0.22 Ω 0603 resistor, and the 1206 input pair with two 22 µF 25 V 0805s. The input damps as well as before and stays damped in the cold. The tallest capacitor drops from 5.8 to 2.0 mm, and the capacitors take 31 % less board, for $0.31 more per board at 100.
- Then try a 24 x 20 mm molecule in a fresh Adom AI Flow placement and route, and solve the result in Adom Fields. Only a real layout can say whether the freed 7 mm column of the can buys the 4 mm step.
Would we tell everyone to move to ECAS? No. We would tell them to look up their MLCCs' DC-bias curves before trusting a label, and to keep an input bulk capacitor's ESR in mind before swapping it for something better on paper. Polymer earns its place where stable capacitance or low height matters more than cost and footprint.
How it was done
The whole study ran in one AI session, which did the following:
- Planned the study and framed the question as variants with tests, rather than a single swap.
- Ran one Adom Parts Search query per technology (Mouser, DigiKey, JLCPCB) and checked lifecycle through the DigiKey API.
- Read Murata's withdrawal notice in a pup browser window.
- Pulled measured C-DC bias, |Z|(f) and R(f) curves for 11 parts from Murata SimSurfing's data service.
- Ran each variant through the board's own ngspice models: several hundred simulations in all.
- Solved the board in Adom Fields with the measured curves, as built and with the ECAS, for the rail impedance on the real copper.
Two simulation traps are worth knowing if you repeat this. A load step that switches in zero time measures the capacitors' inductance spikes, not the loop, so step at a realistic slew (1 A/µs here). And read ripple through a 20 MHz bandwidth limit, as a scope would; otherwise an ideal switch's edge across a capacitor's ESL reads as volts of ripple.
| folder | contents |
|---|---|
tools/murata_curves.py |
fetches the measured curves for each part |
tools/parts.py |
each capacitor's C at bias, ESR and ESL, price, footprint and height |
tools/variants.py |
every capacitor set and the four tests; python3 variants.py reruns them all |
tools/pdn_compare.py |
the VIN impedance chart from the two Adom Fields solves, plus the damping resistor |
fields/ |
the Adom Fields spec (with capCurves) and the curve folder it reads; the board's Yageo and Samsung MPNs map to their Murata twins' curves |
tools/plots.py |
the charts and results/summary.json |
results/ |
every result as JSON: each set, the damping sweep, the can's ESR corners, the Cff corners |
research/murata/ |
the measured curves, one JSON per part |
Honest limits:
- ngspice switch models are ideal apart from 150 pF on the switch node and a loss resistance across each capacitor's ESL.
- The cable is a lumped 0.5 or 1 µH with 50 mΩ.
- Capacitor curves are Murata's 25 °C measurements.
- Temperature corners are estimates.
- Nothing here has been measured on hardware yet.