Skip to the content
4 layers100 × 70 mmSync buck, the printed circuit board in 3D

This project showcases a synchronous buck converter module, demonstrating advanced power PCB layout skills. It features techniques like minimized switching loop areas, strategic ground planes, isolated feedback traces. The README includes an efficiency curve mapping performance across various loads.

Size
100 × 70 mm
Copper layers
4
Thickness
1.6 mm
Parts
58
Nets
17
Pads
114
Tracks, vias
146, 79
Pours
5
Routed
Yes
KiCad
KiCad 9.0 (file 20241229)
Created
8 Sep 2026
Updated
14 Sep 2026

Overview

Sync buck is an open-source printed circuit board by karimf1, from the karimf1/buck-converter repository on GitHub, published under the MIT licence. It is a 4-layer board of 100 × 70 mm with 58 parts on 17 nets, routed with 146 tracks and 79 vias.

It is built around the LMR33630ADDA and SMBJ24A. Its parts include 18 capacitors, 8 resistors, 4 diodes, 4 connectors and 2 ICs. It belongs with the power boards of this library.

Main components

2

README

README.md

sync-buck — 12 V → 5 V / 3 A synchronous buck, laid out twice

The same synchronous buck converter laid out twice on one board — once properly, once with the switching loop stretched, the feedback trace run beside the switch node, and the thermal vias cut from six to two — so the cost of bad power layout can be measured instead of asserted.

board


Key features

  • Two electrically identical channels on one board — same schematic, same parts, same values, same footprints. Only the copper differs, which is what makes the comparison a measurement rather than an anecdote.
  • Three specific layout sins in channel B, each isolated and quantified: a stretched input-capacitor loop, feedback routed beside the switch node, and a starved thermal via field.
  • Kelvin sense pads on VIN and VOUT at both channels, so efficiency is not contaminated by lead and connector drop.
  • Probe points in signal + ground pairs at 2.54 mm pitch, so switch-node ringing can be caught with a ground spring instead of a 150 mm ground lead.
  • Every component value derived from the operating point, not copied from the datasheet's typical-application circuit — with DC bias derating and input capacitor RMS current carried through.
  • 4-layer stack with a solid L2 ground plane, deliberately slit under channel B's hot loop and left continuous under channel A's.

The idea

Three claims about power layout, in order of how much they matter:

  1. The hot loop is the input capacitor loop, not the output loop. Inductor current is continuous; input current is a chopped square wave at 3 A. Only Cin → high-side FET → low-side FET → Cin carries di/dt, and only its enclosed area sets the switch-node ringing. Most people who "know" to minimise loop area cannot say which loop.
  2. FB is the highest-impedance node on the board and it sits millimetres from the noisiest one. Coupling into it does not blow anything up — it shifts the regulation point and adds jitter, the kind of fault that survives to production.
  3. The thermal vias under the exposed pad are the heatsink. There is no other path out of the package.

Channel B breaks all three on purpose. Everything else about the two channels is identical.

What actually differs

Measured out of the board file:

channel Achannel B
input cap → VIN pin, total reach3.51 mm19.41 mm
enclosed hot-loop area2.23 mm²12.33 mm²
vias in the hot loop02
thermal vias under the exposed pad62
FB on the top layer within 10 mm of SW / L1.2 mm26.7 mm
FB on shielded (inner-referenced) layers30.4 mm18.9 mm
closest FB-to-SW approach, same layer3.76 mm2.27 mm
L2 ground plane under the hot loopcontinuous10 × 12 mm slit
schematic, parts, values, footprintsidenticalidentical
channel A hot loopchannel B hot loop
A — the 100 nF sits across pins 1 and 2, no vias in the loop.B — the same part, 9 mm away, reached through two vias.

Specification

parametervaluewhy
Vin7 – 18 V (12 V nominal)wide enough that the duty cycle visibly moves in the data
Vout5.02 V, resistor-set40.2 k / 10.0 k from a 1.0 V reference
Iout3 A15 W — a real thermal problem, still safe on a bench
fsw400 kHzfixed on the LMR33630A
Inductor ripple24 % at 12 V, 30 % worst case at 18 V10 µH, Bourns SRP6060FA-100M
Output ripple6.9 mV pk-pk predicted at 3 A2 × 22 µF, derated to 33 µF at bias
Board100 × 70 mm, 4-layer, 1.6 mm, ENIGL2 is a solid ground plane

400 kHz was chosen for measurability, not optimality. It keeps the fundamental three decades below the parasitic ring so the two are separable on a lab scope, and it keeps the inductor big enough to hand-place. A product would switch faster and say so.

An internally compensated controller was chosen deliberately. Giving up the control loop is what lets the project be about layout and measurement instead of about compensation networks. It is a scoping decision, and it costs the project any claim about crossover frequency or phase margin.

Where the component values come from

Nothing is copied from the typical-application circuit. design/spec.py is the single source of truth — the derivation and the schematic generator both import it, so a value cannot be changed in one place and stale in the other.

Two details worth pulling out, because they separate people who have built things from people who have read about them:

  • DC bias derating. A 22 µF X7R 1210 rated 25 V is not 22 µF at 5 V — it is closer to 16 µF. The design uses the derated numbers throughout; using nameplate would under-predict output ripple by 33 %.
  • Input capacitor RMS current. Iout·√(D(1−D)) peaks at 1.5 A inside this input range. That is real current into a ceramic, and it is why there are two 1210s rather than one.

What the board is built to measure

The instrumentation choices matter as much as the circuit, because a bad measurement of a good board is indistinguishable from a bad board:

  • Kelvin sense pads on VIN and VOUT at both terminals. At 3 A, 50 mΩ of lead and connector resistance is 150 mV — 3 % of the output, straight into the efficiency number as an error.
  • Probe points in signal + ground pairs 2.54 mm apart, so every measurement can be taken with a ground spring. A scope probe's stock 150 mm ground lead is 100–150 nH and rings at 50–100 MHz — the same band as the ringing you are trying to measure. Whatever you see is partly your probe.
  • Currents read as millivolts across a 10 mΩ four-wire shunt, not on the meter's current range. A DMM's DC-volts function is its most accurate; its current shunt is its least, and it sits in series with your circuit.
  • Efficiency is a ratio of four measurements, so its uncertainty is √(Σ(δx/x)²) ≈ ±1 %. "92 % ± 1 %", never "92.34 %" — quoting four significant figures from a handheld meter is the fastest way to say you have not thought about measurement.

The result the board exists for is measurement 4: switch-node ringing, A against B, same scope, same probe, same geometry.

Honest limitations

  • No conducted EMI measurement. That needs a LISN and a spectrum analyser. Any near-field comparison is relative, between two channels at fixed geometry, and proves nothing about compliance.
  • Scope bandwidth will be a caveat on every ringing number. Switch-node ringing is typically 50–200 MHz; a 100 MHz scope under-reads it. The A-vs-B ratio is the result, not the absolute millivolts.
  • One sample of each channel. No unit-to-unit spread, no temperature sweep, no life test.
  • 12 V here means a bench supply, not a vehicle bus with its load-dump transients. Nothing about this is automotive- or safety-qualified.

Possible improvements

  • Isolate the three sins into three channels instead of one. Channel B breaks all three rules at once, so a difference in the data cannot be attributed to any single cause. Three B-variants — long loop only, bad FB only, starved vias only — would turn one qualitative result into three quantitative ones. It costs board area and one more build.
  • Move to an externally compensated controller in a v2. Once the layout question is answered, the interesting remaining variable is the loop, and an external network makes phase margin something you design rather than inherit.
  • Add a shunt-and-amplifier on-board instead of an external four-wire shunt. It removes a pair of clip leads from the efficiency measurement, which is where a chunk of the ±1 % uncertainty lives.
  • Raise fsw in a second build. 400 kHz was chosen to make the ringing easy to separate on a slow scope. Having measured it there, repeating at 1–2 MHz would show whether the A/B gap widens the way the loop-inductance argument predicts.
  • Add a thermal test point pad next to each exposed pad, so junction temperature can be inferred from a thermocouple rather than only from an IR camera's view of the package top.
  • Populate a common-footprint inductor family so the inductor's own contribution to the ringing can be swapped out and separated from the layout's.

Parts

58
QtyValueFootprintReferences
410uF/50VCapacitor_SMD:C_1210_3225MetricC2, C3, C12, C13
422uF/25VCapacitor_SMD:C_1210_3225MetricC6, C7, C16, C17
210.0k 1%Resistor_SMD:R_0402_1005MetricR2, R12
2100kResistor_SMD:R_0402_1005MetricR3, R13
2100nF/16VCapacitor_SMD:C_0402_1005MetricC4, C14
2100nF/25VCapacitor_SMD:C_0402_1005MetricC8, C18
2100nF/50VCapacitor_SMD:C_0402_1005MetricC1, C11
210uHInductor_SMD:L_Bourns_SRP6060FAL1, L11
21uF/16VCapacitor_SMD:C_0603_1608MetricC5, C15
22.2kResistor_SMD:R_0603_1608MetricR4, R14
222pF C0GCapacitor_SMD:C_0402_1005MetricC9, C19
240.2k 1%Resistor_SMD:R_0402_1005MetricR1, R11
2FBTestPoint:TestPoint_Pad_D1.5mmTP5, TP15
2INPUTTerminalBlock_Phoenix:TerminalBlock_Phoenix_MKDS-1,5-2-5.08_1x02_P5.08mm_HorizontalJ1, J11
2LMR33630ADDAPackage_SO:Texas_HSOP-8-1EP_3.9x4.9mm_P1.27mmU1, U11
2OUTPUTTerminalBlock_Phoenix:TerminalBlock_Phoenix_MKDS-1,5-2-5.08_1x02_P5.08mm_HorizontalJ2, J12
2PGTestPoint:TestPoint_Pad_D1.5mmTP6, TP16
2SMBJ24ADiode_SMD:D_SMBD1, D11
2SWTestPoint:TestPoint_Pad_D1.5mmTP1, TP11
2SWGTestPoint:TestPoint_Pad_D1.5mmTP2, TP12
2VINGTestPoint:TestPoint_Pad_D1.5mmTP8, TP18
2VINSTestPoint:TestPoint_Pad_D1.5mmTP7, TP17
2VOGTestPoint:TestPoint_Pad_D1.5mmTP10, TP20
2VOSTestPoint:TestPoint_Pad_D1.5mmTP9, TP19
2VRIPTestPoint:TestPoint_Pad_D1.5mmTP3, TP13
2VRIPGTestPoint:TestPoint_Pad_D1.5mmTP4, TP14
2greenLED_SMD:LED_0805_2012MetricD2, D12

Similar boards

  • 10 layers224.5 × 160 mmHaasoscope pro adc fpga board, the printed circuit board in 3D

    Haasoscope pro adc fpga board

    drandyhaas/HaasoscopePro

    Affordable 2 GHz 3.2 GS/s 12 bit open-source open-hardware expandable USB real-time oscilloscope

    586 parts485MIT
  • 2 layers116 × 74 mmC64-psu, the printed circuit board in 3D

    C64-psu

    jamarju/c64-psu

    Efficient Commodore 64 power supply with overvoltage + overcurrent protection

    50 parts53
  • 2 layers175 × 115 mmLaser Controllerv2 only for pcb test, the printed circuit board in 3D

    Laser Controllerv2 only for pcb test

    Andre1Becker/GalvOS

    Open-source ESP32-S3 laser show controller — built to replace proprietary OEM controllers in RGB galvo laser projectors

    122 parts11GPL-3.0
  • 4 layers120.2 × 224.7 mmPipcb, the printed circuit board in 3D

    Pipcb

    Coen-Lab/exp-control-board

    Hardware and firmware for an experimental control board for scientific rigs

    272 parts3AGPL-3.0
  • 4 layers102.8 × 67 mmEntegre fc, the printed circuit board in 3D

    Entegre fc

    ErdemA33/stm32f405-flight-controller

    Open-source ArduPilot-compatible flight controller for fixed-wing UAVs. STM32F405, 4-layer KiCad design, built in 5 modular stages.

    129 parts3
  • 2 layers59.1 × 75.7 mmPS MotorControllerInterfacePCB, the printed circuit board in 3D

    PS MotorControllerInterfacePCB

    lhr-solar/PS-MotorControllerInterfacePCB

    Provides power conversion/protection and fan control for the Prohelion WaveSculptor 22 motor controller

    127 parts3
  • 6 layers46.5 × 47.2 mmNAVCORE SoOP, the printed circuit board in 3D

    NAVCORE SoOP

    kfilipekk/iridium_drone

    STM32H743 flight controller that navigates without GPS from Iridium satellite Doppler

    217 parts2MIT
  • 4 layers40.5 × 37 mm4 LAYER POWER SUPPLY copy, the printed circuit board in 3D

    4 LAYER POWER SUPPLY copy

    KareemOye/USB_POWER_SUPPLY_PCB

    12 V to 5 V USB step-down supply based on the LM2596-ADJ, designed in KiCad

    30 parts1

Questions

How big is the Sync buck PCB?

It measures 100 × 70 mm and has 4 copper layers, 1.6 mm thick.

What parts does the Sync buck use?

58 parts, built around the LMR33630ADDA and SMBJ24A. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.

Where can I download the Sync buck KiCad files?

From the karimf1/buck-converter repository on GitHub, at the commit this page was made from: the board is sync-buck.kicad_pcb and its schematic sync-buck.kicad_sch. PCBKit opens the board in the browser, with nothing to install.

Can I use the Sync buck design in my own project?

Its repository publishes it under the MIT licence: you may reuse it on that licence's terms (keep the author's notice, and for a copyleft licence share your changes alike).