The usual answer is a table, and most tables are the same IPC-2221 formula at different settings. That's a fine starting point for an outer-layer trace on an ordinary board. It's too generous for some boards, far too cautious for internal layers, and it says nothing about voltage drop, which is often the limit that bites first.
Short answer: on 1 oz copper on an outer layer, allowing a 10 °C rise, about 0.30 mm (12 mil) for 1 A, 0.78 mm (31 mil) for 2 A, 1.37 mm (54 mil) for 3 A and 2.77 mm (109 mil) for 5 A. Allow 20 °C and they shrink by about a third; use 2 oz copper and they halve. Then check the voltage drop: at 1 A, 100 mm of that 0.30 mm trace drops 175 mV.
The table
Width for a continuous current, from the IPC-2221 formula (I = k·ΔT^0.44·A^0.725, k = 0.048 outer, 0.024 internal). The first three columns are outer layers, the last is an internal layer.
| Current | 1 oz outer, 10 °C | 1 oz outer, 20 °C | 2 oz outer, 10 °C | 1 oz internal, 10 °C |
|---|---|---|---|---|
| 0.5 A | 0.12 mm (5 mil) | 0.08 mm (3 mil) | 0.06 mm (2 mil) | 0.30 mm (12 mil) |
| 1 A | 0.30 mm (12 mil) | 0.20 mm (8 mil) | 0.15 mm (6 mil) | 0.78 mm (31 mil) |
| 2 A | 0.78 mm (31 mil) | 0.51 mm (20 mil) | 0.39 mm (15 mil) | 2.03 mm (80 mil) |
| 3 A | 1.37 mm (54 mil) | 0.90 mm (35 mil) | 0.68 mm (27 mil) | 3.56 mm (140 mil) |
| 5 A | 2.77 mm (109 mil) | 1.82 mm (71 mil) | 1.38 mm (54 mil) | 7.19 mm (283 mil) |
| 10 A | 7.19 mm (283 mil) | 4.72 mm (186 mil) | 3.60 mm (142 mil) | 18.71 mm (737 mil) |
A coupon board with one 40 mm outer-layer trace per current, each at the 1 oz, 10 °C width from the table, beside a 0.25 mm signal trace for scale. Built and rendered in KiCad 9; it passes KiCad's design-rule check.
Going from 1 A to 5 A takes the trace from barely wider than a signal trace to over 2.7 mm. At 10 A the 1 oz outer width passes 7 mm, which is the point to stop widening traces and use a copper pour, heavier copper, or both.
How to read it: rise, copper weight, layer
Temperature rise is how much hotter than its surroundings the trace may run. It's a choice, not a property of the board. 10 °C is the cautious default; 20 °C shrinks every width by about a third.
Copper weight is thickness: 1 oz is about 35 µm, 2 oz is twice that. Capacity follows cross-section, so 2 oz roughly halves the width. On budget fabs the outer layers are 1 oz unless you ask.
Layer is where IPC-2221 is most conservative. Its formula halves the constant for internal layers, which makes an internal trace about 2.6 times as wide as an outer one for the same current.
What IPC-2152 changed
IPC-2221's external chart comes from US National Bureau of Standards tests in the 1950s, and its internal chart was never tested: it was derived from the external one. IPC-2152, published in 2009, replaced both with new measurements, and two findings matter for everyday boards.
First, internal traces aren't the poor relations IPC-2221 makes them: in the new tests internal and external traces carrying the same current ran at close to the same temperature rise, internal ones slightly cooler, so the 2.6× internal penalty is mostly margin. Second, the trace is only part of the story. Board thickness and copper planes near the trace change the temperature rise as much as the width does, which is why IPC-2152 gives a baseline chart (3 oz copper, still air, no planes) plus modifiers for copper weight, planes, board thickness and environment.
In practice: the IPC-2221 table is a safe outer-layer answer for a board in open air. Over or beside a solid plane, a trace runs cooler than the table says. In a sealed box, it runs hotter.
Wide enough for heat is not wide enough for voltage
The width tables size a trace so it doesn't get too hot. They don't look at how much voltage it loses on the way. Here's 100 mm of each heat-sized trace from the first column, with the copper at 35 °C:
| Current | Width | Resistance | Voltage drop | Heat in the trace |
|---|---|---|---|---|
| 1 A | 0.30 mm | 175 mΩ | 175 mV | 175 mW |
| 2 A | 0.78 mm | 67 mΩ | 135 mV | 269 mW |
| 3 A | 1.37 mm | 38 mΩ | 115 mV | 345 mW |
| 5 A | 2.77 mm | 19 mΩ | 95 mV | 473 mW |
175 mV is about 5% of a 3.3 V rail, gone before the current reaches the load. For a supply that feeds a regulator that may not matter; for the rail a microcontroller or sensor runs from, it often does. Widen the trace, or bring the load closer, until the drop is small enough.
And the ground path carries the same current back. A wide supply trace with a thin ground return, or a ground pour cut into islands, just moves the drop to the other side of the circuit.
When the table is wrong
- Short necks are fine. Thermal modelling by Brooks and Adam shows a trace that narrows for a few millimetres to enter a pad runs cooler than a long trace of that width, because the wide copper on each side draws heat away. The benefit shrinks as the neck gets longer; the table is for long runs.
- Pulsed and switching currents heat a trace by their RMS value, not their peak. A motor that draws 5 A at start-up for half a second doesn't need a 5 A continuous trace, but a regulator's switching loop needs a short, wide path for reasons that have nothing to do with heat.
- Vias are the bottleneck. A power path that changes layer through one small via is limited by that via, not the trace. Use several.
- Enclosures and hot neighbours raise the starting temperature. A 10 °C rise on a board already at 60 °C beside a regulator is a 70 °C trace.
- Above about 10 A on 1 oz copper, the widths stop being practical traces. Use a pour, heavier copper on the power layers, or a bus bar.
Doing it for your own numbers
Every number above comes from the same code as PCBEditor's free trace width calculator and trace resistance calculator, so you can plug in your own current, copper, rise and length and get the same answers. In PCBEditor itself, stating a current in the description ("the motor rail carries 3 A") sizes that power net with the same IPC-2221 formula at a 10 °C rise; past 6 mm it lays 6 mm and tells you to use a pour, 2 oz copper or a bus bar instead. The full guide is on pcbeditor.com.
Which do you size by first, heat or voltage drop? I'd like to know how often the drop is what actually sets your power trace widths.
