Contact Us
  • Home
  • BLOG
  • 50 Ohm Impedance Tolerance on RF PCBs: What the Number Controls

50 Ohm Impedance Tolerance on RF PCBs: What the Number Controls

Every RF board is specified at 50 ohm, and almost no RF board is exactly 50 ohm. The question that decides yield, and often whether a design works at all, is how far off it is allowed to be and whether the fabricator can hold that window repeatedly. This page covers what the tolerance number actually controls, where it is lost in fabrication, and how to specify it so the answer is measurable instead of a promise.

1. What the 50 ohm number is doing

A 50 ohm trace is a transmission line, not a wire. Its impedance is set by four things: the width of the trace, the dielectric thickness between the trace and the reference plane, the dielectric constant of the laminate, and the copper thickness. Change any one and the impedance moves. That is the whole problem in one sentence, because every fabrication step touches at least one of them.

Impedance tolerance is therefore not a cosmetic spec. On a 50 ohm line feeding a power amplifier, a 10 percent error produces a reflection coefficient of about 0.05, which is a return loss near 26 dB. On a narrowband filter or an antenna feed, that same error shifts center frequency and changes the match at the band edges. On a 30 GHz link it also changes the phase, and phase is what the system is often built around.

2. The three tolerances people confuse

SpecificationWhat it controlsTypical value
Characteristic impedanceTarget value of the line, for example 50 ohm single ended or 100 ohm differential50 ohm plus/minus 5 to 10 percent
Impedance tolerance across the panelVariation between coupons, between layers and across the panel areaWithin the same window as the target
Lot to lot repeatabilityShift between production runs of the same stack-upThe number nobody specifies and everybody notices

The first is on the drawing. The second is what the fabricator's coupon test proves. The third is what causes a design that passed qualification to fail on the fifth order, and it is the one worth writing into a supply agreement.

3. Where the tolerance is actually lost

Trace width after etch. A trace is drawn at a width and comes out narrower, because etching removes copper laterally as well as vertically. On 0.15 mm traces a 0.02 mm loss is 13 percent of the width, and impedance rises with narrower traces. The fabricator compensates in the artwork, but the compensation depends on copper thickness, etch chemistry and how the panel is loaded. This is a process control item, not a CAD item.

Dielectric thickness. Prepreg flows during lamination. The finished dielectric thickness is a function of prepreg style, resin content, press cycle, copper distribution on both sides and the number of layers. A stack-up quoted on nominal values can come out several percent thin where copper density is low, and thin dielectric means low impedance.

Dielectric constant. Dk is not a constant. It varies with frequency, with temperature, and with resin content. On PTFE-based laminates the variation is small and well documented; on FR-4 it is larger, and the Dk used in the design calculation is usually a value at 1 MHz or 1 GHz, while the board runs at 10 GHz or higher. The fabricator can only hold the Dk it is given, so if the design used the wrong Dk the error is built in before the first panel is pressed.

Copper thickness. Plating adds copper to the surface and to the walls. A line drawn for 1 oz base copper finishes thicker after plating, which lowers impedance. This is why the impedance target must be specified on the finished board, not on the artwork.

Reference plane integrity. A reference plane with a slot under the trace, or a plane that is split, changes the return path. The line is no longer 50 ohm regardless of what the coupon says. Coupons are tested on a dedicated strip where the plane is solid, so a coupon can pass while the product fails. This is the most common source of "the coupon said 50 ohm" arguments.

4. How the measurement is made

Two methods matter. Time domain reflectometry sends a fast edge down the line and reads the reflection; it shows impedance as a function of distance, which is how you see a discontinuity at a via or a connector footprint. The coupon method uses a test coupon on the production panel, measured with a TDR or a vector network analyzer, and reports a single number per layer.

What to ask for in the report: the coupon design used, the layer tested, the position on the panel, the instrument, and the pass or fail against the drawing. A report that says "50 ohm, pass" with no position and no coupon design is not a measurement, it is an assertion. On controlled impedance orders we require a coupon per controlled layer, measured on the production panel, with the raw trace attached to the lot documentation.

5. What the tolerance costs

Tighter tolerance is not a pricing trick. To hold plus/minus 5 percent instead of plus/minus 10 percent, the fabricator has to control etch compensation more closely, verify prepreg lots, measure finished dielectric thickness, and test more coupons. On a 4-layer digital board this is routine. On a 12-layer mixed stack with PTFE and FR-4 in the same build it is skilled work, and the price difference is real.

The useful question is not "can you hold 5 percent" but "which lines need 5 percent". A DC bias line does not. An antenna feed, a filter, a coupler and a long high-speed differential pair do. Splitting the drawing into controlled and uncontrolled nets lets the tight tolerance be spent where it changes performance.

6. Specifying it so it can be checked

Put these on the fabrication drawing: target impedance and tolerance per net class; the stack-up with material part numbers, prepreg styles and finished dielectric thickness per layer; the reference layer for each controlled line; the coupon design; the measurement method and the instrument class; and the report requirement. If the tolerance is lot to lot rather than per board, say so, because the test method differs.

Also state what happens on a fail. Rework of impedance is not possible after lamination. The options are scrap, deviation with customer approval, or a stack-up change on the next lot. Agreeing that in advance is cheaper than arguing about it during a build.

7. Where we see this go wrong

Three patterns repeat. A design calculated with FR-4 Dk at 1 GHz is built on a high-frequency laminate and comes out 8 percent low, because the stack-up was never recalculated. A stack-up quoted on nominal prepreg thickness is pressed with a copper pattern that flows resin out of the area, and the dielectric comes out thin. And a board passes the coupon but has a plane split under a feed line that was never checked, so the coupon is honest and the product is still wrong.

None of these are exotic. All three are visible before the first panel is pressed, if the stack-up and the plane layout are reviewed together instead of separately.

Send the stack-up and the impedance requirements with the Gerber and we will come back with the coupon design, the material callout and the tolerance we can hold on that build, before anything is ordered.

Related reading

Need a DFM check or a stack-up review on your files? Write to pcba@superb-tech.com or WhatsApp +86 133 9608 1443.