Two boards built to the same Gerber, on the same line, from the same material specification, can measure several tenths of a dB apart in insertion loss. The first reaction is to suspect the measurement. Sometimes it is. Often it is not, and the cause is a material or process variable that was never specified, so nobody was controlling it.
This page is about where lot-to-lot variation actually comes from on an RF board, how large each source is, and what to put in a specification so the number stays inside the window.
1. The four sources
| Source | Typical contribution | Controlled by |
|---|---|---|
| Laminate Dk and Df tolerance | Largest single variable on PTFE and filled materials | Material specification and lot certificate |
| Copper foil roughness | Substantial above 10 GHz, small below 3 GHz | Foil type in the stack-up, not the laminate alone |
| Etch and line width | Moderate; also changes impedance | Etch compensation and process control |
| Finish and plating | Small per board, meaningful at mmWave | Finish thickness and nickel thickness on ENIG |
2. Laminate tolerance is a range, not a number
A datasheet gives a nominal Dk and Df with a tolerance, and the tolerance is usually stated for a specific test method at a specific frequency. The production lot will sit somewhere inside that window, and the window is wide enough to matter: a Df that moves by 0.0005 at 10 GHz changes the loss of a 100 mm line by a measurable amount, and a Dk that moves by 0.1 changes the impedance and therefore the match.
What to ask for: the lot certificate with measured Dk and Df, not just a compliance statement. On high-volume RF programs, the laminate supplier can screen to a tighter window for an agreed premium, and that is often cheaper than the yield loss it prevents.
3. Copper roughness is the hidden variable
At high frequency, current crowds to the surface of the conductor, and the surface is rough. The effective path length is longer than the physical length, so loss rises with roughness. Standard foil and low-profile foil differ enough that a stack-up change from one to the other can move insertion loss by tenths of a dB without any other change.
This is why the foil type belongs in the stack-up specification alongside the laminate part number. A fabricator that substitutes a different foil grade to meet a delivery date has changed the electrical performance, and nothing on the drawing prevented it.
4. Etch, width and impedance
Line width after etch determines impedance and, to a smaller degree, loss. A line that comes out 10 percent narrow is a higher impedance line with a slightly different loss, and both effects show up on the test bench. The compensation applied in the artwork is a process constant that drifts with etchant chemistry and copper thickness, and it has to be verified per lot, not set once.
On controlled impedance orders we measure coupons per controlled layer per lot and keep the record. The useful comparison is not one lot against nominal; it is the distribution across ten lots. A process that is centered but noisy is worse than one that is slightly off nominal but stable.
5. Finish
ENIG at mmWave is a nickel layer in the current path, and nickel is lossy. A 3 micron nickel deposit and a 6 micron deposit behave differently. On a 28 GHz board this is small but measurable, and on a 77 GHz board it is not small. Immersion silver and bare copper with an organic solderability preservative avoid the nickel, at the cost of shelf life and handling constraints.
6. Measurement repeatability first
Before blaming the boards, establish the measurement's own repeatability. The same board measured twice, removed and reconnected, will differ by some amount that has nothing to do with the product: connector torque, cable position, fixture contact, calibration drift. On a bench with SMA connectors and hand torquing, that number can be 0.05 to 0.1 dB at 10 GHz.
Establish it once. Measure one reference board ten times, with a full disconnect and reconnect each time, and record the spread. That number is the noise floor of the comparison. A lot-to-lot difference smaller than the noise floor is not a difference.
7. Building a control chart that means something
Pick one test structure per panel, measure it on every lot, and plot the result. Three signals matter: a step change, a trend, and an increase in spread. A step change means a material or process substitution. A trend means a consumable or a bath drifting. Increasing spread means the process is losing control even if the average is fine.
Keep the test structure identical between lots. A coupon design change resets the baseline and destroys the comparison, which is a common way to lose a year of data.
8. What to specify
Laminate part number with lot certificate including measured Dk and Df. Foil type and roughness grade. Finish type with nickel thickness where ENIG is used. Impedance tolerance and coupon test per layer per lot. Insertion loss target on a defined test structure, with the measurement method and the reference board. And the sample plan: how many boards per lot, measured where, reported how.
Without those six lines, the only way to find out why lot five measures worse than lot one is to section boards and guess. With them, it is a phone call to the material supplier.
Send the stack-up and the loss target and we will tell you which of these variables we can hold per lot and what evidence we will ship with the boards.
Related reading
- 50 Ohm Impedance Tolerance on RF PCBs
- Back Drilling: How Short a Stub Do You Actually Need
- Antenna-in-Package Assembly: What the Board Has to Get Right
- RF Laminate Dk Drift with Temperature and Humidity
- Rogers 4350B vs 4003C: Selection Guide
Need a DFM check or a stack-up review on your files? Write to pcba@superb-tech.com or WhatsApp +86 133 9608 1443.