Reflow Oven in High-Frequency PCB Production: The Profile Is Measured, Not Set

A reflow oven does not have a temperature. It has a recipe — zone set-points, belt speed, gas atmosphere — and what the board actually experiences is the result of that recipe acting on a particular thermal mass. Which is why the profile on the screen is a hypothesis until a thermocouple inside a real assembly confirms it. On a high-frequency build the stakes are higher than on FR-4, because the same heat that melts the solder also expands the laminate by 80 µm through the board thickness, and the barrel has to survive that.

A convection reflow oven is a tunnel with six to ten heated zones and two or three cooling zones, through which a belt carries the assembly at 0.6–1.2 m/min. What is set is the air temperature in each zone. What the board sees is a curve that depends on the zone set-points, the belt speed, the thermal mass of the assembly, and the atmosphere.

That distinction explains the most common production argument in any SMT line: two assemblies run through the same recipe and come out with different peak temperatures, because one is a 0.8 mm bare board with a few passives and the other is a 2.4 mm board with a shielded module on it. The recipe is a machine setting; the profile is a property of the product. Anyone who quotes a peak temperature without saying which assembly it was measured on is quoting a number that does not exist.

The convection figure matters on a high-frequency assembly for a reason that is not obvious. The laminates conduct heat poorly — 0.20 W/m·K for 5880 against about 0.3 for FR-4 and 400 for copper — so the board does not equalise internally as fast as an FR-4 board does. Heat has to arrive through the surface by convection, and a low-conductivity laminate therefore needs a longer soak to reach a uniform temperature, not a faster ramp.

The profile is conventionally split into preheat, soak, reflow, time above liquidus and cooling. The names are standard; the purposes are frequently mis-stated, and one of them — that a soak zone raises the board temperature quickly — is exactly backwards.

The soak stage is the one worth defending. Its job is equalisation, not heating: a thick assembly with a large ground plane will lag the air by 20–30 °C at the end of a fast ramp, and if the ramp continues into the melt the corner of that board reaches liquidus late and stays above it briefly, which is how a cold joint forms next to a perfectly reflowed one on the same panel. A flat soak of 60–120 s at 150–200 °C removes that gradient. On a low-conductivity RF laminate the same logic applies with more force, because the internal gradient is larger to begin with.

Two numbers define the reflow event, and they trade against each other. Peak temperature sets whether the joint wets; time above liquidus sets how much intermetallic grows. Enough of the first and not too much of the second is the whole objective.

For SAC305, solidus is 217 °C and liquidus 220 °C. A peak of 235–245 °C gives 15–25 °C of superheat, which is the accepted window for a eutectic-behaving alloy on a copper finish. A peak of 250 °C adds nothing to wetting and everything to intermetallic growth and to laminate expansion. A peak of 225 °C is technically above liquidus and produces joints that look fine and fail on thermal cycling, because the wetting time was too short to form a proper fillet. Leaded paste is a different profile entirely — 183 °C eutectic, peak 205–215 °C — and a specification that says “reflow at 200–250 °C” covers neither case correctly.

The superheat number also decides what happens to the laminate, and that is where a high-frequency build diverges from a digital one. The expansion through the board thickness is proportional to the temperature excursion, so every degree of unnecessary peak is paid for in barrel strain. Reducing a peak from 250 to 240 °C on a 1.6 mm 5880 board removes about 3.8 µm of Z expansion — small in absolute terms, but it is 3.8 µm out of a barrel elongation budget that is already being spent three times if the assembly is reflowed three times.

The idea that fast cooling prevents a cold joint is a persistent one and it is wrong. A cold joint — a dull, poorly wetted, lumpy fillet — is caused by insufficient heat or insufficient flux activity during melting, or by disturbance while the joint is mushy. Cooling rate does not enter it.

What cooling rate does control is the microstructure and the stress. Faster cooling refines the grain and improves fatigue life up to a point; beyond about 6 °C/s the thermal gradient across the assembly starts to matter, and on a large board with a heavy ground plane the difference between the top and bottom surface during cooling produces the warp that then shows up as a corner joint crack. The practical window is 2–4 °C/s, and the reason to hold it is warpage and stress, not wetting.

For a high-frequency assembly there is a second reason to slow the cool. The Z contraction on the way down is the same 80 µm as the expansion on the way up, applied in reverse to a barrel that has just been formed at peak temperature. A fast cool puts that strain in quickly, while the solder is still near its weakest; a controlled cool lets the assembly contract more evenly. The barrel and its ductility requirement are covered in plating.

The profile that matters is the one measured inside a real assembly with thermocouples attached, not the one the oven displays. Oven displays show air temperature in a zone, and the air is not the board.

Attaching the thermocouple is a skill in itself. The bead has to be in contact with the joint or the pad being measured, not floating in the air above it, and it has to be held with high-temperature tape or adhesive that survives 250 °C without lifting. A bead attached with ordinary tape drifts off the board at about 200 °C and then reports air temperature for the part of the curve that decides the result. Wire diameter matters too: 0.1–0.25 mm is fast enough to follow a 2 °C/s ramp; thicker wire lags and understates the peak.

The delta-T row is where a high-frequency assembly usually fails its first profile. A low-conductivity laminate plus a thick copper ground plane gives a board that is hot at the edges and cool in the middle, or the reverse depending on the plane distribution, and a single thermocouple on the edge will report a comfortable 240 °C while the centre is at 228. The fix is usually a longer soak rather than a higher set-point, because raising the peak to bring the centre up puts the edge well above the laminate’s comfort zone.

Here the reflow oven stops being a soldering tool and becomes a materials test. Going from 25 to 240 °C is an excursion of 215 K, and the laminate expands through its thickness by its Z-axis coefficient times that excursion times its thickness.

The barrel plating has to stretch by that amount without cracking, which is why ductility is specified at 20 % elongation or better and why the aspect ratio ceiling on PTFE is about 6:1. It is also why reflow cycles are a finite resource on a PTFE assembly. Three reflows — one for the top side, one for the bottom, one for a rework — apply the excursion three times, and the third is applied to a barrel that has already been work-hardened twice. A process that plans for two reflows and discovers a third at rework is planning to fail a thermal cycle test it never ran.

Two related points are worth stating because they are frequently confused. PTFE has no glass transition in the usable range — its crystalline melt is around 327 °C — so the usual “keep below Tg plus 25” rule does not apply; the relevant coefficient is constant across the whole excursion, which makes the expansion worse rather than better. And the material cold-flows, so clamping or fixturing an assembly through reflow can thin the dielectric and move the impedance; a fixture that holds a PTFE board flat is a fixture that changes its electrical properties.

Nitrogen is not automatically better; it is better for specific finishes. At 1,000 ppm O₂ or below, the oxidation of an immersion silver surface during the ramp is suppressed enough to keep the wetting force high, and solder balling on fine-pitch paste drops. On an OSP finish the benefit is real but smaller. On ENIG the nickel barrier already protects the joint and the gain is marginal.

The trade is cost and, on a high-frequency board, one more thing: nitrogen changes the heat transfer slightly and therefore the profile. A recipe validated in air does not transfer to nitrogen without re-measuring, and the difference is typically 3–6 °C at the peak. Shops that switch atmosphere to chase a wetting problem often create a peak-temperature problem at the same time. The finish options and their thickness windows are set out in surface treatment.

An oven is accepted on the profile it produces on your assembly, not on its zone count. The following eight checks cover what a zone-count specification leaves out.

Check 4 is the one most often quoted from a datasheet and least often measured. Cooling rate depends on the mass being cooled, so a machine that achieves 4 °C/s on a 0.8 mm test coupon may achieve 1.5 °C/s on a 2.4 mm assembly with a shield can. Measuring it on the real product is the only way to know whether the cooling zones are sized for the work.

1. “The oven reflows between 200 and 250 °C.” That range spans two alloy systems and includes temperatures that are too cold for SAC305 to wet properly and too hot for a leaded paste. A profile is a curve with a peak, a time above liquidus and a cooling rate, not a range.

2. “Fast cooling prevents cold joints.” A cold joint comes from insufficient heat or flux activity, or from movement while the joint is mushy. Cooling rate affects grain structure, warpage and stress. Quoting it as a cure for cold joints leads to ovens sized for cooling rather than for uniformity.

3. “Reflow soldering forms the connections between the layers of a multilayer board.” Layer-to-layer connection is plated copper in a drilled and metallised hole; it is formed at the plating bath, long before any paste is printed. Reflow connects components to the surface. What reflow does to the layers is stress them — 82 µm of expansion on a 1.6 mm PTFE board, applied to every barrel, once per pass.

Three things decide whether a reflow oven is right for a high-frequency build, and none of them is the number of zones. The first is uniformity: can it hold the whole assembly within a few degrees through a soak long enough to equalise a low-conductivity laminate. The second is control at the top: can it land a peak of 235–245 °C with 45–90 s above liquidus, repeatably, on the heaviest assembly in the mix. The third is the cooling curve, which is a warpage and stress parameter rather than a wetting one.

The discipline that ties them together is measurement. A recipe is a machine setting; a profile is measured on the product, with thermocouples attached to the joints that matter and with a delta-T that has been checked. Everything the oven does to the laminate follows from the profile, and the laminate has the last word: 82 µm of expansion per pass on a 1.6 mm PTFE board, once for each time the assembly goes through.

XCE PCB builds and supports assembly on conventional and high-frequency substrates: 2 to 72 layers, PTFE, hydrocarbon and ceramic laminates, 600×1200 mm panels, impedance held to ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G, IPC-6015 and IPC-A-610. The stage sequence around the oven is in the PCBA manufacturing process, and the inspection gates after it in PCBA quality control.