Delamination in reflow is moisture and resin, not solder. A finished FR-4 board carries 0.1 to 0.3 percent of its own weight in water, and at 260 C that water is steam at roughly 1,600 times its liquid volume with nowhere to go, so it lifts the resin off the glass bundle or off the copper below it. Where the resin was also under-cured, or starved at a glass bundle during pressing, the same steam opens the interface at a fraction of that pressure. The defect is made in the bare board and only becomes visible when the assembly is heated.
The second mechanism is expansion. A 1.6 mm FR-4 board grows about 54 um in Z between 25 C and 260 C, taking 50 ppm per degree below Tg and 250 ppm above it. At 0.8 mm the figure is 27 um, at 2.4 mm it is 81 um. The copper plane on the surface grows at 17 ppm and stays where it is, so the resin has to absorb the whole difference across a dielectric that may be 0.1 mm thick. That is why separation starts where a large plane meets a thin layer, and why a board that passes at 235 C can fail at 260 C.
Where the separation starts
Under a large copper plane. Vapour trapped under a plane has to travel sideways to the nearest edge, and that distance is measured in millimetres. The blister appears at the plane edge, or over a via that pierced the plane, because that is where the pressure finds the shortest path. Isolated pads and 0.15 mm traces almost never blister on their own: they have an edge every few hundred micrometres.
On a drilled hole wall. Drilling leaves resin smear and a receded resin ring at the glass bundles, and desmear has to remove both without opening the glass weave. When desmear is short, the plating sits on a weak layer, and the barrel expands 54 um with the rest of the board while the resin around it lets go first at the knee. Hole wall separation and delamination are often the same event seen in two places, which is why a slice taken at the hole explains a blister 20 mm away.
At a routed edge or a V-score. Routing cuts through glass bundles and leaves them exposed with no resin cap over them. That edge is a direct path for water into the laminate, which is why boards stored for months and then reflowed separate at the profile before they separate anywhere else. A 0.2 mm resin margin at the outline costs nothing and closes the entry point.
Why a two hour bake does not fix it
| Where the water is | How much | What removes it |
|---|---|---|
| Warehouse storage, 30 days at 30 C and 70 percent RH | 0.2 to 0.3 percent through the full thickness | 125 C bake to J-STD-033, of the order of a day on 1.6 mm |
| After drilling and plating, before mask | 0.1 to 0.2 percent, highest at the hole walls | 150 C for 4 to 6 hours before the mask step |
| After mask cure | solvent held in a 20 to 30 um film | the cure profile of the ink, not a bake |
| Under a large plane after assembly storage | 0.1 percent is already enough | dry pack plus bake before the first reflow |
Water moves through laminate by diffusion, and diffusion time scales with the square of the distance. Two hours at 120 C dries the outer 0.2 mm of a 1.6 mm board and leaves the core untouched, and the core is exactly where the large planes sit. This is why a bake that worked on 0.8 mm control boards fails on a 2.4 mm backplane, and why the bake record has to name thickness and time rather than just state that a bake happened.
Three things that get read wrong
Measling read as cosmetic. White spots at glass bundles are resin and glass separating, and they pass a visual check because nothing has lifted yet. They fail six reflow passes. If a lot shows measling at goods in, the lot has a press or a cure problem, and the next thing to check is the press cycle count and the ramp, not the board's appearance.
Changing the laminate before checking the hole. When boards delaminate, the first reaction is to blame the material. Run the float test on the same lot first: 288 C for 10 seconds per IPC-TM-650 2.6.8 on an unplated coupon. If the coupon passes and the boards with 0.3 mm holes fail, the material is not the cause and the new laminate will fail in the same place.
Accepting a lot on a visual plus one cross-section. A section proves one plane. Separation between inner layers needs an acoustic image, and the stress step has to happen before the section is cut, otherwise the report describes a board that was never heated. Ask for the float result, six simulated reflow passes at the peak the assembly will actually see, and C-SAM whenever a board carries planes larger than 10 mm by 10 mm.
What SUPERB PCB holds on a build like this
Tg 130, 150 and 170 available, board thickness 0.6 to 3.2 mm, up to 68 layers on the high layer line, a maximum of 7 press cycles, 288 C float on every lot, slice and impedance reports released with the lot, moisture barrier packing with the bake record attached, and warpage held to 0.75 percent for SMT, 0.5 percent on HDI and rigid-flex. Pre-assembly bake to J-STD-033 is applied whenever thickness and storage time call for it.
The process windows are on FR-4 process parameters, the reliability side of the same question is on high reliability PCB, and the Tg decision is in which Tg to use for lead-free reflow. SUPERB Tech reports the float result with the lot because delamination is a property of the lot, and a certificate that says passed without a thickness and a time is not a record.
What to send with the question
Layer count and stack, Tg, board thickness, the largest copper plane and the dielectric thickness above it, smallest hole and aspect ratio, storage time and packing, reflow peak and time above liquidus, number of passes including rework, and the exact location of the separation. Separation at a plane edge points at moisture and press. Separation at hole walls points at desmear. Separation at the outline points at routing and storage. Those three lead to different corrective actions, and none of them is answered by a photograph of a blister.