A solder mask does one quietly important job at reflow: it walls the molten alloy in. Take it away and every joint wants to misbehave — bridge to its neighbour, pull off-pad, or starve while the alloy wicks somewhere it should not. We ran a dense flip-chip build with no mask at all and held 100% post-reflow yield. The approach that looked like the risk is what made the run work.
The build
An RF program brought us a flip-chip assembly on aluminium nitride: four CMOS die, 118 I/O each, 472 joints per board, across 10+ boards. The solder was SAC305 on non-solder-mask-defined pads — NSMD, maskless.
AlN is the reason. Its thermal conductivity is why you pick it for RF power, but it is a ceramic, and the mask processes that work on organic laminate do not transfer cleanly. So the substrate came to us bare, and the question became whether joint geometry could be controlled without a dam.
Why maskless is hard
On a solder-mask-defined pad, the mask opening sets the joint footprint. The alloy melts, hits the mask edge, and stops. Geometry is defined by a physical boundary and reflow just has to not fight it.
Strip the mask and nothing constrains the alloy but the pad metallisation and surface tension. At 118 I/O per die the pads sit close enough that a joint spreading a little too far is a bridge, and a joint wetting a little too far up an adjacent feature is an open. Both failures come from the same cause — an uncontrolled reflow — and both are invisible until inspection.
What actually held it
- Fluxless, single-pass reflow. No flux residue to trap under a die that is about to be underfilled, and one thermal excursion instead of two. Every extra trip through reflow is another chance for a joint that was fine to stop being fine.
- NSMD pads characterised for wetting. With no mask, pad geometry and surface energy are the joint definition. Characterise them and surface tension becomes the tool that self-centres the die rather than the thing pulling it off target.
- SPC on the reflow profile. Ramp rate, peak temperature and time-above-liquidus held inside a controlled window, run to run. This is the part that does the work the mask would have done. A profile that drifts is a bridge you have not found yet.
- Post-reflow inspection on every joint. All 472 per board, confirmed before the board moved on. On a maskless build you do not sample.
The transferable point
Process control substitutes for physical constraint more often than people expect. When you cannot wall the solder in, you hold it with the profile — and you verify rather than assume, because the failure modes are the quiet kind.
It is also worth saying plainly that maskless is not automatically the harder path. It removes a process step, a material, and a tolerance stack. What it demands in exchange is that your reflow window is genuinely characterised rather than inherited.
If you are looking at something similar
Dense flip chip on ceramic, NSMD pads, or a substrate where the usual mask process does not apply — these are the builds where the reflow recipe is the product. Build data, X-ray imagery and design rules are available under NDA.
More on the six attach families we run, including C4 and copper pillar, on our flip-chip assembly page.