Laser Dicing Stealth
# Laser Stealth Dicing
Cutting a finished wafer into individual dies used to mean dragging a diamond-coated blade or a plasma etch straight through the silicon along the scribe lines, removing a strip of material — the kerf — and generating debris, chipping, and mechanical stress along every cut edge. Stealth dicing skips cutting the surface at all. A tightly focused infrared laser, tuned below silicon's bandgap so it passes through the bulk without burning the surface, is focused to a point deep inside the wafer along each scribe line, locally melting and rapidly re-solidifying a thin internal layer. That internal damage layer becomes a built-in weak plane, and a separate, gentle mechanical expansion step cleaves the wafer along it cleanly, with no kerf loss and almost no surface debris.
The physics works because silicon is transparent to the right wavelength, and opaque everywhere the laser is actually focused. A standard IR dicing laser, typically around 1064nm or longer, sits below silicon's absorption edge, so the beam travels through the bulk of the wafer largely unabsorbed until it reaches the laser's focal point — the one place the photon density is high enough to drive nonlinear, localized absorption. At that focal spot, and only there, the silicon heats past its melting point in a volume a few microns across, then re-solidifies within microseconds into a different, more strained crystal structure. Scan that focal point along every scribe line at a fixed depth, and the wafer ends up with a string of these internal modified-layer segments stitched together — a fault line sitting entirely inside the material, invisible from either surface.
The actual separation happens afterward, mechanically, and gently. Writing the internal fault line doesn't pull the dies apart by itself — the wafer is mounted on an expandable dicing tape, and that tape is stretched radially after the laser pass, applying a uniform, modest tensile stress across the whole wafer. The internal modified layer is weaker than the surrounding crystal, so the applied stress concentrates there and the wafer cleaves cleanly along every scribe line simultaneously, the same way a scored piece of glass snaps along the score rather than anywhere else. Because the crack initiates and propagates through a plane the laser already prepared, the resulting die edges come out straighter and with less chipping than a blade cut leaves behind, and since no material was physically removed, dies can sit closer together on the original layout — real area recovered simply by not cutting a kerf.
This matters most for exactly the dies that can't tolerate a rough edge. Thin dies destined for stacked packaging — the kind of ultra-thin logic and memory dies TSMC's CoWoS and SoIC flows or advanced HBM stacks depend on — are mechanically fragile enough that blade-dicing chipping can propagate a crack straight through the active circuitry, not just the scribe line. Disco Corporation, the dominant supplier of both blade and stealth dicing equipment, and laser toolmakers like Hamamatsu that build the focusing optics, have pushed stealth dicing specifically for this thin-die, high-stack-count packaging segment, where kerf loss and edge damage both translate directly into yield loss on dies that already cost the most to build.
STI divot and bird's beak lose silicon at an isolation edge drawn early in the flow; silicide piping and cobalt agglomeration fail features built mid-flow. Stealth dicing operates at the very end, on a wafer that's already a finished, fully processed product — proof that even the last mechanical step before packaging carries its own physics-driven failure mode to engineer around, not just a simple cutting operation.