Wafer Dicing Blade Saw Laser Stealth Dicing Die Singulation Scribe Line Kerf Dicing Tape Low-K Dielectric Thin Wafer Stacked Packaging

Once wafer probe has finished electrically testing every die and recorded a digital wafer map of known-good versus known-bad die, the wafer still has to become a stack of individual, physically separate pieces before any of those good die can go anywhere near packaging. That separation step is wafer dicing, and it has to happen without destroying the very thing wafer probe just finished certifying as good. Every die on a modern wafer is laid out with a deliberate blank margin between it and its neighbors — the scribe line, or street — specifically so a cutting tool has somewhere to travel that isn't active circuitry, and dicing always targets those streets rather than any patterned area.

Wafer Dicing: From One Probed Wafer to Many Singulated Die Probed wafer, digital wafer map marks good vs. bad die Wafer mounted face-up on dicing tape stretched across a frame Method chosen by wafer thickness, BEOL dielectric, die size Cutting always travels the blank scribe-line streets between die — never through patterned circuitry — whichever method is picked next.

Mechanical blade dicing is the mature, decades-proven default: a diamond-coated blade spinning at very high RPM is drawn straight through the wafer along each street, with a water jet running continuously to cool the blade and flush away the silicon particulate the cut generates. The wafer sits mounted on dicing tape the entire time, so even after every street has been cut, the resulting individual die stay held exactly in place on that tape rather than scattering. Blade dicing removes a real strip of material as it cuts — the kerf — which is simply lost as debris, and that mechanical cutting action puts direct physical stress on every die edge it passes through.

Laser stealth dicing takes a fundamentally different approach: instead of cutting at the surface, a laser tuned to a wavelength silicon barely absorbs at the surface is focused to a tight point inside the wafer's bulk, along each street, where it locally modifies the silicon into a weakened subsurface layer without ablating or removing any material at all. No kerf is lost, no particulate is generated at the surface, and no water is involved. Once every street has been scanned this way, the wafer — still on tape — is mechanically expanded, stretching the tape outward just enough that a clean crack propagates along each subsurface modified line, separating the die from each other with far less mechanical and thermal stress than a blade ever puts on the die edge.

Two Ways to Separate a Wafer: Blade Saw vs. Laser Stealth Same streets, same tape-mounted wafer, very different physics along the cut Wafer mounted on dicing tape, scribe-line streets marked streets = blank lanes between die Blade / Saw Dicing Laser Stealth Dicing wafer on dicing tape diamond blade, high RPM water jet kerf: material cut away as debris Blade physically cuts down through the full wafer thickness, removing kerf material and stressing both die edges as it passes. wafer on dicing tape laser focused below surface subsurface layer, then tape expansion cleaves it Laser modifies silicon inside the bulk only; no surface ablation, no kerf loss, no water — tape stretch alone cleaves the clean crack. Blade dicing removes real material at every street; stealth dicing removes none — the die separate along a modified plane instead Thin wafers and fragile low-k BEOL dielectric stacks tolerate stealth dicing's lower mechanical and thermal stress far better Blade / kerf loss Laser / no kerf loss After Singulation: Die Held on Tape, Sorted by the Wafer Map All die still held on tape/frame; green = known-good per wafer map, picked for packaging Known-bad die (red) are identified by the wafer map and discarded — only good die proceed to pick-and-place. Pick-and-place removes individual good die from the tape/frame for packaging
Blade / Saw DicingLaser Stealth Dicing
Kerf / material lossReal kerf width cut away and lost as debris at every streetNo kerf — no material removed, only a subsurface modified layer
Debris / particle generationContinuous silicon particulate, managed with running water-jet cooling and flushNone at the surface; the modification happens entirely within the bulk
Mechanical stress on die edge / BEOLDirect mechanical cutting stress at every edge; low-k dielectric films are mechanically fragile and prone to cracking or delamination under this stressTape-expansion crack propagation only; substantially lower stress on fragile low-k BEOL stacks
Suitability for thin wafersChipping and breakage risk climbs sharply as wafers are thinned for stacked packagingWell suited to very thin wafers; the clean subsurface crack tolerates thinness far better than a blade
Process maturity / costDecades-proven, simple, lower equipment costNewer, more complex optics and process control, higher equipment cost
Typical use caseThicker wafers, robust BEOL stacks, high-volume cost-sensitive dicingAdvanced nodes, thin wafers for stacked/3D packaging, low-k-sensitive BEOL stacks

Choosing between the two isn't arbitrary — it's dictated by exactly the characteristics wafer probe and the fab's own process already fixed before dicing ever starts. A thick wafer with a conventional, mechanically robust interconnect stack has little to gain from stealth dicing's extra cost and complexity, so blade dicing remains the default. But as wafers are thinned aggressively for stacked packaging, and as back-end-of-line interconnect stacks increasingly use low-k dielectric films specifically to cut interconnect capacitance, those same films become mechanically fragile in exactly the way a blade's direct cutting stress punishes — cracking or delaminating under stress a laser-cleaved edge simply never experiences. That tradeoff is why laser stealth dicing, despite being the newer and costlier option, keeps gaining ground on advanced, thin, or low-k-sensitive wafers even though blade dicing remains perfectly adequate everywhere else.

Whichever method cuts the streets, the die coming off this step are still sitting exactly where they were cut, held in place on the dicing tape and frame, which is what makes the next step possible at all. This is also the point where the wafer map built during wafer probe stops being a passive record and starts actively directing the process: equipment reads that map to know, die by die, which of the now-separated pieces are the known-good ones worth sending onward and which are the known-bad ones to discard on the spot, well before any packaging equipment or material gets anywhere near them. The good die are then picked individually off the tape and routed into whatever packaging flow and SKU that particular device is destined for.

Wafer dicing is the physical transition point where a wafer-scale manufacturing process becomes, for the first time, a collection of individual die. The method chosen here has to respect everything the chip's construction already committed to upstream — a thin wafer or a fragile low-k BEOL stack narrows the safe choices sharply, which is exactly why stealth dicing exists as an alternative to the mechanically simpler blade saw. And none of this separation is blind: it's the wafer map handed down from wafer probe that makes dicing economically selective rather than indiscriminate, ensuring that only the die already proven worth packaging are the ones that actually move forward into it.

Take wafer dicing blade saw laser stealth dicing die singulation scribe line kerf dicing tape low-k dielectric thin wafer stacked packaging further

Ask the copilot about this term, or have our engineers assess it against your process.