FinFET Fin Patterning Self-Aligned Double Quadruple Patterning Sadp Saqp Sidewall Image Transfer
Before a gate can be formed, a spacer etched, or an epitaxial source/drain grown, there has to be a fin to build all of it on, and at FinFET nodes that fin is narrower and more tightly pitched than a single conventional lithography exposure can resolve, which is why the fin pattern itself is created by a sidewall-image-transfer technique called self-aligned double patterning (SADP) rather than by printing the fin shape directly. A 193nm immersion scanner can reliably print a line/space pitch down to roughly 80nm, but FinFET-era fins at 22nm and 14nm need pitches closer to 40-50nm, and fins at 10nm-class and 7nm-class nodes need pitches tighter still, so instead of asking the litho tool to resolve the final fin pitch, SADP prints a coarser, easily resolvable "mandrel" pattern and then uses a deposited sidewall film to split that mandrel pitch in half, with the final fin pattern coming from the sidewalls rather than from the original exposure. This sub-lithographic trick is what lets a 193nm scanner that cannot print a 45nm pitch directly still produce fins at that pitch, by turning the pitch-division problem into a thin-film deposition and etch problem instead of an optics problem.
The sequence starts with mandrel lithography: a sacrificial material, typically amorphous silicon or a patterned hard mask, is printed and etched at the coarsest pitch the litho tool can resolve comfortably, roughly twice the final target fin pitch, and this mandrel pattern carries none of the tight-pitch burden itself. A conformal film is then deposited over and around every mandrel line using the same conformal-deposition concept already covered in this batch for the contact etch-stop liner, except here the conformal film is a sacrificial spacer material such as silicon nitride or silicon oxide rather than a stress liner, and its thickness, not the litho tool, is what sets the final fin pitch, since a thinner or thicker conformal spacer directly shifts where the sidewall ends up relative to the mandrel edge. An anisotropic spacer etch-back then removes the conformal film from the top of each mandrel and from the flat field regions between mandrels while leaving vertical sidewall spacers standing against each mandrel's flanks, and this is the same anisotropic reactive-ion-etch physics already covered in gate-spacer-rie-etch-profile-control, where directional ion bombardment clears horizontal surfaces faster than vertical ones; the chemistry and hardware are the same RIE toolkit, just reused here for pitch-splitting a sacrificial mandrel instead of shaping a transistor gate's spacer profile.
Mandrel removal is the step that actually delivers the pitch-doubling benefit: a selective etch strips the sacrificial mandrel material away entirely, leaving behind only the sidewall spacers, which now stand alone at half the original mandrel pitch and twice the original line density, since each single mandrel line produced two spacer lines, one on either flank. Those freestanding spacers are then used as the etch mask for a final pattern-transfer step that cuts the pattern into the hard mask and ultimately into the silicon fin material beneath it, so the fins that come out of this sequence never existed as a single lithographic exposure; they are a geometric byproduct of a sacrificial mandrel's edges. Running the entire SADP sequence a second time, depositing and etching a second conformal spacer around the already-doubled spacer pattern before transferring it again, is self-aligned quadruple patterning (SAQP), and it divides the pitch by four relative to the original mandrel rather than by two, which is why SAQP shows up at 10nm-class and 7nm-class nodes where even SADP's halved pitch is still too coarse for the target fin spacing; SAQP is correctly understood as SADP run twice, a spacer built on top of a spacer, rather than as a fundamentally different patterning scheme.
| Property | Direct single-exposure litho | SADP (self-aligned double patterning) | SAQP (self-aligned quadruple patterning) |
|---|---|---|---|
| Achievable pitch / division factor | Limited by scanner resolution, no division | Half the mandrel litho pitch (2x density) | Quarter the mandrel litho pitch (4x density) |
| Process step count / cost | Lowest; one litho + one etch | Moderate; mandrel litho + spacer deposit + etch-back + mandrel removal + transfer | Highest; SADP sequence repeated twice before final transfer |
| CD uniformity control challenge | Controlled directly by litho CD and focus/dose | Spacer film thickness uniformity across the wafer becomes the dominant CD control lever | Two compounding spacer-thickness budgets; CD variation can accumulate across both cycles |
| Typical adoption node range | Relaxed-pitch nodes where litho alone resolves target pitch | Widely used ~22-14nm FinFET fin patterning | 10nm/7nm-class and tighter, where SADP alone is still too coarse |
The pitch-doubling and pitch-quadrupling benefit is easiest to see geometrically: start with an original lithographic mandrel pitch P, which is the tightest line/space spacing the scanner can print reliably, and after one SADP cycle the sidewall spacers sit at P/2 with twice the line density of the original mandrel pattern, and after a second cycle (SAQP) the pattern sits at P/4 with four times the original density, all without ever asking the scanner to resolve anything finer than P. This is why SADP and SAQP are described as sub-lithographic techniques rather than advanced lithography: the optical resolution limit never moves, but the usable pattern density on the wafer keeps climbing because each additional spacer cycle borrows geometry from the sidewall of the previous cycle's pattern instead of from a new exposure.
Read against the rest of this FEOL arc, SADP/SAQP fin patterning sits one step further back than anything this batch has covered so far: it is the geometric foundation that every later module assumes already exists. The HKMG gate module wraps a gate stack around a fin that this sidewall-image-transfer sequence already defined; gate-spacer-rie-etch-profile-control reuses the exact same anisotropic RIE physics introduced here for pitch-splitting, just redirected toward shaping a transistor spacer instead of a sacrificial mandrel; source-drain-epitaxy-strained-channel-process grows strained silicon or silicon-germanium directly onto the fin sidewalls that SADP or SAQP cut into the substrate; and everything downstream, from silicide formation through CESL deposition, ILD0 gap-fill, and contact plug fill, happens on top of or around a fin whose width, pitch, and sidewall geometry were locked in at this patterning step, long before any of those later modules ever touch the wafer. A fin pitch of roughly 40-50nm was typical at early FinFET nodes using SADP alone, and advanced nodes needing tighter spacing moved to SAQP precisely because halving the pitch once was no longer enough; either way, every transistor this batch has built so far was built on a fin whose geometry was fixed here, which makes SADP/SAQP fin patterning the true starting point of this FEOL process arc rather than an optional prerequisite.