Spacer Process
The spacer process deposits a thin conformal film over an already-patterned feature and then etches it straight back with an anisotropic plasma, leaving material only on vertical sidewalls. That simple geometric trick does two very different jobs depending on where it's used: it protects a transistor's channel edge during implant, and — unrelated to transistors entirely — it doubles or quadruples the pattern density a lithography tool can print on its own.
The etch-back step works because deposition is conformal and etching isn't. A conformal film coats every surface — top, sidewall, and flat field — with roughly equal thickness. An anisotropic (directional) plasma etch, by contrast, strips material fastest where the ion bombardment hits straight down, which is everywhere except the near-vertical sidewalls, where the effective film thickness along the etch direction is far greater. Run the etch until the flat regions clear completely, and what's left standing is a self-formed sidewall spacer, with a width set almost entirely by how thick the film was deposited — no separate lithography mask required at all.
In a transistor, the spacer buys room for a second implant without a second mask. The gate itself self-aligns the main source/drain implant, as the poly process does, but a lightly-doped-drain (LDD) implant needs to sit slightly further from the channel edge than the heavy implant does, to soften the peak electric field and curb hot-carrier degradation. Growing a sidewall spacer on the already-patterned gate, then implanting again, automatically pushes the second implant's edge outward by exactly the spacer width — tunable simply by depositing a thicker or thinner spacer film, with zero additional alignment step.
Self-aligned double and quadruple patterning turned the spacer into a lithography tool. When immersion scanners couldn't resolve the pitches advanced logic nodes needed, the industry repurposed the exact same deposit-and-etch-back mechanic: print wide, easy-to-resolve "mandrel" lines, wrap a spacer around every mandrel sidewall, then etch the mandrel away entirely. What's left is a set of spacer lines at double the density the scanner alone could print — self-aligned double patterning (SADP). Repeat the whole sequence on the SADP result and you get self-aligned quadruple patterning (SAQP), another 2x. Below roughly the 14 nm node, SADP/SAQP (and later EUV, which relaxed but didn't eliminate the need) became load-bearing parts of the lithography roadmap — not because spacers are inherently a patterning technology, but because the geometry that makes a reliable transistor sidewall also makes a reliable, mask-free pitch splitter.
Spacer composition and etch selectivity decide whether the trick actually works. A spacer has to survive the subsequent implant, anneal, and any wet or dry clean without eroding, and in SADP it additionally has to etch at a dramatically different rate than the mandrel material underneath it, or the "remove the mandrel, keep the spacer" step fails. That's why spacer films are chosen as much for etch selectivity against their neighbors (nitride against oxide, or vice versa) as for any electrical property — the chemistry decision is really a patterning-fidelity decision in disguise.
A modern spacer is rarely a single film — it's usually a liner plus a bulk layer, each doing a different job. Seeing the real stack matters because the etch-back has to clear two different materials in sequence, not one:
That two-material clearance sequence is why spacer etch-back is itself a closed, instrumented loop — not a fixed-time strip. Over-etching into the liner risks gouging the junction; under-etching leaves stringers of nitride that short adjacent gates:
And isolated from the process-loop diagram, this is the actual trace the plasma-emission endpoint detector follows through the two-stage etch:
That two-stage discipline is exactly why a spacer's final profile looks nothing like a simple triangle under magnification. Real production spacers typically show a slightly rounded top corner where the bulk etch transitions to liner over-etch, and a distinct "foot" where the liner meets silicon — both artifacts of the recipe switch visible in the endpoint trace above, not defects. Process engineers read that exact profile from cross-section SEM images as confirmation the two-stage etch landed correctly, the same way the endpoint trace itself confirms it electrically.
Read the spacer process as proof that a purely geometric idea — anisotropic etching leaves sidewalls standing — can be repurposed across completely different problems. The same self-aligned sidewall that once existed only to soften a transistor's hot-carrier degradation became, a few decades later, the mechanism multiple-patterning lithography depends on to print features no scanner alone could resolve. And underneath both of those stories sits the same quiet discipline: a two-material filmstack, cleared in two instrumented stages, every single time.