Bird Beak Lithography
# LOCOS Bird's Beak
Before shallow trench isolation existed, fabs separated transistors with LOCOS — local oxidation of silicon — and LOCOS had a built-in flaw baked into its own name's shape. Grow a thick field oxide by letting oxygen diffuse under a patterned silicon-nitride mask, and the oxidizing front doesn't stop cleanly at the mask edge. Oxygen creeps laterally under the nitride before the mask fully blocks it, and the oxide that grows there tapers to a thin wedge pushing into the active area — a cross-section shape so distinctively beak-like that the defect got its name from the silhouette itself, not the physics.
The nitride mask bends instead of blocking cleanly, and that bend is the whole problem. Silicon nitride resists oxidation far better than bare silicon, which is exactly why it works as a field-oxide mask — but it isn't perfectly rigid, and it isn't a perfect oxygen barrier at its edge. As the field oxide grows thick underneath the open area, the expanding oxide pushes up against the nitride edge and physically lifts it, letting oxidant diffuse laterally underneath the mask for some distance before the oxidation front finally dies out. The oxide that forms during that lateral creep thins out gradually from full field-oxide thickness down to nothing, right at the boundary of what was supposed to be the clean edge of the active area.
That lost real estate is exactly why LOCOS couldn't keep scaling. Every bird's beak eats into the active-area width the designer actually drew, and the beak's lateral reach doesn't shrink proportionally with the rest of the layout — it scales more with field-oxide thickness and oxidation time than with the shrinking minimum feature size around it, so as a process node's active-area dimensions got smaller, the beak consumed a larger and larger fraction of the intended transistor width. Nitride edges also transmit mechanical stress into the underlying silicon as the oxide grows and expands against them, generating dislocations right at the active-area boundary when the stress exceeds what the lattice can absorb — a reliability problem riding on top of the area problem.
The industry tried masking tricks before abandoning LOCOS outright. A pad-oxide layer under the nitride (rather than nitride directly on silicon) relieves some of the stress by giving the nitride edge a little vertical compliance; a sacrificial poly-silicon buffer between pad oxide and nitride, in "poly-buffered LOCOS," further damps the stress and shrinks the beak somewhat. Both bought a few more process generations, but neither eliminated the fundamental mechanism — oxidant will always find a lateral path under any finite-thickness mask edge sitting on top of an oxidizing surface, and shrinking the beak with better masking just trades off a larger, more expensive thermal budget against shrinking it by less. Shallow trench isolation sidesteps the physics entirely by etching a physical trench and filling it with deposited oxide instead of growing oxide laterally under a mask, which is why STI replaced LOCOS once trench etch and gap-fill CVD became manufacturable at the dimensions LOCOS could no longer serve — the transition that fabs like Intel and the rest of the industry made industry-wide through the 1990s.
STI isolation has its own encroachment problem, the divot, but it comes from wet-chemistry recess after the trench is already flush rather than from oxidant creeping sideways during growth. LOCOS's bird's beak and STI's divot are the same underlying story told twice — isolation geometry degrading at the active-area edge, once from lateral oxidation and once from wet-clean recession — and the industry's move from one to the other was a trade of one edge-loss mechanism for a smaller, better-controlled one, not an elimination of edge loss altogether.