Nickel Silicide Pipe Defect
# Nickel Silicide Pipe Defect
Silicide formation is supposed to leave a flat, uniform layer of NiSi sitting on top of the silicon, converting the exposed source, drain, and gate surfaces into a low-resistance contact without touching the junction underneath. Nickel doesn't always cooperate. Along certain crystal defects — dislocations, grain boundaries in the deposited nickel film, or contamination sites — nickel atoms diffuse abnormally fast straight down into the silicon, far ahead of the uniform reaction front everywhere else. The result is a narrow silicide spike, a "pipe," punching through the shallow junction it was supposed to sit safely above, shorting source or drain straight to the substrate.
Nickel's own silicide chemistry is what makes piping possible in the first place. Cobalt and titanium silicides, nickel's predecessors in the contact stack, both form through silicon diffusing up into the metal — a self-limiting reaction that's hard to run away with locally, since the metal has to keep being available at the surface. NiSi forms the opposite way: nickel is the dominant diffusing species, moving down into the silicon. That makes NiSi formation faster and lower-temperature than its predecessors, which is exactly why the industry adopted it once junctions got too shallow for cobalt's thermal budget — but it also means any local fast-diffusion path for nickel translates directly into localized, uncontrolled downward silicide growth instead of a self-limiting reaction.
The electrical signature is a device that leaks or shorts despite a design and layout that looks completely fine. A pipe that reaches the junction shows up as elevated leakage current between source/drain and substrate on wafer-level electrical test, and a pipe that goes deeper shows up as an outright short — both failure modes that, traced back through failure analysis, point to a single localized silicide spike rather than any systematic layout or implant problem. Because piping depends on where a crystal defect or contamination site happens to sit, it's a yield-detractor failure mode rather than a design-rule violation: two nominally identical dies can differ purely because one had a stray defect under its silicide and the other didn't.
Process control fights piping through the nickel film and the anneal, not through layout. A thin titanium or titanium-nitride capping layer deposited on top of the nickel before the first anneal — standard on production silicide modules — suppresses agglomeration and limits how much nickel is locally available to feed a runaway spike. Two-step rapid thermal annealing, forming a nickel-rich phase first at lower temperature then converting it to the final low-resistivity NiSi phase in a second, separate anneal, keeps the reaction more uniform than trying to drive straight to NiSi in one step, where local hot spots have more opportunity to run ahead. And pre-silicide surface cleans — removing native oxide and particulate contamination that would otherwise seed a localized fast-diffusion site — directly reduce the population of nucleation points piping needs to start from, which is why wet-clean quality in the silicide module gets tracked as carefully as the anneal recipe itself.
STI divot and bird's-beak both lose active area at a trench or isolation edge. Silicide piping fails a transistor from the opposite direction — a defect in the metal contact layer reaching down and punching through a junction that was otherwise placed and isolated correctly — a reminder that a shallow, high-performance junction creates new failure paths for the very contact layer built to make it useful.