Spend enough time in a fab and you notice a pattern: 90% of thin-film anomalies eventually trace back to fewer than ten typical modes. Identify them fast, and you save several DOE loops and a few wasted lots. This piece organizes the 5 CVD thin-film failure modes I hit repeatedly on the line — and that newcomers ask about most. Each comes with its symptom, mechanism and debugging path, kept as jargon-light as possible.
Thickness distribution across a lot or a single wafer clearly drifts from target, SPC goes out of control, and device electricals (e.g. threshold voltage Vth) show systematic intra- or inter-wafer drift.
Uniformity is jointly decided by gas flow field, temperature field and plasma distribution. Chamber aging, a partially clogged showerhead, uneven heater-plate temperature, or the loading effect can all unbalance the reactant flux profile across the wafer surface.
A sudden spike in the defect map, or ring-shaped, arc-shaped or localized cluster patterns; corresponding devices show leakage, shorts or a yield drop.
Particles usually come from three sources: chamber-wall / fixture flaking (film accumulates on the wall then sheds), gas-phase by-products condensing on the surface, and transfer / load introduced external contamination. Under a stable process, particle counts sit on a low baseline — a spike almost always means "something started flaking."
On high-aspect-ratio structures, the bottom / sidewall film is visibly thin, even showing a seam or void, leading to fill failure and reliability risks downstream.
This is a reaction-kinetics problem. Surface-controlled reactions give good step coverage but limited conformality; mass-transport-controlled reactions exhaust reactants inside deep holes, leaving poor bottom coverage. The higher the aspect ratio, the stricter the demand on precursor diffusion and surface-reaction path.
The film shows cracks, warpage, or delaminates from the underlying layer; it surfaces intensively after subsequent high-temperature or CMP steps.
Stress comes from two sides: intrinsic stress (microstructure during deposition, e.g. columnar grain growth) and thermal stress (CTE mismatch between film and substrate). In multilayer stacks, stress accumulates and, past a critical value, cracks or delaminates.
The film "looks beautiful" but resistivity, dielectric constant, impurity content or density miss spec — ultimately showing up as device leakage, breakdown or Vth drift.
This is the most hidden class, because the appearance is fine. Incomplete precursor decomposition, doping-concentration drift, residual O/C impurities, insufficient density all make a film "right in form, wrong in essence." It's often a downstream symptom of the previous four, but can also arise independently from gas purity or a process-temperature-window shift.
If you're preparing for a semiconductor process / device job hunt, or want to dig into real case studies of a particular failure mode, feel free to reach me via the home page. I'll also be updating "process-device correlation modeling" and "a 3D NAND device perspective" next.
← Back to home