Ask three engineers what “low-stress nitride” means and you get three numbers, two measurement methods and one argument. PECVD dielectrics are defined by properties, not by names — and a specification that lists only material and thickness leaves every property that matters undefined.

Here is what a usable film specification contains, and why each item constrains the others.

1. Stress — a target and a window

Stress is the property MEMS structural films live or die by. A membrane that bows, a cantilever that curls, a released structure that buckles: all stress problems that appear only after release, when the wafer has already accumulated its full value.

State three things:

  • The target, with sign — for example “−50 to +50 MPa” rather than “low stress”. Compressive and tensile failures look different and are not interchangeable.
  • How it will be measured — typically wafer-curvature before and after deposition, at a stated thickness. Stress is thickness-dependent; a number without a thickness is not a number.
  • Stability requirements — whether the value must hold after subsequent thermal steps, and after weeks of storage. Hydrogen-rich films drift; if your flow anneals afterwards, specify the post-anneal value, because that is the one your device sees.

2. Uniformity — across the wafer and across topography

Two different questions get collapsed into one word:

  • Thickness uniformity across the wafer, stated as a percentage with the range definition (±(max−min)/2·mean, or 1σ — say which, they differ by roughly a factor of two) and the edge exclusion.
  • Step coverage over topography — the ratio of sidewall and bottom thickness to field thickness, at your actual aspect ratio. A film that is beautifully uniform on a blank wafer can thin dangerously at the bottom of a deep trench.

If your device has topography, the second number is the one that decides yield, and it is rarely quoted unless you ask.

3. Thermal budget — the constraint that governs everything else

PECVD exists because it deposits at 100–400 °C instead of the 700 °C+ of LPCVD. That freedom has a price: at lower temperature, films retain more hydrogen, are less dense, and etch faster in wet chemistry.

So the specification must start from the ceiling your stack allows — aluminium metallisation, a bonded interface, a temperature-sensitive device layer — and ask what quality is achievable there. A supplier who quotes excellent film properties at 400 °C has not answered your question if your wafer can only take 250 °C.

4. Film quality metrics that map to your failure mode

Rather than asking for “high quality”, specify the property your device actually depends on:

If the film is a… Specify
Passivation layer Breakdown field, pinhole density, moisture barrier performance
Structural / mechanical layer Stress, Young’s modulus, thickness uniformity
Optical cladding or waveguide Refractive index and tolerance, absorption at your wavelength
Hard mask Wet-etch rate ratio to the underlying film, selectivity in your etch chemistry
Interlayer dielectric Dielectric constant, leakage, step coverage

Wet-etch rate (typically in buffered HF, quoted relative to thermal oxide) is the cheapest single proxy for film density, and worth including even when it is not a device requirement — it makes run-to-run drift visible.

5. Particles and productivity, at your thickness

Particle performance is a function of chamber-clean strategy and film thickness. A thick film campaign loads the chamber faster; the clean interval that works for 200 nm may not survive 2 µm. Ask for particle data at your thickness and over a stated number of consecutive runs, together with the clean overhead — because that overhead is part of your real throughput, not a footnote.

Putting it together

A specification a supplier can quote against looks roughly like this:

SiN, 800 nm ±5 % (1σ, 3 mm edge exclusion), stress −100 to 0 MPa measured by wafer curvature at 800 nm, deposition ≤ 300 °C, step coverage ≥ 60 % at 3:1 aspect ratio, BHF wet-etch rate ≤ 5× thermal oxide, stress stable to ±30 MPa after 30 min at 350 °C, ≤ 20 adders > 0.3 µm per wafer over 25 consecutive runs.

Every clause in that sentence is something a process engineer can design toward and a metrology plan can verify. That is the difference between a specification and a wish.

Working out a film specification, or comparing PECVD platforms against one? Send us the device context and we will help turn it into questions a manufacturer can answer precisely.