Pixel Size

Pixel pitch is the one sensor number that touches everything else.

Sampling. The specimen a pixel sees is pitch divided by magnification — a length. Its area is that length squared, so a factor of 1.4 in pitch is a factor of 2 in area. Sensitivity per pixel follows that area directly. Full well scales with the photodiode's area, so larger pixels hold more charge before clipping. Data rate runs the other way: halving the pitch over a fixed sensor quadruples the pixel count and the bytes.

Representative pitches on current scientific cameras: 4.6 µm on the qCMOS class, 6.5 µm on most sCMOS including the Kinetix, and 11 µm on the Prime 95B, which trades resolution for a very large well and high quantum efficiency.

No pitch is best. The question is always which pitch, against which objective, for which measurement.

Sampling is a property of the pairing, never of the camera, and the same sensor moves between verdicts as the objective changes.

At 40× / 1.15 NA the optics resolve 297 nm, so Nyquist asks for pixels of 148 nm or finer. A 6.5 µm pixel gives 162.5 nm — just past the requirement, 1.09× coarse. A 4.6 µm pixel gives 115 nm and oversamples comfortably.

Move the same 6.5 µm camera to 25× and it samples at 260 nm, undersampling badly. Move it to 60× and it samples at 108 nm, comfortably inside Nyquist. Nothing about the sensor changed.

Which is why a camera cannot be evaluated on its own. It can only be evaluated in the configuration you will actually run.

Larger pixels collect more photons per pixel. This gets read as more sensitive, and it is not the same claim.

A pixel with twice the area sees twice as much specimen and collects twice as many photons from it. Per unit of specimen, nothing has improved. What has changed is how the field was divided, and per-pixel numbers computed on different divisions are not comparable — two cameras on one sample can read 37% apart per pixel and about 2% apart per resolution element.

There is one place the advantage is real rather than bookkeeping. Read noise is charged per pixel read, so collecting a fixed patch of specimen with fewer, larger pixels pays it fewer times. When you are read-noise-limited — below roughly 5 detected photons per pixel — that is a genuine gain, and it is the honest argument for a large pixel. Above about 10 photons, where shot noise dominates, it evaporates.

Start from the measurement, not the catalog.

Decide what has to be resolved, and let the objective follow: that fixes the resolution distance d = 0.61λ/NA. Take Nyquist as d/2 and multiply by the magnification you will use — that is the largest pitch that preserves your optics. Then check the photon budget at that pitch: if you land below about 5 detected photons per pixel, either accept a coarser pitch or buy the read noise down.

Worked, for 40× / 1.15 NA: d = 297 nm, Nyquist 148 nm, so pitch ≤ 148 × 40 / 1000 = 5.9 µm. A 4.6 µm camera qualifies; a 6.5 µm camera misses by 10%, which is usually survivable and should at least be a decision.

Two things this ordering protects you from. Buying resolution you cannot use because the photon budget will not support the pitch, and buying sensitivity you cannot use because the sampling throws the resolution away before the sensor ever sees it.