Full Well Capacity

Full well capacity is the maximum charge a pixel's storage node can hold before it saturates — the ceiling directly paired with read noise's floor, and together the two numbers define a sensor's dynamic range exactly: how many distinct signal levels fit between the quietest reliably-measured signal and the largest signal the pixel can hold before it simply stops responding.

Full well capacity is, at its physical core, the same Q equals C times V relationship this series has already built an entire noise mechanism around. A pixel's storage node — a photodiode's own junction capacitance, or a dedicated storage capacitor in some sensor architectures — holds a maximum charge set by that node's capacitance multiplied by the maximum voltage swing the node can tolerate before breakdown, pinning, or simple loss of linearity sets in. This is the identical charge-capacitance relationship kT/C noise already established as fundamental to any charge-storage node in this series — the difference is that kT/C noise asks how much random thermal charge rides along for free, while full well capacity asks how much deliberate signal charge the same node can actually hold at the top end.

Dynamic range is the direct ratio of these two numbers, both measured in the same units, which is exactly why read noise and full well capacity are always specified together rather than independently. Dynamic range equals full well capacity divided by read noise, both in electrons, commonly expressed in decibels as twenty times the base-ten logarithm of that ratio — a sensor with a hundred-thousand-electron full well and a three-electron read noise delivers roughly ninety decibels of dynamic range, meaning it can simultaneously resolve a signal near its own noise floor and a signal tens of thousands of times brighter within the same single exposure, without either end clipping or disappearing into noise.

Pixel size sets full well capacity directly, which is why sensor dynamic range and sensor resolution pull against each other on a fixed die area — the same area-for-capacitance trade this series has already encountered from the noise side. A larger pixel has more physical area to build storage capacitance into, directly raising its full well capacity and therefore its dynamic range, but a fixed sensor die can only fit so many large pixels, trading spatial resolution away to get there. A smaller pixel packs more resolution into the same die area but has correspondingly less room for storage capacitance, lowering full well capacity and compressing dynamic range — the identical area-buys-capacitance logic that already governed kT/C noise reduction and flicker-noise corner reduction, just applied to the ceiling of the signal range instead of its floor or its noise spectrum.

What physically sets the ceiling differs by sensor architecture, and the mechanism matters for how close to the theoretical Q equals C times V number a real pixel actually gets. A simple photodiode's full well is limited by how much charge its junction can hold before the depletion region itself can no longer sustain additional charge without breaking down or losing linearity in its own capacitance-voltage relationship. A pinned photodiode, the dominant structure in modern CMOS image sensors, is instead limited by the pinning voltage — the specific potential at which the photodiode's surface fully depletes — a parameter set deliberately during process design rather than an incidental breakdown limit, which is part of why pinned-photodiode full well capacity can be engineered more precisely than a simple photodiode's can.

This closes the dynamic-range equation the read-noise entry in this series set up but left half-finished. Read noise is the floor a sensor cannot see below regardless of how much light arrives; full well capacity is the ceiling a sensor cannot see above regardless of how carefully the readout chain was designed. Neither number alone describes a sensor's actual usable range — a sensor with an excellent, vanishingly small read noise but a tiny full well saturates almost immediately in bright conditions, while a sensor with an enormous full well but poor read noise cannot resolve faint signals no matter how much headroom exists above them.

Full well capacity is the ceiling this series' entire noise-and-precision-technique thread was always building toward pairing with a floor — read noise sets how quiet a sensor can be, full well capacity sets how much it can hold before giving up entirely, and the ratio of the two is the single number that actually describes what a sensor can do in a real scene spanning both deep shadow and bright highlight within the same frame. Every technique this series has covered, from autozeroing to correlated double sampling to dark current suppression, ultimately exists to push the floor down; full well capacity is the one number in this whole thread that instead describes how high the ceiling was built in the first place.