Flicker Noise Corner

Flicker noise corner is the frequency at which a device's own 1/f noise power exactly equals its thermal noise floor — below that frequency, 1/f noise dominates the spectrum; above it, thermal noise takes over and 1/f noise fades into irrelevance. Unlike thermal noise, which is a thermodynamic constant set by temperature and resistance alone, the corner frequency is a device-dependent number that a designer can actually move by choosing different transistor geometry, bias current, or even carrier type — which is exactly what makes it a real design lever rather than a fixed limit to simply tolerate.

The physical origin of 1/f noise is fundamentally different from thermal noise's origin, which is why the two combine into a spectrum with a genuine corner rather than a smooth single curve. Thermal noise comes from the random thermal motion of carriers in a resistive path, present at every frequency with equal power density regardless of device structure. Flicker noise instead comes from carriers being randomly trapped and released at defect states, predominantly at the oxide-silicon interface in a MOSFET, each trap with its own characteristic time constant; summed across a large population of traps with a distribution of time constants, the result is noise power that falls off as roughly one over frequency rather than staying flat. Because this is a trapping phenomenon tied to physical defects at an interface, it depends on exactly the things interface-defect density depends on — process technology, oxide quality, and critically, how much of that interface a given device actually samples.

Device area is the single strongest lever a designer has over the corner frequency, because a larger device averages over more independent traps. Flicker noise power scales roughly inversely with the product of channel width and length — a device with twice the gate area sees roughly half the 1/f noise power at any given frequency, which pushes its corner frequency down since the same thermal floor is reached at a lower frequency on a quieter 1/f curve. This is exactly why precision instrumentation amplifier input pairs are deliberately built far larger than minimum device size even when raw speed is not a concern — the area cost buys a lower noise corner directly, the same way it bought kT/C noise improvement in the sampled-data case, just acting on a different noise mechanism entirely.

PMOS devices typically show a lower corner frequency than NMOS devices built on the same process, which is a direct consequence of where each carrier type actually conducts. A PMOS transistor's channel in many processes forms slightly below the oxide interface rather than directly against it, a buried-channel effect that physically separates conducting holes from the interface trap states that cause flicker noise in the first place; NMOS inversion layers sit closer to that same interface and sample more of its defects. This is why a precision analog front end's input differential pair is very often built from PMOS devices specifically, trading away NMOS's typically higher mobility and speed for a noise corner that can sit an order of magnitude lower on the same process.

Where the signal band of interest falls relative to the corner frequency is what actually decides whether any of this matters for a given design. An audio amplifier, a DC-coupled instrumentation front end, or a slow precision sensor interface operates entirely below typical corner frequencies that can run into the tens or hundreds of kilohertz, meaning 1/f noise is not a secondary concern but the dominant noise source across the entire band of interest — exactly the regime where autozeroing earns its bandwidth cost, since the technique specifically targets the low-frequency noise concentrated below the corner. A radio-frequency front end or a multi-gigabit serial link operates at frequencies well above any device's flicker corner, where the spectrum is already thermal-noise-dominated and autozeroing would spend real bandwidth chasing a noise component that was never the limiting one to begin with.

This is the parameter that sits directly between the two noise mechanisms this series has already covered from opposite ends. Autozeroing cancels fixed offset and the low-frequency noise concentrated below the corner, but cannot touch thermal noise at any frequency; kT/C noise is purely thermal, a thermodynamic floor present regardless of corner frequency and untouched by anything that works on the 1/f side. The corner frequency is the number that tells a designer which fight is actually worth having in a given application — whether the dominant enemy in the signal band is a trap-driven, device-geometry-dependent noise source that area and carrier-type choice can push down, or a thermodynamic floor that only a larger sampling capacitor can lower.

Flicker noise corner is the boundary line a designer reads before deciding which noise-fighting technique is actually worth its cost — a precision, low-frequency front end lives below it and spends area and bandwidth accordingly, while a wideband, high-frequency front end lives above it and would gain nothing from the same techniques. The number itself is not fixed by physics the way a thermal noise floor is; it moves with device geometry and carrier type, which is exactly why it belongs on the short list of parameters an analog designer actively engineers rather than simply measures and accepts.