Transit Frequency

Transit frequency, written f_T, is the single number that answers "how fast is this transistor, fundamentally" independent of whatever circuit it eventually sits in. It is defined as the frequency at which a transistor's short-circuit current gain falls to exactly one — above f_T the device can no longer amplify current at all, no matter how the surrounding circuit is designed, which makes it the hard speed ceiling every amplifier, mixer, and oscillator stage built from that device has to respect.

The definition is deliberately circuit-independent, which is exactly what makes it useful for comparing devices rather than designs. Measuring current gain as a function of frequency with the output shorted strips away every external load, feedback network, and bias scheme a real circuit would add — what remains is the transistor's own intrinsic behavior, set by its transconductance and its own parasitic capacitances. Current gain starts flat at low frequency, where those capacitances barely load the signal, then rolls off at twenty decibels per decade once frequency rises high enough that charging them starts stealing current that would otherwise reach the output; f_T is simply where that roll-off line crosses zero decibels, read directly off the curve rather than computed from a separate formula.

The formula itself, f_T = g_m / (2π times the sum of the gate-source and gate-drain capacitances), ties this parameter directly to the same characterization sweep gm/Id methodology already depends on. A gm/Id-based design flow extracts transconductance efficiency, intrinsic gain, and transit frequency from one SPICE sweep per device type, sweeping bias voltage and reading all three curves off the same simulated data rather than deriving any of them from a closed-form equation — f_T is not a separate measurement requiring its own characterization effort, it comes along for free with the same sweep that produces the gm/Id curve a designer already needs for sizing.

This is where the real tension in gm/Id-based sizing shows up, because the inversion region that maximizes current efficiency is the same region that minimizes speed. Weak inversion delivers the highest transconductance per unit bias current, which is exactly why a current-starved, low-power design biases there — but weak inversion also means low current density, and f_T depends on current density through g_m's own dependence on bias current, so a device biased for maximum gm/Id runs with comparatively low f_T, its parasitic capacitances dominating over whatever transconductance is actually available. Strong inversion runs the opposite way: current density is high, f_T approaches its peak for that device, but gm/Id falls, meaning more bias current is spent to generate the same transconductance.

A designer choosing a target gm/Id on the lookup curve is, in the same motion, choosing a point on this exact tradeoff — the two numbers cannot be optimized independently because they come from the same underlying physics. A radio-frequency low-noise amplifier or a high-speed serializer output stage needs enough f_T margin above its operating frequency to have any gain left at all, which pushes the design toward strong inversion and accepts the current-efficiency cost; a precision instrumentation amplifier or a slow, high-resolution data converter front end has no such frequency floor to clear, so it biases toward weak inversion and banks the current savings instead. Neither choice is universally correct — it is read directly off which end of the same curve the application actually needs.

f_T is also only a single-pole approximation, and the distinction from f_max matters once a real design pushes close to either limit. The simple g_m-over-capacitance formula assumes current gain is the only thing that matters and ignores gate resistance entirely; f_max, the maximum frequency at which power gain (not current gain) still exceeds one, accounts for that resistance and is typically higher than f_T for a well-designed device, but is also harder to extract cleanly and more sensitive to layout parasitics outside the transistor's own intrinsic model. A device with excellent f_T can still disappoint in an actual RF design if gate resistance was not managed in layout, which is why f_T alone is a necessary but not sufficient figure of merit for genuinely high-frequency work.

Transit frequency is the speed half of the same bias-point decision gm/Id methodology turns into a lookup rather than a derivation — every point on that curve carries both a current-efficiency number and a speed number, and picking one is picking the other whether or not a designer is thinking about it explicitly. It is also the number this entire analog-design series keeps returning to from different directions: Early voltage and channel-length modulation describe how much gain one stage can deliver, gm/Id describes how efficiently that gain is bought, and f_T is the ceiling on how fast any of it can actually run.