BJT Frequency Response: fT and fmax

Every BJT amplifier or oscillator design eventually hits a frequency ceiling set not by the circuit topology but by the transistor itself. Junction capacitances and base transit time bleed away current gain as frequency rises, and at some point the device simply stops amplifying. fT and fmax are the two numbers that quantify that ceiling, and they show up on every datasheet and in every RF or high-speed design decision — from choosing a transistor for a 2.4 GHz LNA to deciding whether a switching regulator's driver stage can keep up with a 1 MHz switching frequency.

Why Gain Falls Off with Frequency

At low frequency, the hybrid-π small-signal model of a BJT reduces to a resistive current source: ic = gm·vbe, and the short-circuit current gain hfe = ic/ib is just β₀ (or hFE), essentially flat. As frequency increases, two capacitances in parallel with the base-emitter junction start to matter:

• Cπ — the diffusion capacitance of the forward-biased base-emitter junction (dominant) plus its depletion capacitance.
• Cμ — the depletion capacitance of the reverse-biased base-collector junction, which also provides an internal feedback path (Miller effect).

As frequency rises, more of the base drive current is shunted through Cπ and Cμ instead of controlling the collector current source, so hfe rolls off. This is a single-pole rolloff to first order, at −20 dB/decade, exactly like the pole in a Bode plot dominated by an RC time constant — here the "R" is the small-signal base-emitter resistance rπ and the "C" is Cπ + Cμ(1+gm·RL-ish Miller term, though for fT the collector is AC-shorted so no Miller multiplication applies).

Defining fT — the Transition Frequency

fT is defined as the frequency at which the short-circuit current gain |hfe| extrapolates to unity (0 dB), with the collector AC-shorted to the emitter (so there's no voltage gain and no Miller multiplication of Cμ).

The standard hybrid-π derivation gives:

fT ≈ gm / [2π·(Cπ + Cμ)]

Equivalently, in terms of the total base-to-collector delay:

1/(2π·fT) = τb + τe + τc + τbc

where τb is base transit time, τe = Cje/gm is the emitter charging time, τc is collector depletion transit time, and τbc = (rc+re)·Cμ is the collector RC delay. In many small-signal devices, τb dominates at moderate currents, and Cπ dominates over Cμ, so the compact formula above is the useful design approximation.

Key behavior: gm increases with collector current (gm = IC/VT), so fT rises with IC — up to a point. At high IC, high-level injection and base widening (Kirk effect) increase τb and Cπ faster than gm grows, so fT peaks at some optimum IC and then falls. Datasheets show this as an fT-vs-IC curve, not a single number — always check the bias point at which fT is specified.

Defining fmax — the Maximum Oscillation Frequency

fT describes current gain, but many RF circuits (oscillators, power amplifiers) care about power gain, because the input and output are impedance-matched rather than current-driven into a short. fmax is the frequency at which the unilateral power gain (Mason's gain, U) extrapolates to unity — the highest frequency at which the transistor can still provide power gain at all, even with ideal lossless matching and feedback neutralization.

The standard relation:

fmax ≈ √[ fT / (8π·Cμ·rb) ]

where rb is the base spreading (ohmic) resistance. This formula shows why fmax depends on transistor layout, not just junction physics: rb is reduced by using multiple base contact fingers, narrow base stripes, and heavily doped base regions — all common RF BJT layout techniques. Cμ is reduced by minimizing base-collector junction area.

Note fmax can be higher or lower than fT depending on the device: high-power BJTs often have fmax < fT because rb and Cμ are larger (bigger junction area for current handling); small RF BJTs optimized for low rb can have fmax > fT.

Worked Example

A small-signal RF BJT is biased at IC = 5 mA, VT = 26 mV. Datasheet/model parameters: Cπ = 8 pF, Cμ = 0.5 pF, rb = 30 Ω.

Step 1 — compute gm:

gm = IC/VT = 5 mA / 26 mV = 192.3 mS

Step 2 — compute fT:

fT = gm / [2π(Cπ+Cμ)] = 0.1923 / [2π × 8.5×10⁻¹²]
   = 0.1923 / (5.34×10⁻¹¹)
   ≈ 3.60×10⁹ Hz = 3.60 GHz

Step 3 — compute fmax:

fmax = √[ fT / (8π·Cμ·rb) ]
     = √[ 3.60×10⁹ / (8π × 0.5×10⁻¹² × 30) ]
     = √[ 3.60×10⁹ / (3.77×10⁻¹⁰) ]
     = √[9.55×10¹⁸]
     ≈ 3.09×10⁹ Hz = 3.09 GHz

Check: fmax < fT here, consistent with a device where rb·Cμ delay is non-negligible compared to the intrinsic gm-driven speed — physically reasonable for a general-purpose RF BJT rather than a specialized low-rb device. Units check out: gm in S, C in F, product gives 1/s → Hz. Order of magnitude (GHz) matches typical small-signal RF transistor datasheets.

This means: for a straight current-gain application (e.g., a wideband current buffer), useful bandwidth extends toward ~3.6 GHz before gain drops to unity. For a matched power amplifier stage, useful power gain runs out around 3.1 GHz. In practice, usable gain with margin is typically limited to roughly fT/10 or lower for a stable, well-behaved design — so this transistor is realistically a roughly 360–720 MHz part (fT/10 to fT/5), not a "3 GHz transistor," despite what the headline number suggests.

Practical Design Implications

• fT is bias-dependent — pick the IC where the fT curve peaks (from the datasheet) if raw speed matters; too low or too high a bias current both cost you bandwidth.
• fT falls with temperature indirectly — gm and junction capacitances both shift with temperature, so high-speed designs should check fT over the operating temperature range, not just at 25°C.
• Cμ (Miller capacitance) dominates real-circuit bandwidth more than fT alone. In a common-emitter stage with voltage gain Av, the Miller-multiplied input capacitance is Cμ(1+Av), which can dominate the actual −3 dB bandwidth far more than the raw fT number suggests — this is why cascode topologies (which hold the collector-base voltage nearly constant) are used to push real bandwidth toward the fT limit.
• fmax sets the ultimate oscillation/amplification limit — a Colpitts or Clapp oscillator, or any tuned power amplifier, cannot usefully generate gain above fmax regardless of matching network cleverness.
• Datasheet fT/fmax are small-signal, near-unity-gain extrapolations — they are not directly the maximum "usable" frequency for a real gain stage; a rule of thumb is to design for operation at fT/5 to fT/10 for adequate gain margin.
• Layout matters as much as the die — package parasitics (lead inductance, pad capacitance) can drop the effective fmax of an RF transistor well below the bare-die number; this is why RF power transistors specify performance in a specific test fixture.

Key Takeaways

• fT is the extrapolated unity current-gain frequency with the output AC-shorted: fT ≈ gm/[2π(Cπ+Cμ)]; it rises with IC via gm, then falls at high IC due to Kirk effect and high-level injection.
• fmax is the extrapolated unity power-gain frequency: fmax ≈ √[fT/(8π·Cμ·rb)]; it depends heavily on base resistance rb and Cμ, i.e., on transistor layout, not just intrinsic junction speed.
• fmax and fT can differ significantly — power devices often have fmax < fT, while low-rb RF small-signal devices can have fmax > fT.
• Miller-multiplied Cμ, not fT directly, usually sets the real bandwidth of a common-emitter voltage-gain stage; cascoding is the standard fix.
• Both parameters are extrapolated small-signal figures of merit, not usable operating limits — practical designs target a fraction (often fT/5 to fT/10) of the datasheet number for adequate, stable gain.