Take apart a switching power supply and you will find MOSFETs. Look inside a precision op-amp or a low-noise preamp and you will usually find bipolar transistors. That split is not fashion. It follows from how each device is controlled, where its losses land, and what happens when it gets hot. Choosing between them comes down to four questions: what drives the control terminal, how much current flows when the device is fully on, how fast it has to switch, and where the heat goes.

The control terminal decides a lot

A bipolar junction transistor is a current-controlled device. You forward-bias the base-emitter junction, roughly 0.6 to 0.75 V, and the collector current follows at hFE times the base current. The trouble is that hFE is a moving target: it swings with collector current, temperature and part-to-part variation, often by a factor of three or more between devices of the same type. Good designs assume the worst case and overdrive the base, working at a forced beta of about 10 instead of the datasheet's typical 100 or more. That overdrive current flows continuously while the transistor is on.

A MOSFET is voltage-controlled. Its gate is a capacitor separated from the channel by a thin oxide layer, so in the steady on-state it draws essentially nothing beyond leakage. But every switching cycle you must move the gate charge, and that sets the real drive requirement:

Gate drive power = Qg × Vgs × frequency

A part with 30 nC of total gate charge driven at 10 V and 100 kHz burns about 30 mW in the driver. Now compare a BJT carrying 10 A with a forced beta of 10: the base needs about 1 A at 0.75 V, so roughly 0.75 W disappears in the base circuit before the load even gets its current. At low currents the BJT looks thrifty. At high currents the MOSFET's capacitive gate starts looking like a bargain.

Conduction losses have different shapes

These two devices lose energy in fundamentally different ways when they are fully on.

A saturated BJT behaves like a fixed voltage drop. Collector-emitter saturation voltage is typically 0.1 to 0.3 V in small-signal parts and higher in high-voltage types, and it stays roughly flat as current rises. Double the current, and the loss roughly doubles. That predictability is useful.

A MOSFET behaves like a resistor. On-resistance, Rds(on), sets the drop, and because P = I²R, losses climb steeply with current. Worse, Rds(on) is specified at 25 °C and can roughly double by the time the junction reaches 125 °C.

Put numbers on it. A MOSFET with 100 mΩ of on-resistance drops 0.1 V at 1 A — better than most BJTs. At 10 A it drops 1 V and dissipates 10 W, while a BJT at 0.2 V dissipates 2 W. So MOSFETs dominate at low voltage and low-to-moderate current, where you can buy your way to very low Rds(on). At high voltages, Rds(on) climbs steeply with breakdown rating, which is why designs above a few hundred volts often reach for IGBTs or cascode arrangements instead.

Speed, ringing and storage time

Saturating a BJT fills the base region with stored charge, and that charge has to be removed before the device turns off. The result is a storage delay followed by a falling tail, and it is the main reason hard-driven bipolar switches are slower than MOSFETs. Clamps that keep the transistor out of deep saturation help, but add parts.

MOSFETs switch fast, and that speed brings its own problems. The gate circuit and the source inductance form a resonant tank, so a fast edge produces ringing and overshoot. Miller capacitance couples drain voltage swings back into the gate and can turn the device on when you want it off. A gate resistor, a tight gate loop and a driver that can sink real current are not optional extras — they are the design.

One more practical difference: a high-side N-channel MOSFET needs its gate driven above the source, which means a bootstrap capacitor or charge pump. A high-side NPN has the same problem. A P-channel MOSFET or a PNP avoids the extra supply but costs you on-resistance or gain respectively.

Heat, runaway and paralleling

When a BJT heats up, its base-emitter voltage falls for a given current — about 2 mV per degree. The hotter device in a group therefore pulls more current, gets hotter still, and can run away. Paralleling bipolars demands emitter ballast resistors or matched dies on a shared substrate.

MOSFETs go the other way. Rds(on) rises with temperature, so a hot device naturally sheds current to its cooler neighbours. That positive temperature coefficient makes paralleling straightforward and is why high-current supplies use banks of MOSFETs rather than one big bipolar.

Neither device is invincible. Bipolars have a second-breakdown region in their safe operating area, and MOSFETs can develop hot spots when operated in linear mode at high drain voltage — a real concern in linear amplifiers and electronic loads, not in hard-switched converters.

Where BJTs still win

Amplification is the bipolar transistor's home turf. Its transconductance is high — roughly 40 millisiemens for every milliamp of collector current at room temperature — so a single stage delivers more gain than a MOSFET biased at the same current. Bipolars also match well, have low offset voltage and low flicker noise, which is why they appear in op-amp input stages, bandgaps and precision current mirrors.

MOSFETs answer a different need: essentially infinite input impedance and input bias currents in the picoamp range or below. That makes them the default for high-impedance sensors, electrometer front ends and anything driven from a source that cannot supply current. They also scale beautifully — CMOS gives you millions of near-free switches on one die. Their weaknesses are lower transconductance per milliamp, higher 1/f noise, offset from mismatch and a gate oxide that ESD will happily puncture.

Matching the part to the job

  • Battery-powered DC-DC converters, load switches, motor drivers: N-channel MOSFET, logic-level if the gate signal is only 3.3 V or 5 V.
  • Mains PFC stages, induction heating, high-voltage high-current switching: IGBT, or a high-voltage MOSFET at lower currents.
  • Driving a relay, LED string or small solenoid from a microcontroller pin: a small NPN with a base resistor is still the simplest robust answer.
  • Low-noise preamps, precision current sources, bandgaps: bipolar.
  • pH probes, photodiodes, electrometer inputs: MOSFET or JFET input devices.
  • Paralleling many devices for hundreds of amps: MOSFETs, for their self-balancing behaviour.

A short pre-commit checklist

  1. Write down what drive your circuit can actually produce, and whether it is referenced correctly for a high-side or low-side switch.
  2. Compute on-state loss at your real current and at the expected junction temperature, not at 25 °C.
  3. Estimate switching loss from gate charge and frequency for a MOSFET, and from storage time for a BJT.
  4. Check the safe operating area graph for the operating point you care about, especially during startup and fault conditions.
  5. Read the curves, not just the headline numbers: Rds(on) at 4.5 V versus 10 V, hFE versus collector current, thermal resistance with your actual copper area.

In practice most designs end up using both. A microcontroller drives a logic-level MOSFET for the power path, and a small NPN handles the housekeeping signal that is easier to switch with a current than a voltage. Knowing why each one is there is what turns a working prototype into a design you can defend.

Photo: Pok Rie / Pexels