Transistors: BJT and MOSFET

A transistor lets a small signal control a large current, so it can amplify or switch. John Bardeen and Walter Brattain built the first one, a germanium point-contact device, at Bell Labs in December 1947; William Shockley's junction transistor followed in 1948, and Mohamed Atalla and Dawon Kahng demonstrated the MOSFET in 1959. The bipolar transistor survives in analog and RF niches, while MOSFETs are essentially every transistor in a processor: an estimated 1.3 × 10²² had been made by 2018, more than any other manufactured object.

The bipolar junction transistor

An NPN BJT is two junctions back to back: a heavily doped n⁺ emitter, a thin p-type base (well under 1 µm in modern parts) and an n-type collector. Forward-bias the base-emitter junction, Vbe ≈ 0.6-0.7 V at milliamp currents, and the emitter injects electrons into the base. The base is so thin that about 99% of them diffuse across before recombining, and the reverse-biased collector junction sweeps them up. The small base current that does flow sets the collector current: Ic = β · Ib, with β (also written hFE) between 50 and 300. A 2N3904 specifies 100-300 at 10 mA.

• Cutoff: Vbe below about 0.5 V; no collector current.
• Active: Vce above about 0.3 V and Ic = β·Ib; Ic rises tenfold per 60 mV of Vbe. Amplifiers live here.
• Saturation: both junctions forward-biased, Vce(sat) ≈ 0.1-0.2 V; a closed switch. Designers overdrive the base with a forced β of about 10 to get there reliably.

The catch is that a BJT needs base current the whole time it is on: 10 mA of collector current at β = 100 takes 0.1 mA of continuous drive, and β itself varies two to three times between parts.

The MOSFET

An n-channel MOSFET (NMOS) starts with a p-type body. Two n⁺ regions, the source and drain, sit a channel length L apart, and above the gap lies a gate electrode separated from the silicon by a thin insulator: SiO₂, or hafnium oxide with an equivalent thickness near 1 nm in modern logic, or tens of nanometres in power MOSFETs. Because the gate is insulated it draws essentially no DC current; the MOSFET is controlled by voltage.

With Vgs = 0 the source-body-drain path is two back-to-back diodes and nothing flows. Raise Vgs and the gate field first pushes holes away from the surface, then pulls electrons in. At the threshold voltage Vth the surface inverts: electrons outnumber the body's holes and form a channel a few nanometres deep from source to drain. Vth is about 0.3-0.5 V in logic transistors, 1-2 V in logic-level power MOSFETs and 2-4 V in standard power parts.

Three regions

• Cutoff (Vgs < Vth): no channel. Only subthreshold leakage flows, falling about tenfold for every 70-100 mV below Vth.
• Triode or linear (Vds < Vgs − Vth): the channel reaches the drain and acts as a resistor set by the gate, I_D = μCox(W/L)[(Vgs − Vth)Vds − Vds²/2]. A switch should sit here.
• Saturation (Vds ≥ Vgs − Vth): the channel pinches off near the drain and the current levels out at I_D = ½ μCox(W/L)(Vgs − Vth)². Amplifiers live here.

Mind the vocabulary: "saturation" means a fully-on switch for a BJT but the constant-current region for a MOSFET.

MOSFETs as switches

Three datasheet numbers matter. Rds(on) is the channel resistance when fully on, specified at stated gate voltages: the IRLML6344 is rated 29 mΩ at Vgs = 4.5 V and 37 mΩ at 2.5 V. Conduction loss is I²·Rds(on), and Rds(on) rises about 1.5× between 25 °C and 125 °C. Vgs(th) is defined at a tiny current, typically 250 µA: it marks where conduction starts, not where the part is on. Gate charge Qg is what the driver must deliver: 10 nC from an MCU pin sourcing 20 mA takes 0.5 µs, while a 2 A gate driver does it in 5 ns. Logic-level parts guarantee Rds(on) at 4.5 V or 2.5 V so a microcontroller can drive them directly.

MOSFETs won for three reasons. They are voltage-controlled, so a gate holding a logic state draws no static current. They shrink well: Robert Dennard showed in 1974 that scaling dimensions and voltages together keeps power density constant while speed rises. And n- and p-channel devices sit side by side in one process, which makes CMOS logic possible. BJTs remain in precision analog circuits such as bandgap references, in RF as SiGe heterojunction transistors, and, with a MOS gate added, as the IGBTs in many high-power inverters.

Power MOSFETs and gate driving

A discrete power MOSFET turns the structure on its side. Current flows from source contacts on top through thousands of channel cells in parallel (planar or trench), then down through a lightly doped n⁻ drift region to the drain on the back of the die. The drift region sets the voltage rating and the parallel cells give milliohm resistance. Because the p-body is tied to the source, every power MOSFET also contains a body diode from source to drain. It conducts whenever the drain is pulled below the source, which H-bridges use for freewheeling, and it has the reverse recovery of any PN diode.

The gate is a capacitor of a few nanocoulombs, so a good drive circuit has a series resistor of a few to tens of ohms to tame ringing, a 10-100 kΩ pull-down to keep the part off while the microcontroller is still in reset, and enough voltage to reach the Rds(on) rating without exceeding the absolute maximum Vgs, typically ±20 V, or ±12 V for many logic-level parts.

Things to try:

• Hold V_DS at 0.5 V and raise V_GS from 1.5 V to 5 V: the on-resistance falls from kilohms to 143 Ω.
• At V_GS = 3 V, sweep V_DS: the current rises until 1.5 V, then flattens as the channel pinches off at the drain.
• Compare the channel power at V_GS = 5 V, V_DS = 0.5 V (triode, about 1.7 mW) with V_GS = 2.5 V, V_DS = 5 V (saturation, about 6 mW): the half-on transistor passes less current yet burns more than three times the power.