Bandgap Reference Basics: How Two Transistors Make a Rock-Steady 1.24 V

Author: Abloom Instruction Date: November 18, 2026

A voltage regulator, an ADC, and a comparator all need the same thing: a reference voltage that does not move when the temperature changes or the battery sags. A Zener diode is the classic answer, but it drifts with temperature. The bandgap reference fixes that by adding two opposite drifts so they cancel — and the voltage it lands on, about 1.25 V, is set by the physics of silicon itself.

Schematic of an op-amp bandgap reference with two PNP transistors of 1 times and 8 times emitter area, a 560 ohm resistor and two 6.2 kilo-ohm resistors giving a 1.24 volt reference, beside a plot of the falling base-emitter voltage and the rising PTAT voltage cancelling to a flat 1.24 volts from 0 to 70 degrees Celsius

Figure 1: An op-amp bandgap reference (a) — Q1 and Q2 are PNP transistors with 1× and 8× emitter area; R3 carries the PTAT current and R1/R2 scale it up. The two drifts cancel (b), leaving VREF at 1.24 V from 0 °C to 70 °C. (Schematic, not to scale.)

Two voltages that drift in opposite directions

The first voltage is the base–emitter voltage of a transistor, VBE. At a fixed current it falls by about −2 mV/°C as the temperature rises — it is complementary to absolute temperature (CTAT). The second voltage is the difference between the VBE of two transistors run at different current densities, ΔVBE. Because VBE depends on the logarithm of the current, that difference is proportional to absolute temperature (PTAT):

ΔVBE = (kT / q) × ln(N)

At 25 °C, kT/q = 25.7 mV. With an emitter-area ratio of N = 8, ln(8) = 2.079, so:

ΔVBE = 25.7 × 2.079 = 53.4 mV

And because ΔVBE is proportional to absolute temperature, it rises by (k/q) × ln(8) = +0.179 mV/°C. One voltage falls 2 mV per degree, the other rises 0.179 mV per degree — so if you add the second one multiplied by the right factor K, the two drifts cancel exactly.

Worked example — cancelling the drift to 1.25 V

The factor K must make the rising term match the falling one:

K = 2.0 / 0.179 = 11.2

Then the reference voltage is the sum of the two:

VREF = VBE + K × Î”VBE = 0.65 + 11.2 × 0.0534 = 0.65 + 0.60 = 1.25 V

In the op-amp circuit of Figure 1, the op-amp forces its two inputs to the same voltage, so the same PTAT current flows in R1, R2 and R3. R3 turns ΔVBE into that current, and R1 scales it back into a voltage:

I = ΔVBE / R3 = 53.4 mV / 560 Ω = 95.4 µA

With standard values R1 = R2 = 6.2 kΩ, the factor is K = R1 / R3 = 6200 / 560 = 11.07, and the reference comes out at:

VREF = 0.65 + 11.07 × 0.0534 = 0.65 + 0.591 = 1.242 V ≈ 1.24 V

The net temperature coefficient is −2.0 + 11.07 × 0.179 = −0.016 mV/°C, so over the whole 0 °C to 70 °C range the reference moves by only about −0.7 mV — less than a tenth of a percent. The small leftover drift comes from the second-order curvature of VBE, which this first-order cancellation cannot remove.

Why the answer is always about 1.25 V

The cancellation only works out if K × Î”VBE is about 0.6 V, and VBE is about 0.65 V, so the sum is always near 1.25 V. That is no coincidence: the sum is the circuit's way of extrapolating the VBE-versus-temperature line back to absolute zero, where it meets the silicon bandgap of 1.165 eV — hence the name. Practical bandgap references land at 1.2–1.3 V, and you will find them inside parts you already use: the LM385-1.2 is a 1.235 V bandgap reference, the TL431 is a bandgap-based 2.495 V reference, and the LM317 regulator's internal reference is 1.25 V. Because the output is fixed near 1.25 V, the circuit needs a supply of at least about 1.4 V to have any headroom left.

Bandgap vs Zener — which reference should you use?

A Zener reference is one part and needs no op-amp, but it drifts. A 12 V Zener moves by roughly +0.07 %/°C — about +8 mV per °C — so a 45 °C rise on a hot afternoon moves it by about 0.36 V. The bandgap in this article moves by less than 1 mV over the whole 0–70 °C range. That is why Zeners are fine for a rough regulator or a protection clamp, while bandgaps live inside ADCs, precision regulators, and battery monitors, where a moving reference would corrupt every measurement. The price is complexity: a bandgap needs an op-amp and two matched transistors, which is why it is built into ICs rather than soldered from discrete parts.

Key points to remember

  • A bandgap reference adds a falling voltage (VBE, −2 mV/°C) to a rising one (K × Î”VBE, +0.179 mV/°C per unit K) so the drifts cancel to first order.
  • Worked example: two BJTs with 8:1 emitter area give ΔVBE = 53.4 mV at 25 °C; K = 11.2 cancels the drift and VREF = 0.65 + 0.60 = 1.25 V.
  • With standard parts, R3 = 560 Ω gives a PTAT current of 95.4 µA, R1 = R2 = 6.2 kΩ gives K = 11.07, VREF = 1.242 V ≈ 1.24 V, net tempco −0.016 mV/°C, drift about −0.7 mV over 0–70 °C.
  • The 1.25 V value comes from silicon's 1.165 eV bandgap at 0 K; practical references sit at 1.2–1.3 V (LM385-1.2 = 1.235 V, TL431 = 2.495 V, LM317 internal reference 1.25 V), and need a supply of at least about 1.4 V.
  • A 12 V Zener drifts about +8 mV/°C (≈ +0.36 V over 45 °C), while the bandgap moves less than 1 mV over 0–70 °C — use a Zener for a rough clamp and a bandgap where the reference must not move.