a voltage reference
A voltage reference is the stable yardstick that a regulator — or an ADC, or a precision meter — measures everything against. A regulator can only hold its output as steady as the reference it compares to, just as a carpenter can only build to the accuracy of the ruler in his hand. If the ruler stretches with temperature, every measurement is wrong; a good voltage reference is a ruler that refuses to change with temperature, supply voltage, load, or time.
The simplest reference is a Zener diode: bias it in reverse past its breakdown voltage and it pins to a roughly fixed voltage. It is cheap but mediocre — its voltage shifts with current and drifts with temperature, and it is noisy. The precision answer is the bandgap reference, a clever circuit that produces about 1.2 V with near-zero temperature drift. Quality is measured in parts per million per degree C (ppm/°C): a cheap Zener might drift 1000 ppm/°C while a good bandgap holds 5 to 50 ppm/°C. Example: a 2.5 V reference rated 10 ppm/°C moves only 2.5 times 10 / 1,000,000 = 25 uV per degree C — a fortieth of a millivolt.
Why it matters: the reference is the silent foundation of accuracy. In a regulator it sets the output voltage; in an ADC or DAC it sets the meaning of every code, so a reference that drifts 0.1 percent puts a 0.1 percent error on every reading the system ever takes. Whenever a design needs to be accurate rather than merely steady — a voltmeter, a battery charger's cutoff, a sensor front end — the reference, not the op-amp, is usually the part that limits how good it can be.
A 12-bit ADC running off a 4.096 V reference resolves 4.096 V / 4096 = 1 mV per step. If that reference is a Zener drifting 1000 ppm/°C, a 10 degree warm-up shifts it by 4.096 times 0.01 = 41 mV — 41 whole steps of error. Swap to a 10 ppm/°C bandgap and the same warm-up costs under half a step.
The yardstick the whole system trusts — its drift becomes everyone's error.
A regulated output is not automatically a good reference. Regulators are built for current and stability of the rail, not the few-ppm accuracy of a dedicated reference; for precision measurement use a real reference part, not the 5 V rail.