Bandgap-Reference
A bandgap reference is a circuit that generates a fixed, highly stable output voltage, roughly 1.2 volts in silicon implementations, that barely changes with temperature or with the chip's supply voltage — a property neither a resistor-based voltage divider nor a simple diode drop can provide on its own, since both shift noticeably as temperature changes. The trick behind a bandgap reference is combining two voltages that drift with temperature in exactly opposite directions, scaled so their drifts cancel, landing on a combined voltage that happens to sit near silicon's bandgap energy and stays remarkably flat across a wide temperature range. How a bandgap reference actually cancels out temperature drift to produce that stable output looks like this:
A few things are worth knowing about why combining two badly-behaved voltages produces one extremely well-behaved one:
A bandgap reference's core trick is adding a voltage that falls with temperature to one that rises with temperature, in exactly the right proportion, so the combined result holds nearly flat — a technique that turns two individually unreliable voltages into one reliable one. A forward-biased diode's voltage drop decreases as temperature rises, a behavior engineers call complementary-to-absolute-temperature, or CTAT, typically falling by about 2 millivolts for every degree Celsius increase. Separately, the thermal voltage — a quantity derived directly from fundamental physical constants and absolute temperature — increases linearly with temperature, a behavior called proportional-to-absolute-temperature, or PTAT. Scaling the PTAT term by precisely the right factor and adding it to the CTAT diode voltage makes the temperature-dependent parts of each cancel almost exactly, leaving a combined voltage that lands near silicon's bandgap energy and barely moves across the chip's entire operating temperature range — which is also where the circuit gets its name.
A bandgap reference's output voltage doesn't just need to resist temperature changes — it also has to resist changes in the chip's own supply voltage, a separate requirement called power supply rejection, and the circuit's feedback loop is specifically what delivers that second property. Real circuits built around a bandgap core use an operational amplifier in a feedback configuration to hold the internal diode currents and voltages stable even as the chip's supply rail itself fluctuates from switching noise, load transients, or battery drain, so that the reference output stays locked to its target value regardless of what's happening on the power rail feeding it. Without that feedback stabilization, any noise or sag on the supply voltage would leak directly into the reference output, undermining the entire point of having a stable reference in the first place — power supply rejection and temperature stability are two separate engineering problems that a well-designed bandgap circuit has to solve simultaneously.
Nearly every precision function on a mixed-signal chip ultimately depends on this single reference voltage, which is why a bandgap circuit, though tiny compared to the rest of the die, is treated as one of the most critical blocks a chip contains. An analog-to-digital converter needs a known reference voltage to know what input voltage corresponds to its maximum digital output code; a voltage regulator needs a known reference to know what output voltage it's actually supposed to be holding; a current mirror's reference current is frequently derived by running the bandgap voltage across a precision resistor. Because so many other precision circuits on the chip trace their accuracy back to this one voltage, an error or instability in the bandgap reference doesn't stay contained — it propagates outward and degrades the accuracy of every block downstream that depends on it, which is exactly why bandgap circuit design gets disproportionate engineering attention relative to its small physical footprint.
Building a genuinely precise bandgap reference runs into real second-order physical effects that the simple PTAT-plus-CTAT cancellation story leaves out, and dealing with those effects is most of what separates a basic reference from a precision one. The simple linear cancellation between CTAT and PTAT terms isn't perfectly flat across the full temperature range — a residual curvature remains that limits accuracy unless additional circuitry specifically compensates for it, a refinement called curvature correction. Manufacturing variation between individual chips also means no two bandgap circuits produce exactly the same output voltage straight off the production line, which is why many precision references include on-chip or factory trim adjustments to correct each individual part back toward the target voltage after fabrication. These refinements are what separate a basic, general-purpose bandgap reference from the high-precision references used in demanding applications like high-resolution data converters, where even a few millivolts of reference error translates directly into measurable accuracy loss.
Read a bandgap reference as two flawed, temperature-sensitive voltages engineered to cancel each other's flaws, not a single inherently stable voltage source: that cancellation, combined with feedback that rejects supply noise, is what produces the one dependable voltage nearly every other precision circuit on a mixed-signal chip quietly depends on, and the real engineering difficulty lies in the second-order curvature and part-to-part variation that the basic cancellation trick alone doesn't fully solve.