What’s the impact of temperature gradient across a layout on a bandgap reference, and how would you mitigate it?

Temperature gradient can seriously affect the output voltage accuracy of a bandgap reference. Since the circuit depends on matching between bipolar transistors and resistors, any temperature difference across them creates mismatch in Vbe and resistance values.

In one of my designs, when one side of the layout was close to a power amplifier block, it ran hotter by a few degrees. That caused the bandgap output to drift by several millivolts. To fix it, I placed the bandgap in a thermally uniform area and used symmetrical placement of devices. That helped to reduce the gradient effect.

You can’t fully remove temperature gradients, but you can minimize their impact through smart layout and shielding. One thing I like to do is place the bandgap far from high-power blocks and surround it with guard rings connected to a quiet ground to isolate it thermally and electrically. For example, when I placed the bandgap near an LDO pass transistor, the temperature gradient caused 5 mV variation. After moving it to a cooler region and adding a thermal guard ring, the variation became negligible. So, placement and isolation are key in my opinion.

In my view, the main problem is local mismatch caused by temperature gradient. When two matched devices (like the bipolar pair or resistor pair) operate at slightly different temperatures, their characteristics shift unequally, and that spoils the precision.What I usually do is use common-centroid layout for transistors and resistors, and I also add dummy devices around them to balance heat flow. For example, in a 1.2 V bandgap design, after we rearranged resistors in a common-centroid form, the offset error dropped almost by half during thermal simulation.