Mid-End Voltage Regulators are Better than Low-End Bandgap Voltage References?

[niconiconi]

It's my observation that low-end bandgap voltage references can perform even worse than a mid-end voltage regulator (which, of course, is also based on a bandgap reference). In particular, the "free" references embedded in the ADC/DAC of microcontrollers are the worst.

For example:

TL431, standard grade:
Initial accuracy: ~2%
Full temperate range: ~3%
long-term drift: no data, but TI said "a typical acceleration test of 500 hours at 150C, the device still meets specifications."

ATmega328P:
Accuracy: 10%
Temperate drift: no data
Long-term drift: no data

PIC16F1574:
Accuracy: 5%
Temperate drift: no data
Long-term drift: no data

Now let's consider some specifications for mid-end voltage regulators.

LM317 (not really mid-end, more like low-end):
Accuracy (full temperate range): ~4%
Temperature drift (0°C to 125°C): 0.7%
Long-term stability (25 °C): 0.3%/1k hr typical, 1%/1k hr max.

LM1117:
Initial accuracy: 1%
Full temperate range: ~2%
Temperature stability: 0.5% typical
Long term stability (125°C): 0.3%/1k hr typical

MCP1700:
Initial accuracy: 2%
Full temperate range: ~3%
Long-term drift: no data

TLV733:
Initial accuracy: 1%
Full temperate range: ~1.4%
Temperature stability: 0.5% typical
Long-term drift: no data

Therefore, my conclusion seems to be that if you need a slightly more accurate reference and you're already using LDO on the circuit board, it's a great choice to use the same power LDO as the reference. It's inexpensive, readily available, and already in the BOM anyway. Either a separate one, or even the same one from the power rail if the load is a micropower load like a single microcontroller. The LM1117 and TVL733 look even better than a low-grade TL431! The resistor divider will introduce more error and drift, but it's still a small error than the built-in MCU references.

Comments? Any trap for young players?

[Someone]

or even the same one from the power rail if the load is a micropower load like a single microcontroller....

Any trap for young players?

There is your trap, regulators are typically specified across some "typical" operating range and often have poor regulation as the output approaches zero current.