# How are voltage references established in low-power designs?

I've learned that most Zener diodes require something on the order to 20mA in reverse bias in order for Zener breakdown to take place. However, devices such as the LM431 "Adjustable Zener" can begin regulating at as little as 0.4mA. I understand that the LM431 is basically a non-inverting op-amp with a reference (nowhere close to a passive Zener diode). How are low-power fixed voltage references established?

Edit 2017-06-23:

To clarify, I would like to know how a reference voltage can be generated inside of an IC such as the LM431. I'm mainly interested in the general design techniques used to generate reference voltages in linear/switching regulators, comparators, op-amps, supply supervisors, etc.

I have seen multiple "textbook" regulator designs, but many of them are designed for discrete implementation, and they generally use a Zener or a forward-biased diode to generate a semi-stable reference voltage connected to the output via negative feedback.

• The heart of many of today's references are derivative works of Widlar's bandgap reference – sstobbe Jun 24 '17 at 1:24
• @sstobe is that related to the Widlar current mirror? – Caleb Reister Jun 24 '17 at 1:45
• That is another one of widlars Incredible contributions to analog design, his mirror allows you to generate low level currents with small resistor values (much smaller die area) and as a bonus have impovided output resistance. I can't find the exact paper but this Bob peace article summarizes widilars work fairly well, tayloredge.com/reference/Ganssle-Pease/bobpease-bandgap.pdf – sstobbe Jun 24 '17 at 2:04
• Buy "Current Sources & Voltage References," by Linden T Harrison, and read it. You'll get everything you want there. – jonk Jun 24 '17 at 4:42
• There are several possible methods- zener (most stable but requires lots of volts and current), bandgap (cheap and easy, trimmable can be adapted for CMOS and can be improved further with higher order corrections) and floating gate (gives me the heeby-jeebies, unsuitable for some applications but great specs- your exact voltage is stored on a tiny capacitor that must remain constant enough for years). There are IC references that are even better such as 3020 Josephson junctions in series pumped with microwaves but they're a bit.. inconvenient.. as they only operate at very cold temperatures. – Spehro Pefhany Jun 24 '17 at 12:41

The typical 'bandgap' or 'Widlar' voltage reference consists of

a) the Vbe of a transistor (which is temperature sensitive)
b) plus a temperature sensitive voltage, designed to track the error in (a)

Together these add up to the ubiquitous 1.22ish volts that you see in adjustable voltage regulators everywhere.

The thermometer for (b) above is made by running two transistors at different emitter current densities. At a 10:1 ratio, the difference in Vbe will be about 60mV. This is multiplied by 10 to 600mV, and added to the Vbe.

See this for a well written article on the topic.

• +1 If you want more detail, this article by Paul Brokaw (of the eponymous reference circuit) goes into it a bit more fully without going nuts on math. For example, he explains why ~1.25 rather than the actual bandgap of Si. – Spehro Pefhany Jun 24 '17 at 12:33

Zeners are very crude references. It's not clear if you are actually asking about voltage references accurate to just a few ppm? ...or just voltage references accurate to +/-10%?

If you want to understand real voltage references then you should read up on devices such as the Intersil X60003 (my personal favorite) and the Analog Devices ADR4540. Both are low power devices.

• I am wondering how such references are actually created. i.e. where do the internal references such as 2.5V for the LM431 or the 1.25V in the LM317 come from? – Caleb Reister Jun 23 '17 at 23:28
• @CalebReister The Xicor (now Intersil) X60008 (I think Jack missed a zero in there) provides a crazy-low $I_Q\le 0.5\:\mu\textrm{A}$, crazy good initial accuracy of $\pm0.01\%$ for A- and B- grade devs, insanely good tempco of $1\frac{ppm}{^\circ C}$ (A- grade) and all that over an industrial temp range. It's insanely good. Julius Blank and his ilk really did some neat stuff at Xicor. The FGA device uses a proprietary floating gate MOS device, analog switches, charge pumps, current sources, a differential charge amp, and some special blocking caps. I could write more, but no space. – jonk Jun 23 '17 at 23:34
• @jonk You have more space if you write an answer. – pipe Jun 24 '17 at 13:37
• @pipe I could. But there are already some really good answers and they are closer to the OP's wish for the LM431 and... besides... the best answer is already chosen. A nice discussion about trapping a tiny bit of charge and then buffering it out as a voltage reference will just have to wait. – jonk Jun 24 '17 at 15:17

Typically the very low power references simply use the bipolars available (sometimes those bipolars are parasitic to the primary FET process), and operate at very low current densities with the awareness of the huge KT noise attendant.

The classic bandgaps depend on matching of the bipolars and of the servo amplifiers (crude opamps) that enforce current tracking across the 2 halves of the bandgap. To achieve matching of the servo amps at very low currents, huge (very wide spacing between source and drain) FETs are used.