Voltage references hold steady
These ubiquitous parts keep getting better, and selecting them involves carefully weighing many specs and trade-offs.
Paul Rako, Technical Editor -- EDN, October 21, 2010
At A Glance
|
| View as PDF |
Voltage references are low-output-power linear
supply regulators that produce a fixed, or
constant, voltage regardless of the loading
on the device, power-supply variations, temperature
changes, and the passage of time.
As a result, voltage references are ubiquitous
in power-supply voltage regulators, data-acquisition
systems, ADCs, DACs, and myriad
other measurement-and-control systems.
Despite their ubiquity, voltage references vary widely in performance.
A regulator for a computer power supply, for example, may
hold its value to only within a few percentage points of the nominal
value, whereas laboratory voltage standards have precision
and stability measurements in parts per million.Decades ago, these references provided initial accuracies of only ±10%, whereas modern reference ICs can provide initial accuracies of 100 ppm, or 0.01%. “We try to make the parts insensitive to line, load, and temperature variations for demanding tasks in the industrial, scientific, and medical markets,” notes Reza Moghimi, application-engineering manager at Analog Devices. Companies with expertise in those markets can also easily address the military and automotive markets, in which accuracy is critical.

Voltage-regulator chips are available
in series and shunt versions (Figure
1 and Reference 1). A series regulator
has two pins for input power and ground; a third pin outputs a fixed or
adjustable voltage. Two-terminal shunt
regulators operate at a current-limited,
fixed voltage. In essence, every voltage
regulator employs a shunt architecture
because a series reference is simply a
shunt reference with a current-feeding
circuit and a buffered output.In the early days of electronics, engineers
used neon glow tubes as voltage
references (Figure 2). The neon glow
lamp comprises two conductive terminals
in a glass container filled with rare,
or noble, gases—chemical elements
with similar properties. Under standard
conditions, they are all odorless, colorless,
monatomic gases, with low chemical
reactivity. The six noble gases that
occur naturally are helium, neon, argon,
krypton, xenon, and radon. The
gases ionize when you subject them to
a voltage of 66 to 200V dc. Once the
ionized breakdown occurs, the voltage
across the lamp drops to a maintenance
voltage of 48 to 80V dc. If the voltage
across the lamps drops below this maintenance voltage, the lamp goes out, and
you must again subject it to the ionizing
breakdown voltage to get it to light
(Figure 3). A neon glow tube works
on as little as 10−12A, or 1 pA, flowing
through it. In 1966, Signalite made
tubes that could provide regulation to
within ±0.5V (Reference 2).
By the 1970s, however, zener diodes,
which are shunt references, had obsoleted
these cold-cathode glow tubes
(Figure 4). Zener diodes take their
name from researcher Clarence Zener,
who discovered the effect (Reference
3). Although some engineers refer
to zener diodes as avalanche diodes,
the two types involve different physics
(references 4, 5, and 6). Zener breakdown
results from charge carriers that
perform quantum mechanical tunneling
through a PN junction. This breakdown
occurs in heavily doped junctions.
High electric fields in the PN junction accelerate charge carriers, causing avalanche
breakdown. These speedy carriers
cause impact ionization, which
in turn causes charge carriers to multiply.
This effect occurs in lightly doped
PN junctions. Zener-diode manufacturers
take advantage of these two
effects by varying the doping in the PN junction to create diodes with different
breakdown voltages. The zener effect
predominates in diodes with voltages
as high as 5.6V, and the avalanche
effect predominates at higher voltages.
The two effects also differ in their temperature coefficients: Zener devices
have negative breakdowns, whereas
avalanche devices have positive breakdowns.
Diodes that break down at 5.6V
combine the two effects and have a
small temperature coefficient because
the positive and negative coefficients
cancel out.
As ICs became popular in the 1970s,
it became essential that they integrate
a shunt voltage reference. Companies
such as Burr Brown, Analog Devices,
and National Semiconductor then used
the approach of burying zener diodes in
their ICs (Figure 5). The IC-process
steps create the device under the surface
layer of the die. Like buried JFETs,
buried zener diodes do not touch surface
defects in the die, meaning that
the diodes operate at low noise levels.
In 1971, US electronics engineer
Bob Widlar, a pioneer of linear-analog-IC design, employed a voltage reference
that he based on the bandgap-voltage
effect that the late DF Hilbiber,
then an engineer at Fairchild Semiconductor,
discovered in 1964. The bandgap
reference has an inherent 1.2V output
voltage, approximately the bandgap
voltage of silicon at 0°K (Figure
6). Devices with other output voltages
simply increase or decrease the voltage
with internal gain circuits. Analog-IC
designer Bob Pease improved on Widlar’s
designs and helped IC designers at
National Semiconductor use the bandgap
circuit in dozens of chips (Reference
7). “In the 1980s, 40 to 60% of
the bandgap [voltage-effect references]
we brought out had old, dumb errors,”
Pease remarks. “Many such errors are
related to IC layout, and we fixed them
with a good design review.”

In 1974, Paul Brokaw, now senior
technologist at Integrated Device Technology,
designed a bandgap voltage that
used feedback to improve accuracy and reduce errors (Figure 7). “I dreamed it
up when trying to make a discrete power
supply, and I wanted to use a lower
reference voltage than a 6.8V zener diode,”
says Brokaw.In addition to the buried-zener and bandgap-type voltage references, JFET-based devices, such as the Analog Devices’ ADR440, are also available (Figure 8). The buried JFETs help these parts achieve noise specs of 1 μV p-p over 0.1 to 10 Hz. Analog Devices’ Moghimi also alludes to a new class of references that the company will introduce this year that employs a different architecture from any of the techniques this article describes.
Other references, such as those from
Intersil, employ floating-gate FETs similar
to the structures that flash memory
uses but programmed to an analog voltage
(Reference 8 and Figure 9). Intersil
buffers the voltage within the device,
so no leakage currents come from
ESD (electrostatic-discharge) diodes
that would bleed off the charge on the
floating gate. These parts use little current but have better noise performance
than low-power voltage references using
conventional architectures. Barry
Harvey, IC-design manager at Intersil,
notes that the references required
some clever tricks in both the process
and the IC design. “Once we perfected
it, we found that leakage off the floating
gate was in the range of attoamps,
[10−18A], even at high temperatures,”
he says. By using a floating gate instead
of a bandgap, Intersil can program one
die to make parts offering dozens of output
voltages.
Voltage-reference specs
Voltage references have two fundamental
specs: load regulation and line
regulation. Load regulation relates to
the change in the output as the part
draws more current. Line regulation refers
to the change in output as the power
supplying the part changes. Transient
regulation, or output impedance,
also relates to load regulation. Output
voltage must stay in range even if your
system draws sudden current pulses
from the reference IC. Some modern
ADCs have reference inputs that draw
large transients from your part. You can
sometimes fix this problem by adding a
large-output filtering capacitor, but you
must be careful that you don’t cause the
reference to become unstable.You must understand several specifications to properly select and apply voltage references. You need not worry about the internal architecture. It is more important to know the specifications of the part, not how the IC company designs it internally. Besides deciding between shunt and series regulators, you must determine whether a zener diode would work in your system. In most cases, you are better off using a specialized voltage-reference IC from an analog-chip company. If you need ultralow power, you should use a series voltage reference, such as a floating-gate device from Intersil. Linear Technology offers the bipolar LT6656, which operates from less than 1 μA of supply current.
After you consider your
power budget and select a
series or a shunt reference
and an output voltage, you
must consider the device’s
initial accuracy—that is,
the accuracy at room temperature
when you first
apply power to the part.
Some adjustable references
let you set the output
voltage or shunt voltage
with one or two resistors.
The accuracy of those resistors
combines with the stated initial accuracy of the chip to
give you a total initial accuracy of your
output voltage. More commonly, you
select a part with a fixed voltage output
of 1.2 to 12V. The initial accuracy of
the device determines how closely all
the parts you buy approach that ideal
voltage output. Using discrete zener diodes
or older legacy references, you can
expect to achieve 10% accuracy, meaning
that you must calibrate or adjust the
circuits in production. Modern parts,
such as Analog Devices’ AD588, have
initial accuracies approaching 0.01%.
This spec is critical in data-acquisition
systems that require accuracies of 16, 18, or even 20 bits. Another factor
driving the adoption of parts with high
initial accuracy is the requirement for
charging lithium-ion batteries.
The design of charger
ICs or the process of
measuring the charging
voltage of a lithium-ion
battery requires a total accuracy
better than 0.5%.
Thus, the voltage reference
should have initial accuracies
close to 0.2% to keep
total system inaccuracy to
less than the 0.5% figure
that battery-cell manufacturers
specify.Once you have specified
the initial accuracy, you can
begin to consider output-voltage
drift. Temperature
drift, which designers often
express as a temperature coefficient
in parts per million
per degree Celsius, reveals
how much the IC’s output
voltage changes as the ambient
temperature changes.
If you specify your system over a large
temperature range, such as automotive
or military circuits, you have to look at the accuracy of the device across the entire
temperature range and add that accuracy
to the initial accuracy of the part.
Once you have specified the initial accuracy and temperature drift of the part, you then have to look at the stability, or output-voltage drift, over time. Most parts change over the first six months of operation and then settle down to a smaller change over time. Again, the output drift adds to the initial inaccuracy and temperature drift. If you want your system to have a tight accuracy over its operating life, you must use parts that have a long-term drift speciation that keeps your system’s reference voltage within the desired limits. You can also average the output of multiple parts to reduce the effect of output drift over time (Reference 9). Some manufacturers take the extra steps of determining, specifying, and measuring temperature drift and long-term stability of a part, and these steps take time and come at a price. For example, Analog Devices tests the ADR425 voltage reference for a long-term stability of 50 ppm/1000 hours.
A related but less appreciated specification is the turn-on settling time of a reference IC. The output of an IC does not instantly stay within specified limits, so firmware engineers should not make readings or do calibration in the first few microseconds of a circuit’s operation. Many parts specify a 10-μsec delay after you apply power to the part.
Another important specification is noise. Because series references are simply op-amp-buffered shunt references, you can expect the output to have noise characteristics similar to those of an op amp. The noise spectrum is flatband at higher frequencies. Because you use voltage references for their dc output, however, most manufacturers specify their products with a peak-to-peak output-noise voltage over a frequency range of 0.1 to 10 Hz, for example. You might be able to reduce this noise by increasing the output capacitance, but you must be careful to not make the reference unstable. As with all op-amp circuits, driving a large capacitive load makes the amplifier oscillate. Analog Devices’ Moghimi wishes that analog designers would more carefully read modern reference data sheets. “Some customers still think it is good to put a large output capacitor on the part,” he says. “Even if it does not cause stability problems, it can cause the temperature coefficient to get much worse.”
Another trick for reducing noise is
to parallel several voltage references
and add the outputs together. Noise is a
random phenomenon, so the noise contribution
of each reference adds in an
rms (root/mean/square) fashion. Thus,
10 paralleled references can potentially
reduce the voltage noise by the square
root of 10, or over a factor of three (Figure
10). State-of-the-art references,
such as Linear Technology’s LTC6655,
have a noise spec of 0.625 μV p-p over
0.1 to 10 Hz.Still another specification, which relates to the temperature coefficient, is hysteresis—an effect in which the output drifts to another level when you heat the part and then cool it down to its original temperature. Manufacturers often specify it as a part-per-million value over a temperature traversal such as 0 to 50 to 0°C. Like all other analog circuits, voltage-references chips also have a PSRR (power-supply-rejection ratio)—how much attenuation the part applies to any noise or changes in the power supply that is feeding the part. This spec is important now that more systems use switching voltage regulators to supply the reference IC. Manufacturers often specify this characteristic as a voltage ratio in decibels at dc or over a frequency range. PSRR always drops off at higher frequencies, often falling to 20 dB or less at 1 MHz. If your voltage-reference chip is operating from a power supply comprising a switching regulator operating at these high frequencies, you must ensure that the ripple and noise on the power supply do not bleed into the reference-voltage output due to poor high-frequency PSRR. You can often fix these problems by putting a linear preregulator IC on the switching-power-supply output that feeds your reference chip. You can also put RC (resistance/capacitance) or RLC (resistance/ inductance/capacitance) filters before the power-supply pin of the reference IC. This approach prevents the occurrence of high-frequency noise in the reference voltage.
Some engineers use Spice models of voltage references; keep in mind, however, that these models have varying quality. Analog Devices, for example, puts the effects of most of the specs into the model. Other companies do not model references at all. Be sure that your model run takes into account all the specs that will affect your design. You may have to do a Monte Carlo Spice run to see the limits of accuracy, but at least you will know the limitations of the parts you are evaluating.
Trade-offs abound
The accuracy and noise of a voltage reference are important parts of the system-design trade-offs you make. For instance, LCD televisions are adopting Class D audio subsystems. Class D amplifiers are similar to switching regulators in that they are more efficient than conventional Class AB amplifiers. One trade-off with Class D amps, however, is that they have worse PSRR than do linear output stages. As a result, you must use either a higher-quality power supply or a more expensive Class D IC with feedback that corrects errors due to power-supply-voltage changes. This trade-off directly affects your choice of voltage reference. You might use a low-noise voltage reference in a power-supply circuit with low ripple and good regulation. You can then use a less expensive, open-loop Class D-amplifier circuit in your audio systems. On the other hand, it may be less costly to use a Class D-amplifier IC that has feedback and good PSRR so that you can use a less expensive power-supply circuit. This trade-off will change over the years and over the power and cost targets that you have for the TV. Using a more expensive voltage-reference circuit can save money in other subsystems or in factory calibration or testing when you manufacture the product in volume.

You can reach Technical Editor Paul Rako at 1-408-745-1994 and paul.rako@cancom.com.
|
References |
|
Talkback
-
OK, I just got off the phone with Scott Stever, the general-purpose instrument planner at Agilent. He verifies what Jim Williams said-- Agilent ages the references before the test equipment goes out the door. Steve said "Anything you do to a reference affects their stability." So they solder the parts into a carrier or sub-assembly that they can age, and then pop it into the instrument after the reference has settled down. Steve also reminded me that references in metal cans tend to be more stable initially, since they are not suffering die stress from the pressure of the plastic encapsulation around the die. And analog is always hard- Steve also mentioned that when you keep a reference in an temp oven to keep it from thermally drifting, it means that higher temperature will make the age drifting worse-- so yet another analog trade-off to deal with. He mentioned that Agilent is constantly looking at surface-mount references and evaluating them compared to metal-can parts, to see if they can be characterized or aged or some special sauce used to allow their use, since this would be much better for battery life in a hand-held instruments, compared to when you have to have a heated metal-can reference. Scott pointed out that each part number is different and part of Agilent's intellectual property is understanding how the parts behave initially, after soldering, and long term. I see why my buddies all love 34401s.
Paul Rako - 2010-1-12 08:58:21 PST -
Well the fine people at Fluke responded to my query and Jim Williams is right, they age the references for 60 days before selling the equipment. Barb Richardson was nice enough to pass along a response from their calibration group:
"Fluke makes the king of voltage references: 732B and 734A system (for the cal lab anyway). The article talks of aging components - certainly a requirement - we age/characterize our V references for 60 days prior to shipping to a customer - and then we only ship them in a powered-up state. We also have the 7000 series references (nearly as good) that use a reference annealing process to carefully restore their characteristics after being shipped cold."
Paul Rako - 2010-18-11 13:42:42 PST -
Jim Williams agrees that 2ppm drift over time is not practical for a brand new part just soldered into a board. He asked me to check with Agilent and Fluke to verify that they age their references for a few months before soldering them in. I have asked my PR pals to verify if this is true and will let you know. I guess I should have been more clear about the time drift specification-- it might only apply after a burn-in after you solder it on your board. Everybody, Pease, Williams, Dobkin agrees that the time drift slows down a few months after soldering-- I have heard 2 to 6 months. Lets see what Fluke and Agilent say. None of this discussion should be confused with temperature drift-- we are talking about time drift, how the part changes output from the time it is soldered in to ten or 20 years later.
Paul Rako - 2010-17-11 10:13:21 PST -
Bob is right. The numbers are pure lies. TI did the same thing for a while. The best references can do a few ppm per kHr but temperature effects will usually swamp that drift. Another problem with references is temperature hysteresis. just cycling between room and hot and back will give 10-50ppm change.
Bob Dobkin - 2010-9-11 12:48:23 PST -
YOU are not alone. YOU are not the only person to be fooled by the Analog Devices "statement" that the ADR293 can do 0.2 ppm per 1000 hours.
I have been told, (by people who sampled some and who BOUGHT some), that 20ppm is a lot more like it. What ADI does (did) was take the actual 200 ppm drift per 1000 hours at + 120 degrees C, and back-date it per Arrhenius equations to pretend it is 0.2 at room temp.
BO - O - O - O - GUS.
I am sending this over to Bob Dobkin because he knows about LM399's, etc., and his people know about the ADR293. It does NOBODY in the whole INDUSTRY any good to have BO - O - O - O - GUS statements like that around. (Not even ADI, who will be CURSED for making exaggerated claims.)
If you ask the ADI guy, "Show us the data that show your typical ADR293 can actually do 0.2 ppm per 1000 hours at room temp", - - - when you hear what he says, you will PUKE.
Somebody ought to make sure the ADR293 and its BOGUS specs get black-balled. I didn't take any data on this, but I watched as other people did the work. I didn't have to do the measurements myself.
Bob Pease - 2010-9-11 12:02:46 PST





















