Fullscreen HTML5Med Res (??MB)HTML4

By Andrew Levido Precision Electronics Part 8: Voltage References Last month, we looked at sampling and aliasing in DACs (digital-to-analog converters) and ADCs (analog-to-digital converters). Now we will describe how voltage references work, as they are critical to the precision of both ADCs and DACs. W e have covered a fair bit of ground so far in this Precision Electronics series, but we have not looked at the important topic of voltage references in detail. We mentioned them in our discussion of analog-to-digital and digital-to-analog conversion, but now is the time for a deeper dive. In the very first article in this series, we discussed the difference between precision and accuracy. We learned that precision describes the repeatability or reliability of a measured quantity and is all about understanding and quantifying sources of error. On the other hand, accuracy is how closely a measured value matches the ‘true’ value (or an accepted proxy) of the quantity. Most of what we have covered so far has been concerned with precision rather than accuracy. Voltage references are one of the places where these two come together. Put simply, voltage references provide a fixed voltage source at some defined value. This sounds simple, but there is a lot to unpack in that statement. Initial accuracy – or not The “defined value” part is where accuracy comes into play. A voltage reference is expected to provide a known output with some specific level of accuracy; that is, some measure of how close its output voltage is to the ‘true’ value. This requirement implies the source has been calibrated somehow against a known standard. In many cases, this calibration is done for you at the factory, but sometimes you are expected to do it yourself. An example of the former is the MAX6225ACSA reference we used in the DAC circuit in a previous instalment of this series. This is a 2.5V reference with an initial accuracy of ±200ppm (±0.02%). This means the voltage will be between 2.4995V and 70 Silicon Chip 2.5005V at 25°C – a figure achieved by calibration at manufacture. Calibration would have been performed by trimming the chip’s output voltage against a voltage standard with even higher accuracy. This standard will have been itself calibrated against an even better standard, and so on, in a chain ultimately traceable to the international standard definition of the volt. Not all voltage references are trimmed to a standard voltage at manufacture. Some very expensive ultra-precision voltage references have woeful initial accuracy. The legendary LTZ1000, which will cost you the best part of $150, has an initial output voltage anywhere between 6.9V and 7.5V, but has extraordinary precision. The intention is that you will calibrate the device in which you use this reference against some external standard, then take advantage of its incredible stability to ensure it stays that way. Voltage reference errors Whether you use a pre-calibrated or post-calibrated voltage reference, knowing its stability is critical. As you would expect, changes in temperature influence voltage references; achieving a stable voltage over temperature has been one of the driving forces behind the development of voltage reference technologies. You will therefore see a figure for temperature drift in the voltage reference data sheet in the familiar absolute (V/°C) or relative (ppm/°C) units. In many cases, you will also see a ‘thermal hysteresis’ figure. This is the maximum voltage change seen after cycling the device over some fixed temperature range – typically (but not always) from 25°C to 50°C and back. This is important because of the focus on accuracy; if the voltage changes slightly with temperature, we at least want it to return to a consistent value Australia's electronics magazine each time it reaches a given temperature. To add to the growing list of errors, voltage references also drift with time, even if the temperature is held constant. Most precision voltage reference data sheets therefore include a figure for ‘long-term stability’ or similar. This can be expressed as an absolute or relative change in voltage per root thousand hours (ppm/√kh) or per thousand hours (ppm/kh). The odd unit of root-kilohours is used because long-term stability typically has a logarithmic decay characteristic, with more drift in the first 1000 hours than in the second and so on. Because of this early drift, some manufacturers of precision equipment using will ‘burn in’ the voltage references for a period before using them. This improves the long-term accuracy of the references (since the higher initial drift is done with), and allows the manufacturer to weed out any that show too much variation. To put some context around these errors, consider the MAX6225ACSA. The initial accuracy was ±200ppm with a tempco (temperature coefficient) of ±2ppm/°C and a thermal hysteresis of 20ppm. The worst-case longterm drift is ±20ppm/kh. These are all pretty good figures, but the LTZ1000 is in a different class. This latter chip includes an on-chip heater to keep the reference at a stable temperature, resulting in temperature drift in the ±0.05ppm/°C range. No thermal hysteresis figure is given, since the reference is always held at a constant temperature. Long-term stability is quoted as 0.28ppm/√kh. I should note that building a device using the LTZ1000 or similar ultra-­ precision references is no trivial task. To get the most out of the chip, you must employ some ridiculously expensive high-precision, low-tempco resistor dividers and worry about all siliconchip.com.au sorts of crazy details. Take this quote from the data sheet as an example: The Kovar input leads of the TO-5 package form thermocouples when connected to copper PC boards. These thermocouples generate outputs of 35µV/°C. It is mandatory to keep the ... leads at the same temperature, otherwise 1ppm to 5ppm shifts in the output voltage can easily be expected from these thermocouples. Air currents blowing across the leads can also cause small temperature variations... There is a whole online community dedicated to getting the best possible performance out of these and similar devices. Just search for “voltnuts”. Series and shunt references Voltage references fall into two major categories: three-terminal (series) or two-terminal (shunt) devices, both of which are shown in Fig.1. Series references ‘regulate’ an input voltage to produce the reference output. As you might imagine, the output is influenced by changes in the input voltage and by the load current, so for series regulators you will see figures in the data sheet for line and load regulation (ie, how stable the output is despite variations in input voltage and output current, respectively). In the case of the MAX6225ACSA, the line regulation is ±7ppm/V for a Vin above 10V and the load regulation is ±6ppm/mA. You should therefore ensure the input to a series voltage reference is well-regulated and keep the load current low and stable. Achieving the latter can be a bit of a juggling act: you can add an op amp buffer to minimise the load current, but this will itself introduce errors that might outweigh those produced by the reference’s load regulation. You have to crunch the numbers to work out what is optimal. On the other hand, shunt references maintain a constant voltage drop while current is flowing through them. The voltage drop is influenced by the device current, and you can see from Fig.1 that this current is determined by both the source and the load currents. Shunt references therefore usually include a single figure for output voltage change with current that encompasses both line and load regulation. You should therefore aim to keep the device current constant if you want to achieve the best results with shunt-type references, most likely by siliconchip.com.au employing an active current source and by keeping the load current fixed. Changes in shunt current with temperature will add to the shunt element’s inherent temperature drift, so your current source needs to be relatively stable with temperature. Zener references The zener diode, with its well-­ defined reverse breakdown voltage, is the simplest form of shunt reference. This breakdown characteristic occurs because of two different mechanisms: the zener effect at low voltages, and the avalanche effect at higher voltages. The zener effect has a negative tempco, while the avalanche effect has a positive tempco. As both these effects are present in zener diodes with breakdown voltages around 5-7V, it is possible to have these temperature effects more-or-less cancel each other out by careful selection of the diode and its operating conditions. For example, a BZX55C5V1 5.1V, 400mW zener diode has a tempco of between +0.02% and -0.02% (±1mV/°C). By contrast, a BZX55C12 (12V) zener has a tempco of +0.11% (+13mV/°C). The initial accuracy of the 5.1V zener won’t be great, but fed with a constant current, its voltage stability will be surprisingly good. In fact, the LTZ1000 and MAX6225ACSA both use internal zener diodes as the basis of their reference. You can temperature-compensate a zener diode by putting it in series with a forward-biased standard diode, as long as you choose a zener with a tempco of about +2mV/°C. That’s because the tempco of a regular diode’s forward drop is about -2.1mV/°C. Otherwise, you can buy temperature-­ compensated zener references like the LM329, or even the LM399, which includes an on-chip heater. We mentioned above that you need to maintain a constant current through a zener reference. For example, the BZX55C5V1 has a dynamic resistance of up to 35W, so a change in bias current of just 1mA will shift the output voltage by 35mV – as much as a temperature rise of 35°C. Fig.2 shows a clever circuit that uses the zener itself to provide a stable bias current and allows the output voltage to be adjusted to boot. The zener voltage is amplified by the non-inverting amplifier to produce an output voltage, Australia's electronics magazine Fig.1: voltage references are available as either three-terminal series pass devices or twoterminal shunt devices. In either case, keeping the input voltage or current (and the output current) constant is critical to getting the best accuracy. Fig.2: in this circuit, the zener bias current is derived from its own stable output. R1 and R2 allow the output voltage to be amplified if necessary. Vout = Vz (1 + R1 ÷ R2). The output voltage is then used to establish the zener bias current via R3, Iz = (Vout – Vz) ÷ R3. You must use a single supply with this circuit – with a split supply, it could settle into a second stable state with the zener forward-biased. If you are worried about start-up, you can add a 1MW or greater resistor from the zener’s cathode to the positive supply. You will often see the term ‘buried zener’ to describe precision zener references. This just means that the diode junction (where the reverse breakdown occurs) is formed below the surface of the semiconductor and covered with a layer of diffusion material, resulting in a more stable device with lower noise. Band-gap references Band-gap voltage references are June 2025  71 much more common than zener references these days, especially inside integrated circuits. Pretty much every voltage regulator, linear or switching, uses one. Developing a semiconductor voltage reference with a low tempco was no trivial task, but it came down to adding a voltage with a positive tempco to another voltage with a negative tempco just like the compensated zener. The idea was first used commercially by Bob Widlar in 1971. Bob Widlar was an erratic genius who pioneered the first commercially successful op amps, comparators and three terminal-voltage regulators, among many others. Part of his success was to recognise and work with the strengths of the IC production process. He understood that it was very hard to manufacture components of precise absolute value in silicon, but was relatively easy to make components with precisely matched values. The Ebers-Moll large-signal BJT model (a really useful model I strongly recommend you study) tells us that the base-emitter voltage (Vbe) of a transistor is related to its collector current (Ic) by the relationship Vbe = Vt loge(Ic ÷ Is). Vt is the thermal voltage (proportional to absolute temperature) and Is is the reverse saturation current (also highly temperature-dependant). With a fixed collector current, the base-emitter voltage has a negative tempco of -2.01mV/°C, since the tempco of the Is term dominates. However, the difference between the base-emitter voltages of two transistors Fig.3: the simplest band gap reference consists of just three transistors. The voltage across R3 (Vbe1 – Vbe2), and therefore that across R2, has a positive tempco, offsetting the negative tempco of Q3’s base-emitter voltage. 72 Silicon Chip with different collector current densities has a positive tempco. This can be achieved by using two identical transistors with different collector currents. If you are mathematically inclined, it is pretty easy to see why by simplifying the expression Vbe1 – Vbe2 = Vt(loge[Ic1 ÷ Is] – loge[Ic2 ÷ Is]). This becomes Vbe1 – Vbe2 = Vt × loge(Ic1 ÷ Ic2). Note that Is has disappeared, leaving Vt with its positive tempco the only temperature-­ dependent term. Fig.3 shows how this phenomenon can be used in practice. Identical transistors Q1 and Q2 have collector currents in a 10:1 ratio because of the values of R1 and R2. Both transistors have the same base voltage, so the voltage at the emitter of Q2 must be Vbe1 – Vbe2. The 10:1 Ic ratio means this voltage will be 2.3Vt (around 60mV). The voltage across R2 will therefore be 23Vt or about 600mV. The output voltage, Vout, will be this voltage plus the Vbe drop in transistor Q3 for an output of around 1.2V. Importantly, the tempco of Vout will be that of Vbe3, around -2.01mV/°C, plus 23 times that of V t (23 × +86.2µV/°C = +1.98mV/°C), resulting in an overall tempco of -27.4µV/°C, which is around 10ppm per °C. Not bad for three transistors. This type of reference is called a bandgap reference because its output voltage for zero tempco corresponds with the theoretical bandgap voltage of the semiconductor material (~1.14V for silicon). Paul Brokaw developed an improved circuit in 1974, overcoming some of the limitations of the Widlar circuit, which could only produce a 1.2V output and required a fairly constant supply current. Brokaw’s circuit is shown in Fig.4. Brokaw’s circuit uses transistors with identical collector currents (due to identical R3s), but with differing physical on-chip areas to achieve different current densities. The voltage across R1 is the difference in base-emitter voltages, Vt × loge(N) where N is the ratio of transistor areas. The voltage across R2 is therefore Vt × loge(N) × 2(R2 ÷ R1), which has a positive tempco due to Vt. The reference output voltage will be this voltage plus the base-emitter voltage of Q1 with its negative tempco. By choosing the right values for R1, R2 and N, you can cancel the temperature Australia's electronics magazine dependencies as we did before. Fig.5 shows the same circuit configured to provide output voltages higher than the nominal 1.2V bandgap voltage. You can buy off-the-shelf bandgap references for around 60¢ each. The Microchip LM404x series, for example, are packaged as shunt references and available with output voltages from 1.225V to 5.000V. The 2.5V version in C grade has an absolute accuracy of ±0.5%. The versatile and very popular TL431 is even cheaper. This is also a 2.5V reference with a ±0.5% initial accuracy (C grade) with a typical tempco of ±10ppm/°C. You frequently see these devices in the voltage regulation circuits of low-cost flyback switch-mode power supplies. Many Silicon Chip projects have used them too. For example, the DC Supply Protectors project (June 2024; siliconchip.au/Article/16292) used one to set the over-voltage threshold. The 500W Power Amplifier (AprilJune 2022; siliconchip.au/Series/380) also used two, as part of the load-line protection circuitry. Exotic references Another more exotic voltage reference technology is the ‘JFET pinchoff’ reference. These work on a similar principle to the bandgap reference in that the difference in pinch-off voltage of two JFETs has a negative tempco that offsets the positive tempco of a current source. The ADR420 reference uses this technique to achieve an initial accuracy of ±400ppm (B grade) with a tempco of ±3ppm/°C. Its long-term stability is 50ppm/kh. The big advantage of this type of reference over bandgap references is their very low noise. Another interesting reference technology is the floating gate reference. These rely on a Mosfet (actually an array of Mosfets) with a well-­insulated ‘floating’ gate. At manufacture, charge is applied to the gate which, like a capacitor, charges to a particular voltage. The Mosfet then acts as a high-­ impedance voltage follower to read out this voltage. The voltage will remain stable as long as the gate charge does not change. The only commercial examples I am aware of are in the ISL2090 series. The 2.50V B-grade version has an initial accuracy of ±0.02%, a tempco siliconchip.com.au of ±7ppm/°C and long-term stability of 20ppm/kh. By my calculations (assuming a gate capacitance of 100pF charged to 1V), the gate leakage must be something less than 12 electrons per hour! Amazing. Voltage reference noise Depending on your application, you may have to take voltage reference noise into account. Unlike op amps, there is little consistency in how manufacturers specify the noise in their voltage references, especially at very low frequencies (say, below 10Hz). This is the area we usually operate in, and it is the territory where 1∕f noise tends to dominate, meaning we can’t just extrapolate from a wideband noise voltage to a noise density figure. Because we are dealing with DC signals, we can almost always add some filtering to reduce the voltage noise. Many references come with a ‘noise reduction’ pin that you bypass to ground with a small capacitor to improve the noise performance. For example, the MAX6225ACSA has such a pin which, if bypassed with a 1µF capacitor, will reduce the noise density above 100Hz from around 40nV/√Hz to under 15nV/√Hz. If we add an external filter, we have to make sure it does not adversely influence the reference voltage. Fig.6 shows one example of how we could do this. R1 and C1 form a low-pass filter with a -3dB frequency of 0.016Hz. The bottom end of C1 is bootstrapped by R2/C2 so that the voltage across C1 is zero in the steady state. If we did not do this, the leakage current through the capacitors would cause a voltage error as it is dropped across R1. You should also use a lownoise precision op amp buffer so that you don’t add new errors. Miscellany Precision voltage references can be expensive, so it is worth treating them with respect. Below are a few considerations you may need to be aware of, depending on your application. Solder shift: the worst thermal shock a voltage reference is likely to experience is during the assembly process, particularly if it is reflowed onto the printed circuit board (PCB). Like thermal hysteresis, this can cause a permanent change in the output voltage, known as ‘solder shift’. You generally won’t find information siliconchip.com.au on this in the data sheets, but there are a few app notes out there that discuss it. It is only going to be of concern for very high precision applications but is worth knowing about. You might want to hand-solder the reference if your application falls into this category. Start-up time: many references use their own output to provide stabilised bias conditions (like the circuit in Fig.2), so they include start-up circuitry to ensure everything comes up in an orderly fashion. This means that the output voltage might not reach its final stable value for some time. It is not unusual for this time to be tens or even hundreds of milliseconds. Your application should be aware of this and not use the reference until it is stable. Board flex: flexing a PCB that contains a precision reference can produce a measurable change in the output. One manufacturer suggests this can be as much as a 60ppm peak-topeak change for a reference mounted on a standard 1.6mm FR4 board that is 100mm wide board and flexed up and down by 1.8mm. You can minimise such mechanical stress by fixing boards down firmly and/or by using slots in the board to mechanically isolate the reference. Leakage: leakage currents across the surface of a printed circuit board can cause errors in precision references. References with noise-filtering pins (eg, the MAX6225ACSA) can be especially vulnerable, since these usually expose a high-impedance summing node to the outside world. A few tens of nano amps flowing into or out of one of these nodes can shift the output voltage by hundreds of ppm. Flux contamination or skin oils are more than enough to allow this level of current leakage, so it pays to clean your precision boards thoroughly and to keep your fingers off them once you have. Fig.4: the Brokaw band gap reference uses two transistors with the same collector currents, but of differing areas to produce a voltage across R2 with a positive tempco to offset the negative tempco of Q1’s base-emitter voltage. Fig.5: the Brokaw band gap reference can easily be adapted to produce higher output voltages. Conclusion Precision voltage references are unique in that they are one component that combines both precision and accuracy, allowing the device they are used in to deal in absolute quantities. In the next and final article of this series, we will zoom out and look at the big picture – how one might go about the high-level design of a precision electronics device from a wholes­ystem perspective. SC Australia's electronics magazine Fig.6: this RC filter reduces the noise voltage produced by a voltage reference by limiting the bandwidth to 0.016Hz. R1 and C1 are the filter, while R2 and C2 bootstrap the bottom of C1 to eliminate its leakage current, which would be otherwise be dropped across R1, causing the reference voltage to drop. June 2025  73