September 1, 1997

Stay off the HOT SEAT when choosing temperature sensors

DAN STRASSBERG, SENIOR TECHNICAL EDITOR

Many people are well-versed in contact temperature sensors. But that doesn't get you off the hook when you need to use one. Then you have to be an expert. It pays to know what your options are and where to turn for information.

For as long as people have been measuring temperature, physicists and engineers have been developing ways to do it better. Dozens of techniques have evolved, including several that require no contact with the medium whose temperature you are sensing. Still, where the sensor can touch the solid, liquid, or gas whose temperature it must measure, four technologies predominate. The older ones are thermocouples (TCs), thermistors, and resistance-temperature detectors (RTDs). IC sensors constitute a newer class.

Over the years, a lot has been written about contact temperature sensors (see box "For more information..."). Despite the surfeit of reference material, many EEs are so familiar with ICs that they choose IC sensors without giving the other options much thought. For a growing number of applications, ICs clearly are the best choice. But there are plenty of situations in which ICs don't work. The most obvious is when the temperatures you need to measure are greater than 200°C or less than about ­55°C. In fact, choosing a temperature sensor usually isn't a trivial exercise, and avoiding application pitfalls requires care.

Still, standard practice in your industry generally influences your sensor choice. When several types of sensors can do the job, the type that your company has used in previous products often becomes the preferred choice for new applications. For example, IC-temperature-sensor manufacturers bemoan the automobile industry's commitment to thermistors. Although thermistors are small, inexpensive, reliable, and not too hard to apply, semiconductor-industry sources point to important cost advantages for ICs in certain applications. According to these sources, ICs can offer lower cost when the chip that houses the sensor also performs related functions.

Table 1 compares several characteristics of TCs, RTDs, thermistors, and IC temperature sensors. Although the table sometimes suggests that one type of sensor is superior in a particular way, qualifying statements almost always follow. Selecting a temperature sensor requires careful study, because there are so few characteristics in which one technology stands out as superior.

TCs are most common

Of all the types of temperature sensors, TCs probably see the most wide use. They owe their popularity to simplicity, small size, interchangeability, relative ruggedness, suitability to a broad range of temperatures, and moderate cost. Of the four predominant types of sensors, TCs are the only ones that require no power or excitation, and this absence of excitation means that TCs dissipate no power. Power dissipation can warm a sensor and cause measurement errors. Also, TCs connect to the measurement circuit with only two wires; some other types of sensors require three or four wires.

TCs are moderately priced, though they are more expensive than most thermistors and IC sensors. In small quantities, electrically insulated TCs made of common types of TC wire in standard sizes--for example, 1-ft lengths of 24-AWG wire--cost approximately $6 each. TCs of the same composition and wire gauge are interchangeable with one another, although variations in material purity can affect TC accuracy and cost.

A TC consists of a pair of wires made of dissimilar metals, which occupy different positions in the electromotive series. The TC senses the difference between the temperatures at two points. One of these points is the hot junction, or "bead," where the dissimilar wires are welded together. The other is the cold junction. In practice, the cold junction is not usually one physical point; rather, it is a pair of points where the TC wires connect to a measurement circuit. Normally, the measurement-circuit wiring is copper. For a TC to accurately measure the hot-junction temperature, both junctions between a TC wire and the copper wiring must be at the same temperature: the cold-junction temperature.

Laboratory setups sometimes surround the cold junction with an ice bath to hold the junction at 0°C. In such cases, the voltage at the measurement-circuit terminals represents the hot-junction temperature. Most modern TC-based measurement systems make no attempt to control the cold-junction temperature, however. Rather, these systems allow the cold junction to assume the temperature of the room or the surrounding equipment. Such systems use a sensor other than a thermocouple to accurately measure this temperature. The hot-junction temperature then equals the sum of the cold-junction temperature and the temperature difference that the thermocouple measures.

Try to avoid extensions

If at all possible, TCs should connect directly to the measurement system without extension wiring. If you must use extensions, you need ones made of special wire with special connectors. To avoid creating unwanted dissimilar-metal junctions at unknown temperatures, the extension-wire materials must be the same as those of the TCs. Transitions from TC wire to copper should take place at a known location within the measurement system. At this point, the two sides of the TC wire must be close both physically and thermally. That way, there is only one cold-junction temperature, and you can accurately measure it.

System designers used to prefer RTDs for cold-junction-temperature measurement because RTDs offer the best stability and accuracy. But IC sensors have captured a lot of the cold-junction-sensing business because most cold junctions are near room temperature, at which ICs work well. More important, with prices beginning at about $0.50 (1000), ICs are usually less expensive than RTDs and are accurate enough for most applications.

So, why you should measure temperature with a TC? After all, measuring the TC's cold-junction temperature requires another sensor. Why not simply use that other sensor to measure the unknown temperature? There are several reasons. Usually, TC-based systems measure temperatures at many points--sometimes hundreds or thousands. Therefore, with a carefully constructed isothermal termination panel for a large number of TCs, you can prorate the cost of the cold-junction sensor among many channels.

Moreover, the cold-junction sensor may be too large, fragile, or expensive for use in direct measurements. Also, the sensor may work only over a limited temperature range and may require three or four connections, whereas a TC requires just two.

Despite many advantages, TCs' low output voltage--typically only a few millivolts--is a drawback. Relatively inexpensive IC op amps that offer ultralow initial-offset voltage, low noise, and offset-voltage drifts less than 1 µV/°C make this problem less important than it used to be. Still, good op amps, though key elements in solving the TC-signal-conditioning problem, are not a panacea.

To achieve the necessary CMRR, you usually need an instrumentation amplifier. In practical circuits, instrumentation amps generally achieve higher CMRR than do op amps because, unlike op amps, instrumentation amps have a pair of input terminals that are unencumbered by connections to a feedback network. Although good instrumentation amps contain high-quality op amps, the op amps in today's systems are usually just functional blocks within single-chip instrumentation amps.

Despite their advantages for conditioning low-level signals from sources such as TCs, instrumentation amps are not always adequate, however. Sometimes, a-chieving the necessary CMRR and common-mode voltage tolerance (CMVT) requires a more complex device: an isolation amplifier (iso amp). Iso amps provide ohmic isolation between their inputs and outputs. Often, this isolation withstands hundreds of volts or more. Iso amps are usually not ICs. Rather, they are hybrid circuits comprising several IC chips, usually in combination with specialized miniature transformers or LEDs and photodetectors. By their nature, iso amps are more expensive than op amps or instrumentation amps.

In many TC measurements, however, spending extra money on ohmic isolation is a good idea. For example, if the structure whose temperature you are measuring is metallic, is not easily grounded, and has high capacitance to the ac line, its line-frequency potential can be appreciable. To avoid using an iso amp, your first temptation might be to electrically insulate the thermocouple bead from the structure. Unfortunately, nearly all good electrical insulators are also thermal insulators. Therefore, if you insulate the thermocouple bead, you will, at best, slow the thermocouple's response to temperature changes and, perhaps, introduce steady-state errors.

Because of iso amps' cost, system designers usually try to share an amplifier among many channels. In TC-based measurement systems, sharing an amplifier is often practical because most TCs measure temperatures that change rather slowly. Therefore, measuring the temperature at each point as infrequently as once every few seconds, or even once per minute, can be acceptable.

To share the amplifier or signal-conditioner among many channels, most multichannel systems first route the signals through analog multiplexers, which switch the amplifier inputs among many signal sources. Some multiplexers ("flying capacitor" types, for example) also provide isolation, allowing you to use a less expensive amplifier that doesn't incorporate isolation.

By removing the common-mode voltage from the input signals, a flying-capacitor multiplexer also relieves the shared amplifier of the need to provide CMRR. But, regardless of the technology, if a measurement system's CMRR is inadequate, the usual result is excessive line-frequency interference superimposed on the signal you are trying to measure. Although this interference normally begins life as a common-mode voltage, it can become a normal-mode signal at the input to the amplifier that is supposed to amplify the TC's output. In such cases, the interference can saturate the amplifier during portions of each line-frequency cycle.

If the amplifier saturates, its output is not an amplified version of the TC voltage (even if only the input-stage saturates and the saturation is not continuous). This problem suggests that filtering the amplifier output is pointless and that you need a passive lowpass filter ahead of the amplifier. In fact, the most practical place for the filter is ahead of the multiplexer. If you locate the filter there, you need have little concern over the filter's settling time. However, if you use a solid-state multiplexer, you should evaluate the offsets that can result from capacitive coupling of the multiplexer control signals into the filter.

After you have amplified the TC signal, you still need to linearize it. If you need to measure temperature with an accuracy of about 1°C, you cannot assume that the temperature is linearly proportional to the amplified TC output voltage.

Although linearization in the analog domain definitely works, you can rarely justify an analog approach in this era of low-cost computing. Instead, you should linearize the data after you have converted the amplified TC signal into the digital domain. The main methods of linearization are look-up tables and polynomials. Polynomial correction is the more popular method, though it no longer offers the cost advantages it once did. The technique became popular when the computer memory needed for storing look-up tables was expensive. Storing just a few coefficients and taking a little extra CPU time for the calculations was fast enough and, a decade or more ago, was clearly less costly than using a look-up table.

Once you have amplified, digitized, and linearized the TC voltage, you have a numeric representation of the hot-junction-to-cold-junction temperature difference. Converting the data to degrees Celsius or Fahrenheit then requires only multiplying by a scale factor. When the temperature difference is expressed in degrees, correcting for the cold-junction temperature requires only simple subtraction.

Figure 1 compares the accuracy of the most popular types of TCs with that of wirewound platinum RTDs (PRTDs), also called platinum resistance thermometers (PRTs). Wirewound PRTDs are the most accurate contact temperature sensors in common use; TCs are rarely accurate to better than about ±1°C. More expensive, special TCs made of higher purity versions of the same materials can cut the errors roughly in half. Although the reproducibility of TC measurements is better than the measurement accuracy, if you need to hold absolute errors to a fraction of a degree Celsius or Fahrenheit, TCs are a questionable choice.

The platinum standard

The name "RTD" applies to devices based on a metal's known, positive, relatively linear temperature coefficient of resistance (TCR). The most common metal in RTDs is platinum (Pt). At room temperature, Pt wire whose purity exceeds 99.999% exhibits a TCR of 0.3926%/°C. To obtain this TCR, the wire must not have been subject to strain during winding, and temperature changes must not introduce strain.

A relatively recent development is the thin-film RTD, made by evaporating or sputtering a metal film onto a substrate--usually ceramic. Like wirewound devices, thin-film RTDs are often made of Pt. However, other materials, such as nickel-iron (NiFe) alloys, are more common in thin-film RTDs than in wirewound devices. Besides lower cost, an advantage of such alloys is a higher TCR.

Because the film can be quite thin, thin-film RTDs can have higher resistance than wirewound devices. Thin-film devices with a room-temperature resistance of 1000 ohms are common, whereas a common value for wirewound types is 100 ohms. The higher resistance enables the use of lower excitation currents for equivalent output voltages. Also, because platinum is an expensive precious metal, Pt RTDs' price depends strongly on the amount of material they use. Thin-film devices' use of less material makes them inherently less expensive than wirewound units.

A further advantage of thin-film-RTD elements is their low thermal mass, which theoretically makes them capable of very fast response. In practice, however, not all devices achieve such response. Plastic packaging, a common choice in cost-sensitive applications, can result in units whose time constants are many seconds, compared with 0.1 sec or so for the element itself. In fact, one manufacturer of thin-film RTDs reports that low-cost thin-film devices often exhibit time constants longer than those of more massive wirewound units. RTD prices also vary widely and can be as high as hundreds of dollars. However, good starting numbers for small quantities are $4 for low-cost thin-film sensors and $60 for wirewound units.

An area in which wirewound RTDs still outshine thin-film units is stability. An RTD's TCR depends on more than just the TCR of the wire or film. Mechanical stress on the element also affects the TCR. It is challenging to construct units in which the temperature changes you are trying to measure do not also place strain on the sensing element. Manufacturers report that the task is more difficult with thin-film units than with wirewound ones.

Although RTDs are known for their linear resistance vs temperature, this linearity is still inadequate in many applications. As with TCs, you can use look-up tables or polynomials to make the corrections. Polynomials are, by far, the more popular approach.

RTD signal conditioning requires some thought. When you measure the RTD's resistance, you must not measure the resistance of the connecting wires. The most obvious approach is to force a known current through the RTD and measure the voltage across the RTD in a four-wire circuit. This way, you exclude the voltage drop in the leads that carry current to the RTD.

Because self-heating can interfere with accurate measurements, you should keep the current that you force through the RTD as low as possible. Pulsed excitation can sometimes help. If you can accurately control the duty cycle and keep it low, you may be able to reduce the effects of self-heating by powering the RTD only during measurements.

When you set the excitation currents to produce the same output voltage for a given temperature change, a 1000 ohms thin-film RTD dissipates one-tenth the power of a 100 ohms wirewound device. Still, thin-film devices aren't always better from a self-heating standpoint, because their thermal resistance can be higher.

To determine the self-heating effects, you need to know more than the RTD's power dissipation. You also need to know the thermal resistance between the RTD element and the medium that carries heat away from it (normally the object or substance whose temperature you are sensing). The thermal resistance depends on, among other things, the medium's specific heat or thermal conductivity and whether the medium is a solid, a liquid, or a gas. For liquids and gases, speed of motion is important. For solids, an equally important consideration is the thermal resistance between the sensor and the sensed object. This resistance depends on the surface roughness of the two objects.

If you measure the RTD's resistance using a Wheatstone bridge and a detector whose input resistance is high, you should need only three wires--not four--to connect the RTD. The wires that carry current to and from the RTD must be the same gauge, length, and material and must follow the same path. The wire that carries current from the excitation source to the RTD forms part of the bridge arm that includes the RTD. The wire that carries current back from the RTD is part of the bridge arm in series with the RTD. You measure the bridge-unbalance voltage between the point where this last wire contacts the RTD and the common point of a pair of precision bridge-completion resistors in the measurement circuit.

Temperature to resistance

Like RTDs, thermistors use resistance as a surrogate for temperature. Some people refer to thermistors as RTDs because both types of sensors depend on variations in electrical resistance. Usually, though, thermistors and RTDs are considered separate categories. Although you can obtain positive-temperature-coefficient (PTC) thermistors, such devices see much less use for temperature measurement than do negative-temperature-coefficient (NTC) devices. A major use for PTC thermistors is in circuit-protection devices, that is, self-resetting solid-state "fuses."

Thermistors' low cost, small size, simplicity, reliability, fast response, and relative ease of use make the devices top contenders for many temperature measurements. Despite a strong challenge from IC sensors, thermistors maintain an important position in applications that don't involve TCs' and RTDs' wide temperature ranges and that don't demand the utmost in accuracy and stability. Many thermistors cost much less than $1 each in moderate quantities.

Thermistors' TCRs are greater than those of RTDs, and thermistors' room-temperature resistance is also usually higher, especially when compared with that of wirewound RTDs (Figure 2). The relatively high resistance coupled with high TCR enables low self-heating and simple signal conditioning. Generally, you don't need three- or four-wire measurement circuits or high-stability op amps in thermistor circuits. One shortcoming of thermistors is interchangeability; although some thermistors are available from multiple sources, many are sole-sourced.

Of the four major types of sensors, thermistors also have the least linear output vs temperature. EEs used to expend considerable effort designing networks that combined thermistors and resistors to produce linear changes in resistance or current vs temperature over limited ranges. Today, however, most applications include µPs for purposes only tangentially related to temperature measurement. These processors usually have unused clock cycles. Therefore, the µPs can linearize thermistor outputs at no extra cost via table look-up or manipulation of equations.

New but not so new

IC temperature sensors have been around as commercial products for about two decades. By the standards of the sensor industry, they are a recent innovation. By semiconductor-industry standards, however, they are far from new. But, like virtually everything the semiconductor industry touches, IC temperature sensors are evolving rapidly.

IC manufacturers are now moving to integrate temperature sensing with other functions that relate to measuring and controlling environmental conditions. Because most temperature measurements are now digital, the first step was combining temperature sensors with ADCs.

The phenomenon that underlies the IC temperature sensors in today's complex environmental-management chips is no different from that exploited by the earliest IC temperature sensors. IC temperature sensors depend on the accurately known temperature dependence of the forward voltage of silicon junctions. This voltage depends on both the absolute temperature and the current through the junction. Near room temperature, the voltage changes by approximately ­2.2 mV/°C.

Among the most ambitious environmental-management ICs are ones that aim to lower the cost of PC ownership in large companies. One such chip is the $5.50 (1000) LM78, which National Semiconductor designed for bonding to a PC's CPU during computer assembly. The chip monitors the CPU temperature, the airflow through the system unit, the rotational speed of the fans, and the power-supply voltages. In the future, similar chips may also control the fan speed, though only in a coarse way--high or low speed. Such chips may someday also reduce the CPU clock speed or disable functions that consume lots of power if the CPU temperature gets too high.

Did you open the case?

A case-intrusion-sensor input allows the LM78 to answer the first accusatory questions that information-technology personnel always ask PC users unfortunate enough to encounter hardware problems: "You didn't open the case, did you? Well, did you?" Of necessity, the airflow, fan-speed, and case-intrusion sensors are separate devices, as are the sensors that measure temperatures of ICs other than the CPU. The LM78 provides both ISA-bus and serial interfaces.

IC manufacturers have sold millions of simpler temperature-sensor ICs to PC suppliers for monitoring the temperature of the CPU chips in Pentium-class PCs. The cavity-down, PGA packages of most Pentium-class µPs provide a good location for sensor ICs. The sensors go on the motherboard beneath the µP-package cavity in the center of the square "window frame" of device pins. Sensor-IC vendors report that, unfortunately, the Pentium II's proprietary package provides no similarly convenient spot for a temperature sensor.

Accurately measuring temperature with a plastic-cased, multileaded, environmental-management IC can present some problems. Until you discover that your system doesn't work as well as you thought it would, you might not even imagine that such problems could exist.

Many of the plastic cases around low-cost ICs and some thin-film RTDs are good thermal insulators. (Generally, the tinier the case, the less insulation it provides.) Most heat enters and leaves the chip via the leads--especially if the chip has many. Thus, if you solder a sensor IC to a pc board, it is quite likely that the chip's temperature will be close to the board's. If you want to sense the pc-board temperature, that's fine. But if you want to sense the temperature of something else--say, a heat sink--you may have trouble. Even with the IC package in intimate contact with the heat sink, the sensor may report a temperature closer to that of the board than that of the heat sink.

Acknowledgment

Mike Fraser, senior engineer at IOtech Inc, provided the material that formed the basis of Table 1 and offered other suggestions and information on proper application of temperature sensors.

| EDN Access | Feedback | Table of Contents |