John Austin
Senior Product Engineer

Ezana Haile
Senior Applications Engineer
Microchip Technology Inc.
Chandler, Ariz.

Edited by Robert Repas

Designers face many chal lenges when placing thermal-sensing components in embedded systems. The first comes in choosing the temperature sensor. It is often not obvious whether a thermistor, resistive thermal detector (RTD), thermocouple, or other integrated sensing device makes the most sense. Key design criteria such as temperature accuracy, power consumption, and system cost affect the choice. Good decisions come from an understanding of how each device measures temperature along with its advantages and disadvantages.

Both thermistors and RTDs use the basic principle that the resistance of a material changes as the temperature changes. The change in resistance divided by the change in temperature is known as the temperature coefficient of resistance and is expressed as Ω/°C. Ideally, the temperature-sensing material should keep the same temperature coefficient over its complete measurement range creating a linear response. Unfortunately, the temperature coefficient of most materials changes for different temperatures producing a nonlinear measurement.

Many materials have a positive temperature coefficient (PTC). A PTC means resistance rises as the temperature climbs. Other materials may have a negative temperature coefficient (NTC). The resistance of NTC material falls as the temperature climbs.

The most common device used to sense temperature is the thermistor. Thermistors are made from semiconductor materials that can have either PTC or NTC characteristics. They are highly sensitive to changes in temperature and exhibit a quick response to any fluctuation. Thermistors are fairly accurate over a limited temperature range, say from 0 to 70°C. However, beyond that limited range there is significant nonlinearity.

Temperature monitoring circuits that use thermistors typically send a constant current, I, through the thermistor. The current creates an I RTH voltage drop across the thermistor (VTH) that represents thermistor temperature. The thermistor voltage feeds a low-pass filter and unity-gain buffer amplifier. The low-pass filter removes system noise from the sensor output while the unity-gain buffer amplifier drives resistive or capacitive loads without affecting the VTH value.

More complex electronics can boost thermistor accuracy over a wider temperature range. Multiple gain stages compensate for nonlinearity by applying different gains to different temperature ranges. However, this makes the expanded measuring range more costly.

Current through the thermistor must remain low. While greater current can boost resolution by increasing VTH, it also affects reading accuracy through a process called self-heating. Current passing through the thermistor creates I2RTH power that internally heats the thermistor shifting its resistance and thus its reading.

Another sensor that uses a resistance change to measure temperature is the resistive-temperature detector, or RTD. While the material in thermistors is a semiconductor, RTDs use a metallic element typically drawn into a fine wire or deposited as a metallic film. The most common metals are nickel and platinum. But they may also use other metals and special alloys.

The RTD advantage lies in its characteristic stability and repeatability. Standards and specifications for various RTDs are maintained worldwide by several national and international standards organizations. The accuracy of an RTD system with careful scaling, calibration, and resistance-totemperature conversion can run ±0.25% or better over a range of several hundred degrees Celsius.

Basic RTD circuits require a constant current source for biasing an analog circuit such as an instrumentation amplifier. Like the thermistor, the RTD develops an output voltage equal to I RRTD. However, the change in V/°C is usually much less than the thermistor. The instrumentation amplifier boosts the output voltage to a usable level. RTDs can be expensive, due to the cost of the instrumentation amplifier and manual calibration of the range, gain, and offset voltages. The amplifier output typically feeds an analog-to-digital converter (ADC) for digitization.

Another technique foregoes the instrumentation amplifier and ADC in favor of a direct conversion of resistance to frequency. For example, a relaxation oscillator circuit uses the RTD as the resistor in an RC timing network. As the RTD resistance changes, the time required to charge and discharge the capacitor varies, controlling the output frequency of the oscillator. A microcontroller can directly measure the period of the oscillator output and convert the time into a temperature.

Like thermistors, RTDs can suffer from self-heat errors if too much current passes through them. However, the overall resistance of an RTD is typically less than a few hundred Ohms compared to the thousands of Ohms in the thermistor. That helps reduce the I2R effect and the amount of heat generated. Unlike thermistors and RTDs that need an external bias current to generate an output voltage, thermocouples convert heat energy directly into electrical energy. The simplest thermocouple system consists of a thermocouple, a cold junction, and a special temperature- calibrated meter called a pyrometer.

Thermocouples have an operating temperature range that spans from –270 to 1,820°C. The American Society for Testing and Materials (ASTM) classifies thermocouples that are commercially available in terms of performance. For example, types E, J, K, N, and T are basemetal thermocouples used for temperature measurements from –270 to 1,372°C. Types S, R, and B are noble-metal thermocouples that work from –50 to 1,820°C.

Thermocouples use two metal alloys to measure temperature. For example, one wire might be a nickel-chromium alloy (chromel) and the other a nickel-aluminum alloy (alumel). Contact between the different metals generates a voltage across the contact. In the thermocouple the two metals are welded at one end and open at the other. The magnitude of the voltage generated by the welded tip varies in relation to the temperature. Higher temperatures generate higher voltages that can be measured at the open end using a voltmeter.

Some thermocouples can maintain a consistent μV/°C output over a given temperature range. However, many thermocouples are highly nonlinear and so need special linearization techniques over their full temperature range.

Every thermocouple actually has two junctions. The welded tip, or hot junction, senses the actual temperature. But when the leads are brought back together to measure this value, another junction is formed known as the cold junction. What’s measured is the difference between these two junction voltages. For accurate readings, the cold junction temperature must be known.

Early thermocouples placed the cold junction in an ice-water bath, setting the reference at 0°C. That’s obviously not practical for modern electronics. Instead, electronic thermocouple systems monitor the temperature of the cold junction using another precalibrated temperature device, such as a thermistor or RTD. The electronics then use that value to adjust or compensate the reading from the hot junction. The technique is called cold-junction compensation.

The full-scale voltage range of a thermocouple is typically less than 100 mV. Therefore, thermocouples that control machinery and monitor processes require high-performance analog signal conditioning. The leads of the thermocouple may act as an antenna for electrical noise, so many industrial applications connect the thermocouple to the instrumentation system through EMI filters for noise suppression.

Auto-zero and chopper amplifiers find use here for signal conditioning because of their low-offset voltage and common-mode-rejection (CMR) specifications. Highvalue resistors link the amplifier input terminals to the positive and negative power supplies. The high value of the resistors have no affect on the reading under normal operation. Should the thermocouple become disconnected, though, the resistors pull the amplifier beyond its full-scale value to indicate an open-circuit failure.

It should be apparent that thermistors, RTDs, and thermocouples need many additional components to take effective readings. To help simplify the designer’s task, manufacturers have combined temperature sensing with integrated circuit technology to create a new breed of silicon-based temperature sensors. Temperature-sensor ICs combine many useful features to help simplify design and lower overall system cost.

Most of these temperature sensors fall into three classes — logic, voltage, and serial-outputs. The different outputs let system designers meet the needs of different applications over the standard operating range of –55 to 150°C.

Logic-output temperature sensors become the device of choice when simple on-off control is all that’s needed. Sometimes referred to as temperature switches, logicoutput temperature ICs trigger an action when they reach a specific temperature. They come in either a “hot” option that toggles the output as temperature climbs or a “cold” option where the output toggles as temperature drops. Typically, the outputs are not latched. Therefore the switch turns off when the temperature returns to pretrigger conditions.

The trigger temperature is programmed by the value of the resister connected between the setpoint input and the supply voltage. An internal hysteresis setting of a few degrees Celsius prevents output chatter when the device switches. The hot and cold options ensure the hysteresis value has the proper relationship below or above the temperature setpoint.

Voltage-output temperature sensors produce an output voltage proportional to temperature. The analog output is typically given as a ratio of output voltage per degree Celsius. Some common temperature coefficients are 6.25, 10, and 19.5 mV/°C. Temperature-to-voltage converters can sense temperatures from –55 to 150°C. A temperature offset feature lets the device read negative temperatures without requiring a negative supply voltage. Typical operating currents are in the tens of microamps. The low current minimizes self heating from power dissipation and maximizes battery life.

The final IC temperature sensor typically connects to a host microcontroller via a two or threewire serial-data digital interface. The serial output ICs possess an integrated ADC that converts the analog value of the internal sensing element to a digital output. They can achieve temperature accuracies as high as ±0.5°C with a resolution of <0.1°C.

Many serial-output temperature sensors provide user-programmable functions, such as over and under temperature alerts and integrated EEPROM for general-purpose data storage. These features can help simplify a design, boost design flexibility, improve temperature-sensing accuracy, and lower overall system cost.

The outputs that alert over and undertemperature conditions work like those in the logic-output temperature sensors. However, instead of setting the limits with a resistor, the microcontroller loads the setpoints into a comparison register in the temperature sensor using the serial interface. At the desired temperature limit, the sensor flags the host controller about the over or undertemperature condition. The same output can also turn on a light or control a fan without having the microcontroller monitor temperature continuously. This frees up the host microcontroller and helps simplify software and hardware development.

Many applications today need less than 0.5°C temperature accuracy over a fairly wide temperature range. For higher accuracy, a calibration lookup table can correct the sensor reading at specific known temperatures. The number of calibration points depends upon the temperature range, the accuracy required, and the nonideal characteristics of the sensor.

Some temperature-sensor ICs have integrated EEPROMs to store sensor correction data. For example, the MCP98242 by Microchip Technology has 256 bytes of EEPROM built in that can store look-up tables and other generalpurpose data.