Circuit designers need to keep temperature excursions to a minimum on increasingly hot, crowded boards. Integral digital temperature sensors and some control circuitry can do the trick.
Claire O’Keeffe and Donal McNamara
Analog Devices BV
Printed circuit boards are getting hotter and hotter as more and more is squeezed onto them. The thermal properties of these boards have thus far been monitored and controlled by assorted conventional devices—thermistors, RTDs, and thermocouples.
Thermistors have long been favored for their very small form factor, low cost, and high sensitivity. On the negative side, they operate over a limited temperature range and are often difficult to replace due to lack of industry standards. They also require compensation circuitry to overcome nonlinearity. RTDs are generally used where accuracy and stability are crucial, but rarely where cost is a deciding factor. Thermocouples are ideal for monitoring extreme temperatures, but get poor marks in accuracy and stability and must be thoroughly tested under controlled conditions. (For a detailed comparison of RTDs, thermistors, thermocouples, and IC temperature sensors, see the sidebars, “Summary of the Main Differences” and “Strengths and Weaknesses.”)
Thanks to advances in IC technology, board designers have a new alternative to discrete temperature sensors: digital temperature sensors. These devices, competitively priced, available in tiny packages, highly accurate, and able to operate over a wide temperature range, are beginning to replace their precursors in many commercial applications.
Digital Temperature Monitoring
In contrast to analog sensors, digital temperature detectors are complete in themselves, requiring no additional circuitry for signal conditioning or linearization. They can be directly connected to the microcontroller, saving design time, PCB area, and expense. They can flexibly reduce their current consumption, particularly useful in battery-operated applications where minimal power consumption is important. The user can also program temperature limits (THIGH and TLOW) where an alarm is required. If a programmed limit is exceeded, an interrupt is generated and the microcontroller can take action. Many IC design systems integrate ADCs and DACs onto a single chip to save on board space and reduce cost. Some have built-in algorithms that drive cooling fans at the optimum speed while consuming the least amount of power and keeping noise to a minimum. These sensors are also particularly useful for closed-loop applications in general.
Principles of Digital Temperature Sensing
Digital temperature sensors are based on the principle that the base-emitter voltage, VBE, of a diode-connected transistor is inversely proportional to its temperature. When operated over temperature, VBE exhibits a negative temperature coefficient of approximately –2 mV/°C. In practice, the absolute value of VBE varies from transistor to transistor. To nullify this variation, the circuit would have to calibrate each individual transistor. A common solution to this problem is to compare the change in VBE of the transistor when two different current densities are applied to the emitter of the transistor (see Figure 1).
Figure 1. The temperature sense transistor is incorporated onto the IC’s die and configured as a diode. Two current sources (I and N × I) are switched through the sense transistor to determine VBE, which is gained up and converted into a digital format. The bias diode-transistor prevents any ground noise from filtering through the gain switches.
The formula used to extract the temperature from this technique is:
With I1 = NI2, the formula becomes:
current ratio of I1/I2
charge on an electron
absolute temperature in Kelvin
N, k, and q are all known constants, so:
VBE = (constant) (T)
T = (constant) (VBE)
The bias diode configured transistor in Figure 1 prevents ground noise from interfering with the measurement by biasing the negative input of the amplifier up one diode drop from ground. The maximum VBE from the sense transistor is in the tens of millivolts, so a gain stage is required to allow effective
Figure 2. The remote temperature sensing transistor is external to the IC and can be implemented by any standard transistor. If the sense transistor is far from the IC, a twisted shielded cable should be used as the interface. The low-pass filter prevents any noise from going through the gain amplifier.
processing of the signal by the ADC. This gain stage uses a switched capacitor circuit and a chopper amplifier. The ADC converts the analog proportional-to-absolute temperature signal into a digital word in °C.
The operating principle described thus far is for digital temperature sensing ICs that incorporate the sensing transistor. This method is usually called local temperature sensing. It is also possible to have the transistor external to the IC, a configuration commonly known as remote temperature sensing (see Figure 2). When an external transistor is used, an extra parameter must be added to Equations 1 and 2. This parameter is the transistor’s ideality factor, nf, a measure of its deviation from ideal behavior. The equation for remote temperature sensing is:
nf is not a constant, and so introduces an error into the temperature reading. Some digital temperature sensors are factory trimmed to compensate for a particular nf value, with the trimmed figure given in the sensor data sheet. The ADM1032, for example, is trimmed for an nf value of 1.008. If the nf of the transistor used does not match 1.008, the following formula can be used to calculate the error introduced at a temperature T°C:
This offset error can be written to a register in the ADM1032, and is automatically added to the temperature measured.
When using remote temperature measurement, standard PNP transistors (e.g., 2N3906) can be configured as diodes, and operate exactly like the local temperature sensors. A low-pass filter of fC = 65 kHz is included in the signal path since the remote transistor is external to the IC and noise can be picked up by D+ and D–, the lines connecting the sensor with the remote thermal
Figure 3. In the preferred layout of the sense transistor signal tracks, the gnd lines help prevent signal coupling from adjacent high-frequency lines.
transistor. For optimum performance, it is best to position the remote transistor close to the D+ and D– pins of the temperature-measuring IC. If proper precautions are taken, the remote transistor can be positioned several feet away from the D+ and D– pins. These precautions may entail using shielded twisted-pair cable if the remote thermal diode is far from the PCB. If the diode is on the same PCB, the D+ and D– traces should be routed in parallel with a gnd plane and kept away from noisy signals such as clocks or high-speed data lines (see Figure 3).
One major source of error associated with using a remote sensing transistor is the series resistance on the D+ and D– lines. This resistance causes an offset voltage that must be individually calibrated out when using a 2-current source temperature sensor. Offset registers onboard the IC are used to nullify this offset voltage and are programmed by the user after calibrating the path to the remote transistor. Leakage on a PCB also contributes to temperature error, so refer to the data sheet for optimum PCB layout.
Temperature Sensor Applications
The burning application question is how will temperature measurement benefit a system. In the not-so-distant past, designers would have used heat-dissipating techniques such as heat sinks on power amplifiers, copper plating on PCBs to spread the heat, or air vents in the housing to allow circulation of cooling air. If these techniques did not do the job, discrete “set and forget” temperature sensors would be designed into the system to detect dangerous temperature levels. When the temperature reached a point deemed destructive, the sensor would switch off the affected circuit, turn on a fan, or take other corrective action.
Digital temperature sensors allow the master devices in a system to track the temperature profile of circuits more efficiently than ever before. Temperature sensors are becoming more accurate (±0.5°C max.), a lot smaller (SOT23, at 2.9 × 2.8 mm and SC70, at 2 × 2.1 mm), and more integrated (ADCs, DACs, and fan control). These sensors have become part of control loops where the salient concerns are protection of circuitry and increased efficiency.
One typical example is temperature control of crystal oscillators in mobile phones (see Figure 4).
Figure 4. A common application in mobile phones is to monitor the temperature of the crystal oscillator. Temperature greatly affects the frequency output of these oscillators, so the processor uses a look-up table to maintain a constant frequency over all temperatures.
Here, the communication frequency must remain stable over the full operating temperature range because if it changes, the phone cannot communicateýwith its network. Placing a digital temperature sensor as close as possible to the oscillator allows the system processor to monitor its temperature. The sensor output is a 10-bit digital word that gives a temperature reading with 0.25°C resolution. The software uses this reading in a formula, then plots the transfer function of oscillator frequency vs. temperature. The frequency output of the oscillator can be voltage controlled; if the temperature changes, the processor will instruct the power management system to alter the control voltage and thus maintain a stable frequency over temperature.
Semiconductor manufacturers commonly incorporate into their temperature sensors extra functions previously handled by specialized ICs. Because an ADC is already part of a digital temperature IC, and used for only a small part of the time to measure temperature, some IC fabricators have included analog input channels by simply putting a multiplexer before the ADC input. The ADCs of digital temperature ICs typically have slow conversion times, so the analog input channels can measure only slowly varying signals. High-speed signal conversion is not possible. A common application for these analog inputs is system voltage monitoring, where an accuracy of 1% FSR would be expected.
Newer devices combine DACs and temperature sensors into one package. The reasoning is that once the temperature of a thermal zone has been measured, a decision must be made on the significance of this temperature and appropriate action taken depending on the temperature level (see Figure 5).
Figure 5. Many control systems use a DAC in the control loop to alter a device’s performance as a result of a temperature reading. In the example shown here, ADI’s ADT7316 measures the temperature of one thermal zone and the microcontroller uses the DAC to alter the speed of the fan that controls this thermal zone.
Here, we have a typical remote thermal transistor in an area of the circuit—near the processing ICs or power amplifiers—where the temperature is most likely to increase. Every circuit has an optimum performance temperature, usually kept under control by a fan, and the DAC output of the ADT7316 in Figure 5 can be programmed to control fan speed by means of an op-amp. The higher the voltage to the fan, the faster it rotates. As the temperature drops to within its specified limits, the fan can be slowed or turned off. Running it all the time increases the system’s noise level. Complete temperature monitoring systems have been designed to monitor and control the speed of many fans as well as measure critical supply voltages and multiple thermal zones including any diode-connected transistor connected on a high-performance IC or discrete transistor.
Time is the most measured parameter in the world, and temperature is second. Circuit designers have several candidate technologies from which to choose as they try to cool their boards while balancing efficiency and price considerations. There are now various thermal simulation packages on the market that can give a fairly good indication as to where hot spots are going to occur. Once a potential hot spot is identified, the designer can insert a digital temperature sensor and some control circuitry that will counteract temperature excursions. Interfacing these devices is easy via I2C or SPI protocols by the master device for constant monitoring of this critical parameter.
Summary of the Main Differences Among RTDs, Thermistors, Thermocouples, and IC Sensors
Metal oxide ceramic
Two dissimilar metals
Cost of sensor (relative)
Moderate to low
Moderate to low
Cost of system (relative)
Moderate to low
Moderate to low
-200°C to 850°C
-100°C to 500°C
-270°C to 1800°C
-55°C to 150°C
10%, 2°C typ.
Strengths and Weaknesses of RTDs, Thermistors, Thermocouples, and IC Sensors
Moderate temperature range
Cost: 4–10 × more expensive than thermocouples
Not suitable for high-vibration environments
O/p values prop to temp, no additional circuitry required
Addressability, data storage and retrievability
Available in tiny packages
Limited temperature range
Good resolution, resistance vs. temperature characteristics is large
High resistance allows for longer lead wires without lead wire compensation
Available in tiny packages
React fast to changes in temperature
Resistance vs. temperature characteristics nonlinear, require additional circuitry
Limited temperature range
Lack of industry standards makes thermistors difficult to replace
Wide temperature range
Must be tested under controlled conditions
Thermocouple extension wires must be used in hooking up to equipment, hence results in error when ambient temperature changes