Electronic Design

Sense junction temperatures without calibration

Transistors can be used as a low-cost temperature sensors. Sensing temperature with a silicon junction often exploits the fact that the forward voltage drop has a temperature coefficient of −2.2 mV/°C. The problem with this method is that the actual junction voltage at a given temperature is subject to wide variation, requiring a calibration whenever different devices are used. Manual calibration can be tedious and also involves accuracy-degrading potentiometers.

Sensing junction temperature without calibration relies on the predictability of junction voltage variation at two currents over temperature, described by the familiar:

Rearranging to yield T, assuming a 10:1 ratio for I1/I2, and assuming 8.7248−5 for K/q, this simplifies to a good approximation for most common bipolar transistors (2N2222, 2N2907, 2N3904, 2N3906, etc.):

This result is in degrees Kelvin; simply subtract 273 to convert the result to degrees Celsius.

As an idea of the amplitudes involved, a 10:1 current ratio will yield a 59.6-mV difference between the junction drop at high current and low current, at 25°C, with a temperature coefficient of +200 µV/°C.

You can stimulate a junction to yield this information by toggling it between two currents, as the circuits described here do. Then you simply measure the dV between the two current values and use the above equation to describe the temperature of the junction. This small signal represents a slight challenge when amplifying, since you are trying to observe a tiny signal (the tempco) contained in a larger signal (the dV/I), riding on top of a big imprecise diode drop with a comparatively huge tempco (0.4 to 0.7 V, −2,2 mV/°C). These constraints favor a means of stimulation that allows ac coupling the dV/I voltage to an amplifying circuit. These variations, available supply voltage, and the dynamic range of the device (ADC) that this circuit will feed dictate the maximum allowable gain of the amplifier.

The basic approach for junction temperature sensing with a hardware monitor IC such as the LM80 is shown in Figure 1. The general purpose output of the LM80 is used as a control line for a circuit which continuously supplies 10 µA of current to the measurement junction when the control line is low.

To make a measurement, the control line is momentarily taken high, supplying an additional 90 mA to the junction for a total of 100 mA. The amplifier/conditioner circuit is designed to extract and amplify the difference between the low-current and high-current junction voltages. This voltage then is sensed by one of the analog inputs of the LM80.

The circuits described here are all designed to feed their output into a National LM80 Personal Computer Hardware Monitor IC. The LM80 is simply an analog-to-digital converter, converting the voltage to digital form. Once in a computer, the actual temperature is simply a matter of software calculation. The LM80 provides a 0- to 2.56-V full-scale range with 10-mV LSB (least significant bit).

The LM80 presents certain timing challenges because of its “Round-Robin” input sampling, which prevents the user from knowing exactly when the LM80 is converting a given input. Usually this requires considering the conversion time to be as long as 1.5 seconds, a time that requires impractical capacitor values.

You can achieve better control timing problem by using IN0 of the LM80, which is the second item converted during its Round-Robin loop, just after the LM80 converts its internal temperature. In addition, the LM80 should be stopped prior to making a junction measurement. By starting the LM80 at exactly the same time the control line high is issued, the maximum delay before conversion will be 224 ms. It’s these conditions that dictate that the control pulse should go high, then as soon as 224 ms elapse, read the LM80, and take the control line low.

While additional time is required past the 224 ms (this is system-dependent; in the system these circuits were tested on, a 133-MHz Pentium running Visual Basic software, it took around 250-300 ms total), the amplifier/conditioner need only have accuracy within those first 224 ms.

Amplifier/Conditioner A of Figure 1 is simply an ac-coupled amplifier. While Amplifier/Conditioner A is the simpler of the two options, it requires a quality 1-µF input coupling capacitor with low leakage and low dielectric absorption.

Amplifier/Conditioner B in Figure 2 is a peak detector circuit. This reduces the required capacitance on the input coupling capacitor by an order of magnitude with a trade-off of complexity. For the circuit as tested, using an X7R ceramic for CPEAK worked fine. Q2 provides a reset of the peak detector between acquisitions. Though this doesn’t take the peak detector to a “clean” zero, it drops the peak detector to a value lower than those which it is measuring (although it limits the lower temperature range of the circuit to around −55°C).

The most critical factor in achieving the necessary accuracy is ensuring that the actual current ratio is identical to that used in the equation to calculate temperature. In the example circuits shown, the ratio is 10:1. A current ratio mismatch of 1 % will cause an error of approximately ±2.5°C at 25°C. In any design it will be necessary to carefully determine the exact voltage available to RLOW and RHIGH. The resistors selected for RLOW and RHIGH should be 1% or better.

One means of providing the absolute best possible accuracy would be to drive both RLOW and RHIGH from logic lines coming from the same chip, and equally load them (Fig. 3). This causes the variations in source voltage to the two resistors to track better. The line to RLOW is a “dummy” line that’s kept high at all times. RLOAD is a resistor to match loading between the RLOW and the RHIGH line, and the diode in the RLOW line ensures a complete match in the behavior of both lines. RLOAD should be selected to match the current out of both outputs based on the exact voltage available at those outputs.

Used with the LM80 as shown, a gain of 30 would provide a signal that would just reach the upper limit of the dynamic range of the LM80 at 125°C (2.55 V). A gain of 50 (50k for RGAIN) would provide a signal proportional to 10 mV/K. This output can be offset by setting the offset resistor shown (ROFFSET) to 88.7k to provide a convenient analog signal of 250 mV at 25°C. This requires a negative supply rail on the op amp to read negative temperatures. Or, in single supply circuits, select a different, convenient offset such as 0.5 V (ROFFSET = 107 k), so that 750 mV corresponds to 25°C. The scaling and offset possibilities are endless.