Temperature is one of the commonest quantities to be measured electronically. There are a large number of phenomena which may be used in temperature sensing but in integrated circuits the relationship between voltage and current in a diode is one of the easiest to exploit. Usually the diode used is the base-emitter diode of a bipolar junction transistor (BJT).
The base emitter voltage of a bipolar junction transistor (BJT) is related to its collector current by the formula
where k is Boltzmann's Constant (1.38×10-23 J/K), T is the absolute temperature, q is the charge of an electron (1.6 ×10-19 C), and Is is a current related to the geometry and the temperature of the transistor. (This assumes a voltage of at least a few hundred mV on the collector, and ignores Early effects.)
The current Is is strongly temperature dependent so this measurement cannot easily used to derive the device temperature.
But if we reduce the current in the transistor by a factor of N, or if we take N transistors identical to the first and allow the total current Ic to be shared equally among them, we find that the new base emitter voltage is given by
and the expression for the difference between the two base emitter voltages is proportional to absolute temperature but does not involve Is.
So if we have a silicon transistor which is located so that it is at the temperature we wish to measure all we need do is to connect its base to its collector and pass two different currents through it, with a ratio of N:1, and measure the difference in the two voltages, which is proportional to the absolute temperature of the transistor junctions. This technique is often used to measure the temperature of large digital processors, which often contain a transistor intended to be used for chip temperature measurements. (It is more accurate to use a transistor connected in this way than to use a simple diode as the current gain of the transistor minimises resistive errors.)
To make a self-contained IC temperature sensor we can arrange a circuit, as shown in the diagram above, containing N+1 NPN BJTs, a temperature stable resistor, and a current mirror using two PNP BJTs. The total current in the circuit will be proportional to its absolute temperature, as will the voltage across the resistor (which is the difference between V and VN in equation ).
This circuit is the very widely used basic "band-gap" temperature sensor. The circuit in the diagram is incomplete, because it is stable with all the transistors turned off and no output, and therefore requires turn on circuitry to ensure that when the supply voltage is applied the transistors will turn on and not simply remain non-conducting.
The AD590, AD592, and AC2626 use this exact circuit (plus starter) and are two terminal temperature sensors with current equal to 1 µA/K when a voltage between 4 and 30 V is applied to them. Other IC sensors may have PTAT (proportional to absolute temperature) voltage output (TMP-01), offset voltage output where some specified voltage corresponds to s specified temperature (TMP-35/36/37), PWM (TMP-03/04/06), or digital outputs with various resolutions - but almost all of them use this basic temperature sensing technique.
©James Bryant - March 2008
Analog Devices Inc.