17.

3 Resistance Temperature Detectors (RTD) 597

Table 17.1 Temperature Temperature reference point description 

C
reference points
Triple pointa of hydrogen 259.34
Boiling point of normal hydrogen 252.753
Triple point of oxygen 218.789
Boiling point of nitrogen 195.806
Triple point of argon 189.352
Boiling point of oxygen 182.962
Sublimation point of carbon dioxide 78.476
Freezing point of mercury 38.836
Triple point of water 0.01
Freezing point of water (water-ice mixture) 0.00
Boiling point of water 100.00
Triple point of benzoic acid 122.37
Freezing point of indium 156.634
Freezing point of tin 231.968
Freezing point of bismuth 271.442
Freezing point of cadmium 321.108
Freezing point of lead 327.502
Freezing point of zinc 419.58
Freezing point of antimony 630.755
Freezing point of aluminum 660.46
Freezing point of silver 961.93
Freezing point of gold 1064.43
Freezing point of copper 1084.88
Freezing point of nickel 1455
Freezing point of palladium 1554
Freezing point of platinum 1769
a
Triple point is the equilibrium between the solid, liquid
and vapor phases

17.3 Resistance Temperature Detectors (RTD)

Sir Humphry Davy had noted as early as 1821 that electrical resistances of various
metals depend on temperature [2]. Sir William Siemens, in 1871, first outlined the
use of a platinum resistance thermometer. In 1887 Hugh Callendar published a
paper [3] where he described how to practically use platinum temperature sensors.
The advantages of thermoresistive sensors are in simplicity of the interface circuits,
sensitivity, and long-term stability. All such sensors can be divided into three
groups: RTDs, semiconductors and thermistors. They belong to class of the abso-
lute temperature sensors, that is, they can measure temperatures which are
referenced to an absolute temperature scale.
The RTD term is usually pertinent to metal sensors, fabricated either in form of a
wire or a thin film. Nowadays, this class also covers some semiconductor materials
598 17 Temperature Sensors

with a pronounced sensitivity to temperature (e.g., germanium). Temperature

dependence of resistivities of all metals and most alloys gives an opportunity to
use them for temperature sensing (Table A.7). While virtually all metals can be
employed for sensing, platinum is used almost exclusively because of its predict-
able response, long-terms stability and durability. Tungsten RTDs are usually
applicable for temperatures over 600  C. All RTDs have positive temperature
coefficients. Several types of them are available from various manufacturers:

1. Thin film RTDs are often fabricated of thin platinum or its alloys and deposited
on a suitable substrate, such as a micromachined silicon membrane. The RTD is
often made in a serpentine shape to ensure a sufficiently large length/width ratio.
2. Wire-wound RTDs, where the platinum winding is partially supported by a high
temperature glass adhesive inside a ceramic tube. This construction provides a
detector with the most stability for industrial and scientific applications.

Equation (4.58) gives a best fit 2nd-order approximation for platinum. In indus-

try, it is customary to use separate approximations for the cold and hot
temperatures. Callendar-van Dusen approximations represent approximations of
the platinum transfer functions:
For the range from 200 to 0  C
 
Rt ¼ Ro 1 þ At þ Bt2 þ Ct3 ðt  100 Þ : ð17:13Þ

For the range from 0 to 630  C it becomes identical to Eq. (4.58)

 
Rt ¼ Ro 1 þ At þ Bt2 : ð17:14Þ

The constants A, B, and C are determined by the properties of platinum used in the
construction of the sensor. Alternatively, the Callendar-van Dusen approximation
can be written as
 3

t t t t
Rt ¼ R0 1 þ α 1  δ 1 β 1 ð17:15Þ
100 100 100 100

where t is the temperature in  C and the coefficients are related to A, B, and C as

 
δ
A¼α 1þ
100
ð17:16Þ
B ¼ αδ  104
C ¼ αβ  108 :

The value of δ is obtained by calibration at a high temperature, for example, at the

freezing point of zinc (419.58  C) and β is obtained while calibrating at a negative
temperature.
17.4 Ceramic Thermistors 599

To conform to the ITS-90, the Callendar-van Dusen approximation must be

corrected. The correction is rather complex and the user should refer for details to
the ITS-90. In different countries, some national specifications are applicable to
RTDs. For instance, in Europe these are BS 1904: 1984; DIN 43760-1980; IEC
751:1983. In Japan it is JISC1604-1981. In the U.S.A. different companies have
developed their own standards for the α-values. For example, SAMA Standard
RC21-4-1966 specifies α ¼ 0.003923  C1, while in Europe DIN standard specifies
α ¼ 0.003850  C1, and the British Aircraft industry standard is
α ¼ 0.003900  C1.
Usually, RTDs are calibrated at standard points which can be reproduced in a
laboratory with high accuracy (Table 17.1). Calibrating at these points allows for
precise determination of approximation constants α and δ.
Typical tolerance for the wire-wound RTDs is 10 mΩ which corresponds to
about 0.025  C. Giving high requirements to accuracy, packaging isolation of the
device should be seriously considered. This is especially true at higher temperatures
where the resistance of isolators may drop significantly. For instance, a 10 MΩ
shunt resistor at 550  C results in the resistive error of about 3 mΩ which
corresponds to temperature error of 0.0075  C.

17.4 Ceramic Thermistors

The term thermistor is a contraction of words thermal and resistor. The name is
usually applied to metal-oxide sensors fabricated in forms of droplets, bars,
cylinders, rectangular flakes, and thick films. Thermistors can be also fabricated
of silicon and germanium (Sect. 17.5). A thermistor belongs to class of the absolute
temperature sensors, that is, it can measure temperature that is referenced to an
absolute temperature scale. All thermistors are divided into two groups: NTC
(negative temperature coefficient) and PTC (positive temperature coefficient).
A conventional metal-oxide (ceramic) thermistor has a negative temperature
coefficient, that is, its resistance decreases with increase in temperature. The NTC
thermistor’s resistance, as of any resistor, is determined by its physical dimensions
and the material-specific resistivity. The relationship between the resistance and
temperature is highly nonlinear (Fig. 4.18).
A ceramic thermistor (thermo-resistor) is fabricated of a crystalline material that
essentially is a semiconductor. There is a similarity between a photo-resistor and a
thermo-resistor in the way their resistances are modulated. A photoresistor, as
described in Sect. 15.4, is characterized by the energy band gap (forbidden
energies), while a thermo-resistor is characterized by the activation energy. Both
the band gap and activation energy serve as the barriers for electrons, preventing
them from moving, energy-wise, from the valence band to conduction band. For
being able to jump through the band gap, the electron’s energy shall be boosted up
either by absorbing a photon or by gaining extra kinetic (thermal) energy. In more
detail this process is described in Sect. 17.5.
600 17 Temperature Sensors

Whenever a high accuracy is required, or the operating temperature range is

wide, thermistor characteristics should not be taken directly from a manufacturer’s
data sheet. Typical tolerances of the nominal resistance (at 25  C) for the mass
produced thermistors may be rather wide: 5 % is quite common, however for a
higher price, a 1 % or even better thermistors are readily available. Unless it was
produced with tight tolerances, to reach a high accuracy, each low-tolerance
thermistor needs to be individually calibrated over the entire operating
temperature range.
Manufacturers can trim a ceramic thermistor by grinding its body to a required
dimension that directly controls the nominal value of its resistance at a set temper-
ature (typically 25  C). This, however, increases cost. An alternative approach for
an end-user is to individually calibrate thermistors. Calibration means that a
thermistor has to be subjected to a precisely known temperature (a stirred water
bath is often employed3) and its resistance is measured. This is repeated at several
temperatures if a multi-point calibration is needed (Sect. 2.2). Naturally, a
calibration is as good as the accuracy of a reference thermometer used during the
calibration. To measure resistance of a thermistor, it is attached to a measurement
circuit that passes through it an electric current. Depending on the required accu-
racy and production cost restrictions, a thermistor calibration can be based on use of
one of several known approximations (models) of its temperature response.
When a thermistor is used as a temperature sensor, we assume that all its
characteristics are based on the so-called zero-power resistance, meaning that
electric current passing through a thermistor does not result in any noticeable
temperature increase (Joule self-heating) which may affect accuracy of measure-
ment. A static temperature increase of a thermistor due to a self-heating is governed
by the following equation:

N2 V 2
ΔT H ¼ r ð17:17Þ
Rt
where r is a thermal resistance from the thermistor to surroundings, V is the applied
d.c. voltage during the resistance measurement, Rt is the resistance of a thermistor at
a measured temperature, and N is a duty cycle of measurement (for example,
N ¼ 0.1 means that constant voltage is applied to a thermistor only during 10 %
of the time.) For a continuous d.c. measurement, N ¼ 1.
As follows from Eq. (17.17), a zero-power can be approached by selecting high
resistance thermistors, increasing coupling to the object of measurement
(reducing r), and measuring its resistance at low voltages that are applied during
short time intervals. Below in this chapter we will show effects of a self-heating on
the thermistor response, but for now we assume that a self-heating results in a
negligibly small error.

3
Actually, water is not used with the unprotected thermistors. Mineral oil or Fluorinert® electronic
fluid are more practical liquids.
17.4 Ceramic Thermistors 601

To use a thermistor in the actual device, its transfer function (temperature

dependence of a resistance) must be accurately established. Since that function is
highly nonlinear (Fig. 4.18) and specific for each particular sensor, an analytical
equation connecting the resistance and temperature is highly desirable. Several
mathematical models of a thermistor transfer function have been proposed. It
should be remembered, however, that any model is only an approximation and
generally, the simpler the model the lesser accuracy should be expected. On the
other hand, at a more complex model, calibration and practical use of a thermistor
become more difficult. All present models are based on the experimentally
established fact that logarithm of a thermistor’s resistance Rt relates to its absolute
temperature T by a polynomial equation:

A1 A 2 A3
lnRt ¼ A0 þ þ þ ; ð17:18Þ
T T2 T3
From this basic equation, three static transfer functions (models) have been
proposed.

17.4.1 Simple Model

The Simple Model is the simplest approximation of the thermistor transfer function.
Over a relatively narrow temperature range and accepting that some accuracy may
be lost, we can eliminate two last terms in Eq. (17.18) and write it in form [4]:

β
lnRt ffi A þ ; ð17:19Þ
T
where A is a dimensionless constant and β is another constant called the material
characteristic temperature (in kelvin). If the thermistor’s resistance R0 at the
calibrating temperature T0 is known, then the resistance-temperature relationship
(transfer function) is expressed as:

β T T 0
1 1

Rt ¼ R0 e ð17:20Þ

Equation (17.20)is the most popular and widely used thermistor model. As
shown in Sect. 17.5, it also can be derived from the Svante Arrhenius equation
that describes the rate of chemical reactions as function of temperature. An obvious
advantage of this model is a need to calibrate a thermistor at only one point (R0 at
T0). This, however, assumes that value of β is already known, otherwise a two-point
calibration is required to find the value of β:

lnRR10
β¼ ; ð17:21Þ
1
T1  T10
602 17 Temperature Sensors

where T0 and R0, T1 and R1 are two pairs of the corresponding temperatures and
resistances at two calibrating points on the curve of Eq. (17.20). The value of β in
this model is considered temperature independent, but may vary from part to part
due to the manufacturing tolerances which typically are within 1 %.
When the thermistor is used, its resistance Rt is measured. From that resistance,
temperature in K can be computed from an inverse transfer function that is:

R
!1
1 lnS0t
T¼ þ ð17:22Þ
T0 β

Error from the approximation of Eq. (17.20) is small near the calibrating
temperature T0, but increases noticeably while moving away from that point.
Beta (β) specifies the thermistor curvature, but it does not directly describe its
sensitivity, which is the negative temperature coefficient, α that can be found by
differentiating and normalizing Eq. (17.20)

1 dRt β
αr ¼ ¼ 2 ð17:23Þ
Rt dT T
It follows from Eq. (17.23) that the sensitivity depends on both: beta and
temperature. Being a highly non-linear sensor, a thermistor is much more sensitive
at lower temperatures while its sensitivity drops fast with a temperature increase. In
the ceramic NTC thermistors, the sensitivity α varies over the temperature range
from 2 % (at the warmer side of the scale) to 8 %/ C (at the cooler side of the
scale), which implies that an NTC thermistor, albeit a nonlinear sensor, is a very
sensitive device, roughly an order of magnitude more sensitive than any RTD. This
is especially important for the applications where a high output signal over a
relatively narrow temperature range is desirable. Examples are the medical elec-
tronic thermometers and home thermostats.

17.4.2 Fraden Model

In 1998, the author of this book proposed a further improvement of the Simple
Model [5]. It is based on the experimental fact that in many thermistors, the
characteristic temperature β is not a constant but rather a function of temperature
(Fig. 17.9). Depending on the manufacturing process and type of a thermistor, the
function may have either positive slope, as shown in the picture, or negative.
Ideally, β should not change with temperature, but that is just a special case
which can be seen only from the best manufacturers who tightly control composi-
tion of the ceramic material. In such cases, the Simple Model provides a quite
accurate basis for temperature computation. But for a relatively inexpensive sensor,
the Fraden model should be considered.