Thermistor Temperature Sensing: Circuits and Design

Thermistors are among the most widely used temperature sensors in electronic systems, yet their strongly nonlinear resistance–temperature characteristic often makes practical implementation less intuitive than simple RTD or IC sensors. Thermistor temperature sensing converts the resistance of an NTC thermistor into temperature using controlled excitation, an analog-to-digital converter and a sensor-specific resistance–temperature model. This article explains how design engineers can select the sensor, compare current and voltage excitation, implement ratiometric measurement, and control self-heating, conversion errors and diagnostic limitations in embedded, industrial and instrumentation systems.

Key Takeaways

An NTC thermistor provides high temperature sensitivity, but nominal resistance alone does not define its curve, accuracy, operating range or interchangeability.
A voltage divider supplied from the ADC reference provides a straightforward ratiometric interface. Its output is a resistance ratio, not a voltage directly proportional to thermistor resistance.
Choose excitation and the fixed resistor from the complete temperature range, ADC input limits and allowable self-heating. Matching the fixed resistor to the thermistor resistance at the critical temperature is a useful starting point, not a universal requirement.
Use the manufacturer’s resistance–temperature table, beta model or Steinhart–Hart coefficients over their specified range. ADC resolution does not establish temperature accuracy.
Ratiometric measurement suppresses common excitation-scale errors, but thermistor tolerance, fixed-resistor drift, leakage, thermal mounting and self-heating remain in the error budget.
Validate input headroom, settling, open- and short-circuit detection, and installed thermal response before releasing a single-channel or multiplexed design.

The system discussion builds on the Analog Dialogue article by Jellenie Rodriguez and Mary McCarthy and the associated CN-0545 reference design. The AD7124-4 and AD7124-8 examples illustrate an integrated precision front end; the selection principles also apply to other converters whose input, reference and filtering specifications meet the application requirements. [1, 3]

Thermistors vs. RTDs

Thermistors and resistance temperature detectors measure temperature through a change in resistance. The useful distinction is not simply cost or accuracy: sensor material, resistance–temperature characteristic, packaging, calibration and installation determine which technology fits the measurement.

Comparison of negative-temperature-coefficient thermistor, switching positive-temperature-coefficient thermistor and RTD resistance–temperature characteristics — Figure 1. Representative resistance–temperature characteristics of NTC thermistors, switching PTC thermistors and RTDs.

An NTC thermistor decreases in resistance as temperature rises. Temperature-sensing NTC elements commonly use mixed metal-oxide semiconductor ceramics. Conventional metallic RTDs, including platinum types, increase in resistance with temperature and have a substantially more linear characteristic over their useful range. Figure 1 compares these basic responses. [1, 2] A platinum RTD can cover a wider temperature range than many NTC sensing assemblies; the original comparison cites −200 °C to +850 °C for the platinum RTD family. This is not a rating for every probe. NTC operating limits likewise depend on the exact element, encapsulation, leads and assembly, rather than being a direct consequence of the exponential curve. [1, 2] Small NTC beads and chips can provide fast response and high sensitivity in compact assemblies. However, an encapsulated NTC probe is not necessarily faster than a small RTD element. Thermal mass, mounting, immersion depth, airflow, potting and the thermal path to the measured object can dominate the response. Compare thermal time constants only when their test conditions are comparable. [2] Table 1 maps the technologies to common application classes. These are selection tendencies, not guaranteed performance rankings. Precision laboratory instruments may use either a calibrated NTC or an RTD, depending on the required span, interchangeability, stability and measurement uncertainty.

Application class	Typical sensor choice	Main selection considerations
Industrial process measurement over a wide temperature span	Platinum RTD	Established resistance–temperature characteristics, wide-range capability and suitable probe constructions; check the exact assembly rating.
Precision instruments over a restricted temperature span	Precision NTC or RTD	NTCs provide high sensitivity; compare calibration uncertainty, interchangeability, aging and installed thermal behavior.
HVAC, building controls and appliances	NTC	Cost, available probe styles, useful sensitivity, environmental sealing and response appropriate to the control loop.
Overtemperature thresholds and resettable current limiting	Application-specific switching PTC or polymer PTC	Switching behavior, voltage and current ratings, thermal coupling, residual current and reset conditions.
Power-supply inrush limiting	Power NTC; selected PTC arrangements for specific circuits	Use components designed for the required startup energy, current, operating temperature and restart behavior; a sensing NTC is not a power limiter.
High-temperature surface measurement	Suitable RTD or high-temperature NTC assembly	Check separate sensor-head, cable, connector and mounting limits. Selected NTC assemblies specify sensor-head operation up to 300 °C. [6]

Table 1. Typical sensor choices by application. NTC sensing and switching PTC behavior are described in References [2] and [5].

Thermistors also complement other sensors. For example, a thermocouple instrument can use an NTC for cold-junction compensation, provided the NTC measures the relevant terminal temperature rather than an unrelated PCB hot spot. The required temperature span, thermal arrangement and error budget determine the sensor choice. [1]

NTC sensing and switching PTC protection are different functions

The switching PTC characteristic in Figure 1 represents a protection-oriented ceramic device. Its resistance rises strongly near a transition associated with the ceramic’s Curie region. Under overload, electrical self-heating can move the device into a high-resistance state and reduce current, provided the circuit remains within the device’s voltage, current and thermal ratings. It does not necessarily interrupt current completely. [5] The switching temperature is not the PTC’s full operating-temperature range. Nor should a switching PTC curve be treated as representative of every positive-temperature-coefficient sensor. This article concentrates on low-power NTC resistance measurement, not the design of resettable protection or power inrush-limiting components. [2, 5] Table 2 compares NTC sensing, switching ceramic PTCs and conventional metallic RTDs. Cost, size and response depend on the complete assembly; protection components and precision sensing elements are not interchangeable simply because both are called thermistors.

Parameter	NTC temperature sensor	Switching ceramic PTC	Metallic RTD
Primary role	Continuous temperature measurement and compensation	Temperature thresholds, current limiting or self-regulating functions	Continuous temperature measurement
Temperature characteristic	Resistance decreases nonlinearly	Strong resistance rise in the switching region	Resistance increases; comparatively near-linear
Usable temperature range	Element- and assembly-specific; the worked 44031 example covers −50 °C to +150 °C. [3]	Operating limits and switching temperature must be specified separately	Element- and probe-specific; platinum families can cover much wider spans
Sensitivity	High relative resistance change per temperature change	Very high near the threshold, but not generally intended for wide-range precision measurement	Lower relative resistance change than a typical NTC
Response time	Depends on size, package, mounting and medium	Depends on thermal design and operating mode	Depends on element, sheath, mounting and medium
Measurement excitation	Required; keep self-heating within the error budget	Application-dependent; self-heating may be intentional	Required; self-heating remains relevant
Wiring	Often two-wire; verify lead and contact errors	Usually two-terminal	Two-, three- or four-wire according to accuracy requirements
Cost and size	Small, low-cost elements available; precision probes may cost more	Strongly dependent on voltage, current and power duty	Strongly dependent on accuracy class, construction and probe requirements

Table 2. Practical comparison of NTC sensors, switching ceramic PTCs and metallic RTDs. [1–3, 5]

Thermistor Temperature Sensing: Measurement Challenges

A thermistor is a passive resistive sensor: it requires electrical excitation before an ADC can measure a temperature-dependent voltage. The complete channel must control both electrical measurement errors and the difference between the thermistor body temperature and the temperature of interest.

Block diagram of an NTC thermistor temperature measurement signal chain — Figure 2. Typical signal-chain functions in an NTC thermistor measurement system.

Figure 2 summarizes the signal chain. Integrated precision ADCs can combine conversion, buffering, programmable gain, reference generation and diagnostics, but integration does not remove the need to check the sensor, passive components and operating conditions. The main design questions are: [1, 3]

Which nominal resistance, resistance–temperature curve, tolerance and package satisfy the measurement range and environment?
What excitation current or voltage provides sufficient signal without excessive self-heating?
Do the sensor and reference voltages remain inside the ADC’s differential-input, absolute-input and common-mode limits at every temperature?
What buffering and analog filtering are required for the source impedance, interference and update rate?
How will firmware reconstruct resistance, linearize the curve and handle invalid measurements?
Can excitation, reference components or ADC channels be shared between sensors without unacceptable loading, switching errors or settling delays?
Which errors dominate before and after calibration, and how will performance be validated in the installed assembly?

Thermistor Selection Guide

Start with the required temperature range, accuracy, response and environment, then select the resistance and interface. Two thermistors described as 10 kΩ at 25 °C can have different curve shapes, tolerances and mechanical ratings, so they are not automatically substitutes. Nominal resistance is commonly specified at 25 °C and written R₂₅. A 10 kΩ device therefore has a nominal resistance of 10 kΩ at that temperature. The resistance at the cold and hot endpoints must come from the selected device’s resistance–temperature data, not from its nominal value alone. [1, 2] Higher resistance can reduce divider current, but it also increases sensitivity to leakage, input bias current and contamination. Lower resistance may ease some electrical-interface constraints while requiring more care with excitation power. Neither a 10 kΩ threshold nor any other nominal value establishes a universal upper operating temperature. [2, 4]

Package, environment and long-term stability

NTC sensing elements are available as beads, radial-leaded parts, SMD chips and assembled probes. Glass encapsulation, epoxy coatings, potting and metal housings change moisture resistance, mechanical robustness, thermal response and temperature capability. A coating is not automatically hermetic, chemically compatible or suitable for immersion. [2] For surface measurements, the sensor must couple thermally to the target while avoiding excessive heat flow through its leads or surrounding PCB. For ambient-air measurement, nearby regulators, processors and power resistors can create a local temperature different from the intended measurement point. Compare sensor placement and attachment as carefully as the electrical tolerances. Long-term stability depends on material, stabilization, encapsulation, temperature exposure, humidity and mechanical stress. Use the manufacturer’s endurance and environmental-test results for the exact series; a generic annual drift figure is not a reliable prediction for every installation. TDK’s technical information also notes that thermal parameters measured in still air do not remain unchanged after mounting or exposure to another medium. [2] High-temperature capability can be localized. For example, the TDK M1703 assembly specifies a sensor-head maximum of 300 °C, but the cable, connections and installation still require their own assessment. This illustrates why a headline temperature limit should not be applied indiscriminately to the entire assembly. [6]

Accuracy, interchangeability and curve tolerance

Thermistor accuracy may be specified as resistance tolerance at one temperature, beta tolerance, temperature interchangeability over an interval, or a resistance–temperature tolerance band. These specifications describe different things. A tight R₂₅ tolerance does not by itself guarantee a tight temperature error far from 25 °C because curve-shape tolerance also contributes. [2] As a device-specific example, CN-0545 uses a 10 kΩ Type 44031 thermistor over −50 °C to +150 °C and quotes ±0.1 °C accuracy from 0 °C to +70 °C, with ±1 °C over the wider range. These are sensor specifications cited for that design, not guaranteed accuracy for every 10 kΩ thermistor or for the complete installed system. [3] Purchasing specifications should identify the exact curve or coefficients, their resistance units, the tolerance definition and the package. An alternative supplier must meet the required resistance–temperature envelope and environmental requirements, rather than only matching R₂₅.

Linearization: beta vs. Steinhart–Hart equation

The beta model approximates an NTC characteristic using a parameter determined between two temperatures. In Eq. (1), β is the beta value in kelvins, R₁ and R₂ are resistances at absolute temperatures T₁ and T₂, and ln is the natural logarithm. Both temperatures must be expressed in kelvins. [2]

\beta = \frac{\ln(R_{1}/R_{2})}{1/T_{1}-1/T_{2}} \tag{1}

Common beta designations include β_25/50 and β_25/85; manufacturers may specify other temperature pairs. Select the model appropriate to the measurement interval and check its residual error against the manufacturer’s resistance–temperature table. Beta is a fitted description over an interval, not an exactly constant material parameter across every temperature. [2]

For conversion from measured resistance R to absolute temperature T, Eq. (2) uses a known reference resistance R₀ at absolute temperature T₀. When the reference point is 25 °C, T₀ is 298.15 K. Convert the calculated temperature to degrees Celsius by subtracting 273.15. [3]

T = \left[\frac{1}{T_{0}}+\frac{1}{\beta}\ln\left(\frac{R}{R_{0}}\right)\right]^{-1} \tag{2}

For wider spans or tighter interpolation requirements, the Steinhart–Hart relationship in Eq. (3) often provides a better fit. A, B and C are the coefficients for the selected thermistor and fitting interval. Here r is the numerical resistance value in ohms, defined as R divided by 1 Ω, so the logarithm has a dimensionless argument. Coefficients supplied for another resistance unit must not be used unchanged. [2, 3]

\frac{1}{T} = A+B\ln r+C(\ln r)^{3}, \qquad r=\frac{R}{1\,\Omega} \tag{3}

Three calibration points can determine the three coefficients, but they do not prove a universal interpolation-error limit between those points. Verify the model over the intended range. A manufacturer-provided lookup table with validated interpolation is another suitable approach, particularly when the published curve is more accurate than a single beta approximation. [2]

Design and selection checklist

Compare the complete sensing assembly and its interface, not just the element’s nominal resistance. The following checklist applies to low-power temperature sensing; power NTCs and switching PTC protection require separate energy, current and fault-duty checks.

Temperature and curve: minimum and maximum measured temperature, resistance at both endpoints, R₂₅, beta definition, full resistance–temperature data and model validity.
Accuracy: resistance tolerance, curve tolerance or interchangeability, coefficient precision, calibration requirements and long-term drift under relevant conditions.
Thermal behavior: dissipation factor, thermal time constant, stated test medium, mounting method and allowable self-heating.
Electrical interface: maximum measuring load, ADC input and reference limits, leakage, source impedance, filter settling and lead/contact resistance.
Mechanical and environmental performance: dimensions, lead strain relief, soldering process, vibration, humidity, condensation, chemical exposure and permitted immersion.
Catalog versus custom assembly: use a catalog component where its thermal contact, insulation, cable and sealing meet the need. Specify a custom probe when geometry or environmental demands require it, including acceptance tests and change-control requirements.
Qualification and safety: obtain evidence for the exact part and assembly required by the application. A qualification label or insulation test voltage does not alone establish the complete product’s safety or continuous working-voltage capability.
Supply continuity: control the curve identifier, coefficient revision, tolerance grade, cable, connector and manufacturing site where relevant; validate alternatives before approving substitution.

Current and Voltage Excitation

Excitation must produce a measurable signal while keeping the thermistor and electronics within their limits. Current excitation provides direct control of measurement current; voltage excitation with a series resistor often accommodates a large NTC resistance span more simply.

Current excitation and reference-resistor sizing

Figure 5 shows the current-excited arrangement. If the same excitation current I flows through thermistor resistance R_TH and reference resistance R_REF, their voltages are V_TH and V_REF. Their ratio in Eq. (4) removes the common current scale factor in the ideal circuit. [1]

Original schematic illustrating current excitation of a thermistor and precision reference resistor — Figure 5. Current excitation of a thermistor with a resistive reference.

\frac{V_{\mathrm{TH}}}{V_{\mathrm{REF}}}=\frac{IR_{\mathrm{TH}}}{IR_{\mathrm{REF}}}=\frac{R_{\mathrm{TH}}}{R_{\mathrm{REF}}} \tag{4}

Reference-resistor sizing depends on the converter’s transfer function and gain. For an ADC whose positive differential full scale is V_REF/G, where G is gain, the ideal range condition is G × R_TH,max ≤ R_REF. Thus, R_REF ≥ R_TH,max is the unity-gain case, not a universal rule for every converter or circuit. Reference-voltage limits, input headroom and current-source compliance must also be satisfied. [4]

A large cold resistance may require a small excitation current, leaving a small voltage at the hot endpoint. Options include programmable gain, switched current or a different architecture. Dynamic gain or current is not automatically required: first determine whether a fixed setting already meets the required noise and range over the complete span.

The original example uses 441.117 kΩ at −50 °C and 50 µA excitation. The required thermistor voltage is 22.05585 V, or approximately 22.06 V. A low-voltage current source cannot sustain that operating point; it reaches its compliance limit and the intended current is no longer maintained. [3, 4]

The AD7124-4 Rev. E datasheet also lists a 0.1 µA excitation option, so 50 µA is not its smallest listed setting. The 0.1 µA option would require approximately 44.11 mV across 441.117 kΩ, but its suitability must still be checked against the hot-end signal, reference generation, noise, leakage and the applicable current-source specifications. It does not automatically make current excitation preferable. [4]

Voltage excitation with a series resistor

Figure 6 illustrates voltage excitation. A fixed series resistor limits current and forms a divider with the thermistor. The current changes automatically as thermistor resistance changes, avoiding the compliance problem of maintaining a fixed current through a very high cold resistance. [1, 3]

Original schematic illustrating voltage excitation of a thermistor with a precision series resistor — Figure 6. Voltage excitation of a thermistor through a series sense resistor.

Using the same source for divider excitation and the ADC reference makes the measurement ratiometric. With an appropriate input range, unity gain can cover the entire divider output without dynamic gain changes. However, this does not guarantee that every ADC input configuration can accept voltages near either rail, or that hot-end sensitivity will meet the required resolution. [3, 4]

For low nominal resistance, check the maximum divider current and sensor dissipation carefully. For high resistance, evaluate leakage and settling. Current excitation is not inherently immune to all cable interference, and voltage excitation is not restricted to sensors mounted beside the ADC: remote installations require an explicit assessment of wiring, shielding, filtering and fault exposure.

Thermistor resistance range and self-heating

Self-heating is the temperature rise produced by the measurement itself. In Eq. (5), ΔT_SH is the approximate steady-state rise above the surrounding medium, P is electrical power dissipated in the thermistor, and D is its dissipation factor in W/K. A rise in kelvins has the same numerical value as a rise in degrees Celsius. [2]

\Delta T_{\mathrm{SH}}\approx\frac{P}{D} \tag{5}

For voltage excitation V_EXC and series resistor R_S, Eq. (6) gives divider current I and thermistor power P. In these equations, R_S is the same fixed resistor described elsewhere as R_SENSE. These expressions assume the measurement input does not significantly load the divider. [2]

I=\frac{V_{\mathrm{EXC}}}{R_{\mathrm{S}}+R_{\mathrm{TH}}}, \qquad P=I^{2}R_{\mathrm{TH}}=\frac{V_{\mathrm{EXC}}^{2}R_{\mathrm{TH}}}{(R_{\mathrm{S}}+R_{\mathrm{TH}})^{2}} \tag{6}

Retaining the original worked example, let V_EXC = 3.3 V, R_S = 10 kΩ, R_TH = 10 kΩ near 25 °C and D = 1 mW/K. Eq. (6) gives I = 0.165 mA and P = 0.27225 mW; Eq. (5) then gives approximately 0.272 °C self-heating. The thermistor voltage is 1.65 V. These are first-order estimates using the stated resistance and dissipation factor.

A 0.272 °C bias may be acceptable in some monitoring applications but is already larger than a ±0.1 °C total-error target. Being below 1 °C is therefore not an adequate precision-design criterion. Allocate self-heating a specific part of the error budget and validate it in the installed medium. [2, 3]

For a fixed excitation voltage and fixed series resistance, thermistor power reaches its maximum when R_TH equals R_S. Eq. (7) gives that maximum. Consequently, the highest divider current at the hot endpoint does not necessarily produce the highest thermistor dissipation. [2]

P_{\max}=\frac{V_{\mathrm{EXC}}^{2}}{4R_{\mathrm{S}}}, \qquad R_{\mathrm{TH}}=R_{\mathrm{S}} \tag{7}

For the same 3.3 V and 10 kΩ divider, Eq. (7) gives 0.27225 mW. This confirms that the original 25 °C example is also the divider’s maximum thermistor-power point if the resistance range crosses 10 kΩ. The maximum temperature rise can occur elsewhere if the installed dissipation factor changes with temperature or operating conditions.

Lower excitation voltage, a different series resistor or intermittent excitation can reduce heating. For intermittent operation, allow excitation, filters and the ADC to settle before accepting data; average power alone does not establish the instantaneous thermal error. The thermal time constant and dissipation factor remain dependent on the sensor assembly and environment. [2, 4]

Importance of Sigma-Delta ADCs in Thermistor-Based Applications

Low-bandwidth sigma-delta ADCs suit thermistor channels because they combine high resolution with configurable digital filtering. Integrated buffers, references, gain stages and diagnostics can reduce external circuitry, but the selected operating configuration determines the actual performance.

The AD7124 examples include programmable gain, an internal reference, input and reference buffering, excitation functions, calibration modes and diagnostic checks. These features are useful architectural examples, not requirements that make one converter family mandatory. Select the converter from the application’s input range, noise, power, timing and diagnostic needs. [3, 4]

Resolution, noise and filtering

A 24-bit output word does not mean 24 noise-free bits, and noise-free resistance resolution does not equal absolute temperature accuracy. Compare input-referred noise at the intended gain, filter, output rate and power mode, then convert it into temperature resolution using the local sensor sensitivity. [3, 4]

Analog filtering is still required to address interference near the modulator sampling frequency and its harmonics. A simple RC network may be sufficient in a controlled design, but remote cables, switching-power environments and transient exposure can require additional filtering and protection. Digital filtering cannot correct input saturation or eliminate all out-of-band interference. [3]

Filter choice trades noise and mains rejection against settling and throughput. Where 50 Hz and 60 Hz rejection are required, select a filter and rate that explicitly provide it; do not assume all settings do. In a multiplexed system, use the settling time after a channel change rather than the headline single-channel output rate to calculate the available update rate. [3, 4]

Input buffering, headroom and reference loading

A high-impedance divider can interact with an ADC’s input bias currents or switched-capacitor input. Input buffers can reduce source-loading errors, but they introduce their own input-range constraints and do not make source resistance, leakage or settling irrelevant. [3, 4]

For example, the AD7124-4 Rev. E specifies buffered analog inputs at gain 1 between AVSS + 0.1 V and AVDD − 0.1 V. The differential signal can be within full scale while one input pin is outside this valid operating range. A ground-referenced diagram therefore does not automatically justify enabling every buffer on a single-supply implementation. [4]

Check the absolute voltage at each analog and reference pin, the differential range, buffer settings and common-mode restrictions across temperature and faults. Also check reference-output loading when it excites several dividers. Any added headroom resistor or bias network must be included in the resistance-reconstruction equation unless its effect is excluded by the actual sensing connections. [3, 4]

Thermistor Circuit Configuration—Ratiometric Configuration

A ratiometric channel derives the measured signal and ADC reference from the same excitation source. This suppresses a common excitation-scale factor, but the wiring and transfer equation must match the voltage actually measured.

Current-excited ratiometric measurement

Figure 7 shows the current-source form discussed in Eq. (4). The same current must flow through the thermistor and reference resistor for ideal cancellation. Leakage, separate current paths, current-source compliance and reference-input loading can invalidate that assumption. [1, 4]

Original thermistor ratiometric circuit using a constant current source and reference resistor — Figure 7. NTC thermistor ratiometric configuration using current excitation.

The 441.117 kΩ cold-end example illustrates why current-source compliance must be checked before selecting gain or reference resistance. Conversely, at the hot endpoint the reduced resistance can make the signal-to-noise requirement more demanding. Evaluate both ends of the range, not only the nominal 25 °C point.

Voltage-divider measurement and resistance reconstruction

Figure 8 shows the divider concept. Let V_TH be the voltage measured across the thermistor itself, V_EXC the voltage across the complete two-resistor divider, and x their ratio. Eq. (8) gives the ideal relationship and its inverse. [3]

Original NTC thermistor voltage-divider circuit with a fixed series resistor — Figure 8. NTC thermistor measurement using a voltage divider.

x=\frac{V_{\mathrm{TH}}}{V_{\mathrm{EXC}}}=\frac{R_{\mathrm{TH}}}{R_{\mathrm{S}}+R_{\mathrm{TH}}}, \qquad R_{\mathrm{TH}}=R_{\mathrm{S}}\frac{x}{1-x} \tag{8}

Eq. (8) shows why divider voltage is not directly proportional to thermistor resistance. Choosing R_S equal to R₂₅ gives x = 0.5 at 25 °C, but other values are valid. Choose the resistor to balance sensitivity in the critical temperature interval, endpoint headroom, loading and self-heating.

If the thermistor forms the upper divider leg and the ADC instead measures the midpoint relative to ground, the ADC measures the voltage across the lower fixed resistor, not V_TH. Calling the normalized midpoint voltage y, Eq. (9) applies. Reversing the divider orientation reverses the direction of the measured node’s temperature response.

y=\frac{V_{\mathrm{node}}}{V_{\mathrm{EXC}}}=\frac{R_{\mathrm{S}}}{R_{\mathrm{S}}+R_{\mathrm{TH}}}, \qquad R_{\mathrm{TH}}=R_{\mathrm{S}}\frac{1-y}{y} \tag{9}

Convert ADC code into the appropriate voltage ratio using the device’s specified unipolar or bipolar transfer function and programmed gain. A bipolar converter’s midscale code can represent zero differential voltage, not half the reference voltage. Reject saturated or invalid readings before dividing by x, y or their complements. [3, 4]

Original schematic showing a thermistor measurement using a shared excitation and ADC reference — Figure 9. Thermistor measurement with a shared excitation and ADC reference.

Figure 9 illustrates the common-source principle. When V_EXC is also the ADC reference, the ideal result depends on the resistor ratio rather than the absolute excitation voltage. Cancellation is incomplete if the signal and reference see different voltage drops, loading, filtering or timing. Changes in excitation also change thermistor dissipation, so ratiometric operation does not cancel the resulting thermal bias. [2, 3]

Multiple thermistors and shared components

Figure 10 shows multiple thermistors sharing an excitation source. Independent divider branches normally use an individual precision sense resistor per sensor. Their combined current must remain within the excitation source’s drive capability, including loading and startup behavior. [3]

Original multiple-thermistor analog input configuration with shared excitation and separate sensing branches — Figure 10. Multiple thermistor channels sharing excitation.

An external multiplexer or switches can allow a precision resistor to be shared, but switch on-resistance, leakage, charge injection and off-channel paths enter the design. Check whether switch resistance is inside the measured resistance path and whether its temperature dependence is significant. Allow the analog network and digital filter to settle after switching. [3]

The per-sensor update interval includes all enabled measurements, channel settling, diagnostics and any excitation startup delay. Do not equate the ADC’s single-channel output data rate with the update rate of every sensor in a scan. [3, 4]

Open-circuit and short-circuit detection

Fault polarity depends on the divider orientation and measurement connection. With a fixed resistor connected to excitation and the thermistor connected to ground, an open thermistor drives the midpoint toward excitation and a short drives it toward ground. With the thermistor above the midpoint and a fixed resistor below, these midpoint indications reverse.

A differential measurement across the thermistor must be interpreted according to that connection rather than copied from a ground-referenced divider rule. Leakage, protection components, disconnected cables and a failed fixed resistor can also alter fault voltages. Choose thresholds outside the valid temperature range with allowance for tolerances, then test actual open, short and wiring-fault conditions.

Integrated burnout currents, reference checks, saturation flags and interface CRC checks can support diagnostics. They do not automatically identify every fault uniquely, and diagnostic currents may disturb a normal high-impedance measurement. Follow the ADC’s required diagnostic sequence and settling behavior. [4]

Accuracy Budget, Calibration and Validation

Temperature accuracy belongs to the complete channel, not to the ADC or sensor alone. Build the error budget over the required temperature span and electronics operating conditions before choosing tighter components or a higher-resolution converter.

Include thermistor interchangeability or resistance–temperature tolerance, model residual, fixed-resistor initial tolerance and temperature coefficient, aging, lead/contact resistance, leakage, ADC offset and gain error, noise, self-heating and thermal installation error. The fixed resistor follows its own local temperature, which may differ from the remote sensor temperature. [2–4]

For a small equivalent resistance error ΔR, Eq. (10) estimates the corresponding temperature-error magnitude. Here α is the thermistor’s local relative temperature coefficient in K⁻¹, and dR/dT is the local slope in Ω/K. The beta expression for α is an approximation within that model. [2]

|\Delta T|\approx\frac{|\Delta R|}{|dR/dT|}=\frac{|\Delta R/R|}{|\alpha|}, \qquad \alpha=\frac{1}{R}\frac{dR}{dT}\approx-\frac{\beta}{T^{2}} \tag{10}

Use Eq. (10) for local error estimates, not as a replacement for full-range tolerance analysis. Keep worst-case limits distinct from root-sum-square estimates: statistical combination requires justified assumptions about distributions and independence. Thermal gradients and self-heating are often systematic offsets rather than random noise.

Internal ADC calibration addresses converter errors under the conditions and modes supported by the device. It does not calibrate thermistor interchangeability, resistor drift or mounting error. The AD7124-4 is factory gain-calibrated at gain 1 and does not support an additional internal full-scale calibration at that gain; use the current datasheet rather than applying a generic calibration sequence from an older example. [4]

System calibration can remove selected electrical errors, while temperature calibration of the complete assembly can address sensor and installation effects under controlled conditions. A single temperature point does not establish curve accuracy across the full range. Retain coefficient units, curve identity and calibration data with the production configuration.

Validate the assembled channel at representative temperatures and electronics conditions. Compare readings at reduced excitation to assess self-heating, verify response after temperature changes, test every multiplexed channel, and inject wiring faults. Safety-related overtemperature protection requires an appropriate fail-safe response and system-level assessment; ADC diagnostics alone do not establish a safe product. [4, 6]

Conclusion

NTC thermistors provide high sensitivity, compact size and economical temperature sensing when their nonlinear characteristic and thermal installation are handled explicitly. Select the exact resistance–temperature curve and assembly first, then design excitation, the fixed resistor and ADC configuration around the full resistance span and required uncertainty.

A ratiometric voltage divider is often a practical starting point, while current excitation remains useful where its compliance, reference and noise requirements can be met. Neither architecture removes the need to budget self-heating, resistor drift, leakage, model error and thermal gradients.

The final design should reconstruct resistance using the correct circuit equation, convert it with a validated sensor model, and reject invalid measurements. Calibration, fault injection and installed thermal testing complete the process; resolution or a nominal sensor accuracy alone cannot demonstrate system performance.

FAQ

These answers summarize the selection and circuit-design decisions that most often determine thermistor measurement performance.

What are the advantages of an NTC thermistor compared with an RTD or switching PTC?

An NTC thermistor offers high resistance sensitivity, small sensing elements and economical interfaces over a suitable temperature interval. An RTD may be preferable for a wider span or established interchangeability requirements. Switching PTCs primarily serve threshold or protection functions. Response speed and accuracy depend on the exact sensor assembly, calibration and installation rather than the technology name alone.

Are all 10 kΩ NTC thermistors interchangeable?

No. The 10 kΩ designation normally identifies resistance at 25 °C, not the complete resistance–temperature characteristic. Substitutes must also match the required curve, beta definition or coefficients, tolerance envelope, operating range, package and environmental performance. Purchasing specifications should preserve the curve identifier and accuracy definition, and alternative parts should be validated in the actual measurement circuit.

Must the fixed divider resistor equal the thermistor resistance at 25 °C?

No. Equal resistances produce half the excitation voltage across each divider element at that temperature, which is a useful starting point. The best fixed resistance depends on the critical temperature interval, endpoint voltages, ADC limits, leakage and self-heating budget. Calculate the complete range and use the resistance-reconstruction equation that matches the measured voltage and divider orientation.

How can thermistor self-heating be estimated and reduced?

Estimate thermistor power from the actual excitation circuit, then divide it by the installed dissipation factor to estimate steady-state temperature rise. In a fixed-voltage divider, thermistor power peaks when thermistor and series resistance are equal. Lower excitation, a different resistor or intermittent operation can reduce heating, but settling and the final thermal mounting must also be validated.

What errors does ratiometric thermistor measurement cancel?

Ratiometric measurement suppresses the common excitation-scale factor when the sensor signal and ADC reference follow the same source. It does not cancel thermistor tolerance, fixed-resistor drift, lead resistance, leakage, ADC errors or thermal gradients. Excitation changes can still alter self-heating, and unequal filtering, loading or wiring drops can prevent ideal cancellation of source variation.

Is a 24-bit sigma-delta ADC necessary for accurate thermistor sensing?

Not automatically. Select the converter from the required temperature resolution, input range, source impedance, filtering, update rate and power budget. A sigma-delta ADC can integrate useful buffers, gain, reference and diagnostics, but its output word length does not establish absolute accuracy. Sensor tolerance, self-heating and installation errors can dominate even when ADC noise is very small.

How to Design an NTC Thermistor Measurement Channel

Use this procedure to move from the measurement requirement to a validated sensing channel. Perform the calculations with the selected sensor’s actual curve and the ADC’s intended operating configuration.

Design and validate a low-power NTC thermistor temperature measurement channel.

Define the range, accuracy and environment
Specify the measured temperature range, total allowable error, response time, update rate and installation conditions. Identify whether the channel provides monitoring, control or safety-related protection, and determine the required response to invalid data.
Select the exact sensor and curve
Choose nominal resistance, resistance–temperature characteristic, tolerance and package. Obtain endpoint resistances, model coefficients or a lookup table, dissipation factor, thermal response and environmental ratings. Confirm that any proposed substitute meets the complete specification.
Choose excitation and calculate self-heating
Compare current excitation with a voltage divider. Calculate endpoint voltages, current-source compliance where applicable, thermistor power and temperature rise. For a divider, also check the maximum-power point where thermistor and fixed resistance are equal.
Configure the ratiometric circuit and ADC
Derive excitation and ADC reference from the same source where practical. Check each input pin voltage, differential range, buffer headroom, reference loading and source impedance. Choose analog filtering and digital filter settings for the required noise and timing.
Implement resistance conversion and diagnostics
Convert ADC code using the correct gain and coding format, reconstruct resistance for the actual measured voltage, and apply the validated temperature model. Add saturation checks and topology-specific open and short thresholds. Allow settling after excitation or channel changes.
Build the error budget and calibration plan
Include sensor tolerance, model residual, fixed-resistor drift, wiring, leakage, ADC errors, noise, self-heating and thermal mounting. Use only calibration modes supported by the ADC configuration, and separate converter calibration from temperature calibration of the complete assembly.
Validate the installed assembly and release controls
Test representative sensor and electronics temperatures, reduced excitation, thermal response, multiplexed timing and injected wiring faults. Confirm the fail-safe response and environmental suitability. Record the approved curve, coefficients, component grades and assembly details for production and purchasing.