The LM35 is a three-terminal precision analog temperature sensor. It produces an output voltage directly proportional to Celsius temperature at a nominal 10 mV/°C, with no offset term: 250 mV at 25°C, 1000 mV at 100°C. Because the scale is Celsius rather than Kelvin, no large constant needs subtracting before the reading is usable.
The device is trimmed and calibrated at the wafer level, so it needs no external calibration to reach typical accuracy of ±0.25°C at room temperature and ±0.75°C over the full rated span. Quiescent current is 56 µA typical, which keeps self-heating below 0.1°C in still air. Output impedance is 0.1 Ω at a 1 mA load, so a following ADC or comparator sees a stiff source.
The specified temperature range is a property of the grade suffix, not of the part number as a whole. −55°C to 150°C applies to the LM35 and LM35A grades, −40°C to 110°C to the LM35C and LM35CA, and 0°C to 100°C to the LM35D — and each grade is sold in a limited set of packages, so the package on the board also fixes the range the design can claim. The grade table and the package matrix later in this article carry the detail.
What separates the LM35 from a thermistor is linearity rather than accuracy: a thermistor needs a lookup table or a Steinhart-Hart fit, the LM35 needs a multiply. What separates it from a digital sensor is that it hands you a voltage, so the resolution of the measurement is set by your converter and not by the sensor.
LM35 Pinout
Pin names are consistent across all four packages: a positive supply pin, an analog output, and ground. Pin numbers are not.
| Signal | TO-CAN (TO-46) | TO-92 | TO-220 | SOIC-8 |
|---|---|---|---|---|
| +VS | 1 | 1 | 1 | 8 |
| VOUT | 2 | 2 | 3 | 1 |
| GND | 3 | 3 | 2 | 4 |
On the TO-220 part, the middle pin is ground and the output moved to pin 3. TI flags this in the datasheet with a note that the LM35DT pinout also differs from the discontinued LM35DP, so a legacy board designed around the older TO-220 part will not accept the current one without rework.
Two orientation traps apply on top of the pin-number differences.
The first is view direction. The datasheet draws the TO-92 package bottom view, while the other three are drawn top view. Reading the TO-92 figure as though it were a front view reverses the left-to-right order and swaps supply with ground. In the figure above the TO-92 is drawn the way you hold the part: flat face toward you, leads pointing down, giving +VS, VOUT, GND from left to right.
The second is that TI has changed its own view convention. The revision history records the TO-CAN pinout moving from top view to bottom view at revision F, then back to top view at revision H, which also added the pin numbers. Cached PDFs of revisions F and G are still in wide circulation. For the metal-can part, take lead-to-tab orientation from the current mechanical drawing rather than from any pinout graphic you already have on disk.
LM35 CAD Models and Land Pattern
The TO-92 die ships two ways, and the footprint is not the same for both.
Straight-lead parts are supplied in bulk only, and their three holes sit in a single row on 2.54 mm centres. Formed-lead parts are supplied on tape and reel or in ammo pack, and the centre lead is bent so that the middle pad steps 1.27 mm off the line of the outer two. The two layouts also call for slightly different hole diameters, both under a millimetre; take the exact figures off the drawing rather than off a footprint library.
That maps onto order codes directly. An LM35DZ/NOPB is a bulk part and wants the in-line footprint. An LM35DZ/LFT1 or LM35DZ/LFT4 is reeled or ammo-packed and wants the offset footprint. Ordering the reeled version to save on handling, against a board laid out for the bulk version, is a mechanical problem discovered at assembly rather than at design review. TI last revised this package drawing in April 2025, so footprint libraries built from older copies deserve a check.
LM35 Datasheet
The current revision is SNIS159H, dated December 2017, with the package and orderable addendum refreshed on 15 July 2026.
Download: LM35 Precision Centigrade Temperature Sensors, SNIS159H — Texas Instruments, PDF
That link resolves to whichever revision is current, so it will not go stale as TI republishes.
Technical Details
| Parameter | Value |
|---|---|
| Sensor gain, nominal | 10 mV/°C |
| Sensor gain, tested limits | 9.9 to 10.1 mV/°C (A, CA grades) · 9.8 to 10.2 mV/°C (LM35, C, D) |
| Supply voltage, recommended | 4 V to 30 V |
| Supply voltage, absolute maximum | −0.2 V to 35 V |
| Output voltage, absolute maximum rating | −1 V to 6 V |
| Output voltage, actual span | −550 mV at −55°C to 1500 mV at 150°C |
| Output current, absolute maximum | 10 mA |
| Quiescent current at 5 V, 25°C | 56 µA typical, 67 µA limit (A, CA) · 80 µA limit (LM35, C, D) |
| Nonlinearity | ±0.25°C typical |
| Output impedance | 0.1 Ω at 1 mA load |
| Self-heating in still air | 0.08°C |
| Long-term stability at TMAX, 1000 h | ±0.08°C |
| ESD, human body model | ±2500 V |
| Maximum junction temperature | 150°C |
| Storage temperature | −60°C to 150°C (TO-CAN, TO-92) · −65°C to 150°C (TO-220, SOIC) |
The two output-voltage rows describe different things and are frequently conflated. −1 V to 6 V is an absolute maximum rating: how far the pin may be driven from outside before damage becomes likely. It says nothing about what the device produces. The output itself never leaves the −550 mV to 1500 mV window, and on a single supply it never goes below roughly 20 mV.
Thermal resistance varies by more than a factor of four across the package range, which matters whenever the part carries load current:
| Package | RθJA, still air | RθJA, moving air |
|---|---|---|
| TO-CAN, 3 pin | 400 °C/W | 100 °C/W |
| TO-92, 3 pin | 180 °C/W | 90 °C/W |
| SOIC-8 | 220 °C/W | 105 °C/W |
| TO-220, 3 pin | 90 °C/W | 26 °C/W |
At 56 µA and 5 V the internal dissipation is 280 µW, so even 400 °C/W produces about 0.11°C of rise. Draw a milliamp from the output and the arithmetic changes: 5 mW into a TO-CAN in still air is 2°C of self-heating, which swamps the accuracy specification entirely.
LM35 Characteristics by Grade
Specified range and guaranteed accuracy are set per grade:
| Grade | Specified range | Accuracy at 25°C, typ / limit | Accuracy at TMAX, typ / limit | Accuracy at TMIN, typ / limit |
|---|---|---|---|---|
| LM35A | −55°C to 150°C | ±0.2 / ±0.5 | ±0.4 / ±1.0 | ±0.4 / ±1.0 |
| LM35 | −55°C to 150°C | ±0.4 / ±1.0 | ±0.8 / ±1.5 | ±0.8 / ±1.5 |
| LM35CA | −40°C to 110°C | ±0.2 / ±0.5 | ±0.4 / ±1.0 | ±0.4 / ±1.5 |
| LM35C | −40°C to 110°C | ±0.4 / ±1.0 | ±0.8 / ±1.5 | ±0.8 / ±2.0 |
| LM35D | 0°C to 100°C | ±0.6 / ±1.5 | ±0.9 / ±2.0 | ±0.9 / ±2.0 |
All figures in °C. The typical column is what a part will usually do; the limit column is what TI guarantees.
Now cross that against the packages each grade is available in. The −55°C grades, LM35 and LM35A, exist only as LM35H and LM35AH in the hermetic metal can. There is no LM35A in TO-92, no LM35 in TO-220, and no −55°C grade in surface mount at all. The plastic TO-92 offers LM35CZ, LM35CAZ and LM35DZ. The TO-220 offers LM35DT and the SOIC-8 offers LM35DMX, both D grade only.
The practical consequence is that the LM35DZ sitting in most parts drawers, and in nearly every sensor kit, is specified from 0°C to 100°C with a ±1.5°C guaranteed limit at room temperature. It is not a −55°C part, it is not a ±0.5°C part, and no external circuitry changes that, because the difference is in what was tested at the wafer, not in how the device is wired.
Functional Block Diagram
The sensing element is a delta-V<sub>BE</sub> structure: two transistors run at a 10:1 emitter area ratio, so the difference in their base-emitter voltages is proportional to absolute temperature. A first amplifier develops that into a PTAT voltage of roughly 1.38 V, at about 8.8 mV/°C, and a second stage scales it to the 10 mV/°C the pin delivers.
The output amplifier at the far right is a simple class A stage with a typical 0.5 Ω output impedance. It sources current readily. It sinks approximately 1 µA.
That asymmetry is the root cause of most LM35 wiring surprises, and it explains two things at once: why the basic single-supply circuit is specified only from 2°C upward, and why reading below zero requires something external to pull the output pin down. Neither is a defect. Both follow from a topology chosen to keep quiescent current at 56 µA.
How to Use the LM35
The basic connection is three wires. Supply between 4 V and 20 V to +VS, ground to GND, and the output straight into an ADC input or a comparator. In noisy environments TI recommends a 0.1 µF capacitor from V+ to ground, and notes that larger values may be needed depending on how bad the supply is.
This circuit is specified from 2°C to 150°C, not from 0°C. The datasheet gives a minimum temperature for rate accuracy of 1.5°C typical with a 2°C limit, measured in exactly this configuration. Below that the class A output cannot hold the linear relationship, because there is nothing to sink the current the transfer function would require.
To measure below 2°C you need the full-range circuit on the right. A resistor from the output pin to a negative rail provides the sink path, sized as R1 = −VS / 50 µA. With a −5 V rail that is 100 kΩ. Once the pull-down is in place the output tracks properly across the whole grade range, reaching −550 mV at −55°C for the parts rated that low.
A negative output has a consequence downstream that is easy to miss. Almost no single-supply ADC will digitise a negative input, and applying one may exceed its own absolute maximum ratings. Sub-zero measurement therefore means either a level-shifting stage between sensor and converter, or a converter with a bipolar input range. The pull-down resistor solves the sensor problem, not the conversion problem.
On 3.3 V supplies: the recommended minimum is 4 V, and the datasheet includes a curve of minimum supply voltage against temperature showing the requirement rising with temperature. A 3.3 V rail is outside the specification at every temperature. Parts often appear to work at 3.3 V near room temperature; nothing about that behaviour is guaranteed, and it degrades as the device warms.
Temperature Measurement with the LM35
Converting the output is one multiply, but the resolution you get depends entirely on the reference you convert against.
On a 10-bit ADC running from a 5 V reference, one count is 4.88 mV, which is 0.488°C. The sensor’s typical nonlinearity is ±0.25°C, so the converter is the dominant error source by roughly a factor of two. Averaging reduces noise but cannot recover quantisation you never captured.
// LM35 on an AVR-based board, default 5 V reference
const uint8_t LM35_PIN = A0;
void setup() {
Serial.begin(9600);
}
void loop() {
uint16_t counts = 0;
for (uint8_t i = 0; i < 16; i++) {
counts += analogRead(LM35_PIN);
delay(2);
}
float millivolts = (counts / 16.0) * (5000.0 / 1024.0);
float celsius = millivolts / 10.0; // 10 mV per degree C
Serial.print(celsius, 2);
Serial.println(" C");
delay(1000);
}
Switching to the 1.1 V internal reference on an ATmega328P changes the arithmetic considerably. One count becomes 1.07 mV, or 0.107°C, and the sensor becomes the dominant error term as it should be. The trade-off is a ceiling: 1.1 V of range corresponds to about 110°C. The internal reference also carries a tolerance of roughly ±10% part to part, so the gain must be calibrated against a known temperature before the extra resolution translates into extra accuracy.
Note also that the 5 V “reference” on a USB-powered board is not a reference at all. It is the USB rail, and it moves with cable resistance and load. Measuring the actual rail with a meter and substituting that number for 5000.0 typically removes more error than any amount of software filtering.
How the LM35 Behaves in Real Circuits
A 56 µA micropower output stage has limits on what it can drive. Unaided, the LM35 handles 50 pF. Beyond that the output can ring or oscillate, and the usual symptom is a reading that is stable on the bench and unstable once the sensor moves onto a cable.
Two remedies appear in the datasheet. A series resistor, on the order of 2 kΩ, isolates the load capacitance from the output pin and works where the following stage is high impedance. Where capacitance must be driven, a series R-C damper from output to ground does the job: 75 Ω in series with 0.2 to 1 µF. A 0.01 µF bypass on the supply is often added alongside.
For remote sensing the datasheet gives a better arrangement than adding damping to a long run: the two-wire remote circuits of Figures 16, 17 and 19, built around a 200 Ω load resistor and a twisted pair. In that configuration the wiring capacitance forms a bypass from ground to the input rather than sitting on the output node, which is what makes the output relatively immune to cable length and why remote operation appears in the feature list at all. Take the exact wiring from those figures.
The remaining constraint is thermal, not electrical. Glue or cement the device to a surface and its temperature will be within about 0.01°C of that surface — provided the air matches. The copper leads of a TO-92 package are the principal thermal path into the die, so if the leads sit in air at a different temperature, the die settles somewhere between the two. The fix is a bead of epoxy over the leads and nearby wire, so that everything close to the sensor is at the surface temperature. A metal-can part can instead be soldered directly to a metal surface or pipe, at the cost of grounding the circuit’s negative terminal to that metal, since the case is tied to GND.
In a hostile electrical environment the wiring behaves as an antenna and the internal junctions as rectifiers, so relays, motors with arcing brushes and SCR transients will show up as offsets rather than as noise. Supply bypassing and the output damper both help; shortening the analog run helps more.
LM35 Sensor Uses
TI lists four application areas: power supplies, battery management, HVAC, and appliances. All four want a single local temperature in a system that already has a supply rail above 4 V.
Thermal shutdown and derating in a power supply is the cleanest fit, because the threshold can be a fixed comparison voltage and no conversion is needed at all. Battery pack monitoring uses the part the same way, with the advantage that a metal-can device can be bonded directly to a cell. HVAC and appliance control exploit the remote arrangement, with the sensor at the measurement point and the electronics elsewhere.
The LM35 fits badly in three cases. It is a poor choice for battery-powered products that sleep, because the output is always live and cannot be powered down without a supply switch. It is a poor choice for multi-point sensing, since every sensor needs its own ADC channel and its own analog run. And it is a poor choice where sub-zero measurement is the main requirement, because the negative rail and the level shift together cost more board area than a digital sensor would.
LM35 Package Overview and Orderable Variants
| Package | Body size, nominal | Grades offered | Order-code examples |
|---|---|---|---|
| TO-CAN (NDV), 3 pin | 4.699 × 4.699 mm | LM35, A, C, CA, D | LM35H/NOPB, LM35AH/NOPB, LM35CH/NOPB |
| TO-92 (LP), 3 pin | 4.30 × 4.30 mm | C, CA, D | LM35CZ/NOPB, LM35CAZ/NOPB, LM35DZ/NOPB |
| TO-220 (NEB), 3 pin | 14.986 × 10.16 mm | D | LM35DT/NOPB |
| SOIC-8 (D) | 4.90 × 3.91 mm | D | LM35DMX/NOPB |
Two procurement notes come out of the July 2026 addendum.
First, the non-RoHS versions of the two newer packages have moved to not-recommended-for-new-design status: LM35DMX and LM35DT are both NRND, while their /NOPB counterparts remain in active production. For a new board there is no reason to specify the leaded part number.
Second, the op-temp column of the addendum is internally inconsistent for the tape-and-reel TO-92 variants. LM35DZ/LFT1 and LM35DZ/LFT4 are listed as −55°C to 150°C, and LM35CZ/LFT1 and LM35CAZ/LFT4 as −55°C to 150°C, none of which agrees with the Recommended Operating Conditions table for those grades. LM35DH is listed as 0°C to 70°C where the same table says the D grade is specified to 100°C. Where the addendum and the specification table disagree, the specification table is the document that carries the tested and design limits. Read the grade letter, then read section 6.3; do not size a design from the addendum column.
Comparable Parts
None of the parts below substitutes directly for the LM35. The differences are structural rather than incremental.
LM34. The Fahrenheit sibling, 10 mV/°F on the same three-pin topology. Mechanically and electrically interchangeable, and wrong by a factor of 1.8 plus an offset if swapped in without changing the conversion. It is the one part on this list that will appear to work while reading nonsense.
TMP35 and TMP36. Two different parts, usually listed together. Analog Devices describes the TMP35 as functionally compatible with the LM35, with the same 10 mV/°C scale factor and the same 250 mV at 25°C, and the same TO-92 pin order. The TMP36 keeps the 10 mV/°C slope but adds a deliberate 500 mV offset, giving 750 mV at 25°C, which lets it report temperatures down to −40°C from a single supply with no negative rail and no pull-down resistor. It also has a shutdown function, which the LM35 does not. The two catches are that the offset must be subtracted in software and eats ADC range, and that both parts top out at a 5.5 V supply against the LM35’s 30 V, so neither is a candidate where the sensor runs off a 12 V or 24 V rail. The TMP35/TMP36/TMP37 datasheet carries the offset and supply specifications.
LM19 and LM335. Also analog, but neither shares the LM35’s transfer function. The LM19 has a negative, mildly nonlinear slope; the LM335 is a Kelvin-calibrated zener-style device needing a bias resistor. Both require the conversion to be rewritten.
DS18B20. Digital, one-wire, and a different design entirely. Resolution is set inside the part rather than by your converter, many devices share one bus, and no analog run is needed. It is the right answer for multi-point or long-distance sensing and the wrong answer where a bare voltage feeding a comparator is all that is wanted.
NTC thermistor. Cheaper, faster in small packages, and far more sensitive over a narrow band. It is also nonlinear, needs a bias network, and varies part to part, which means either a lookup table or individual calibration. The LM35 trades all of that away for a multiply.
For BOM substitution: only the LM34 is a mechanical drop-in, and it is a functional mismatch. Everything else on this list requires firmware changes, and the TMP36 and DS18B20 also require reconsidering the surrounding circuit.
Manufacturer Information
The LM35 originated at National Semiconductor and passed to Texas Instruments with the 2011 acquisition, which is why older datasheets in circulation carry National branding and National document numbers alongside the TI literature number. TI maintains the part in active production across all four packages in /NOPB form. There is no Q1 automotive-qualified LM35 and no radiation-hardened variant in the orderable list, so a design with either requirement needs a different part number rather than a different grade.
Parts marked LM35 from vendors other than TI exist. Because the entire value of the device is in wafer-level trimming and tested limits, and because grade suffixes are the only thing separating a ±0.5°C part from a ±1.5°C part, an untraceable LM35 is a part with no accuracy specification at all regardless of what is printed on the package.
Frequently Asked Questions
Is the LM35 output range −1 V to 6 V?
No. That is an absolute maximum rating describing how far the pin may be driven from outside before damage. The output the device produces spans −550 mV at −55°C to 1500 mV at 150°C, and stays above roughly 20 mV on a single supply.
Is the LM34 a drop-in replacement for the LM35?
Mechanically yes, functionally no. The LM34 is calibrated at 10 mV per degree Fahrenheit. Substituting one for the other without changing the conversion produces a plausible-looking reading that is wrong at every temperature except −40°, the single point where the two scales coincide.
Why do some LM35 datasheets state 10 mV/°F in the transfer function?
That was an error in the equation in revisions up to F, corrected at revision G. The same revision also corrected the Power Supply Recommendations section, which had said 4 V to 5.5 V rather than the correct 4 V to 30 V. If a PDF in your files shows either of those, it predates August 2016 and should be replaced.
Is the LM35 discontinued?
No. All /NOPB order codes across the TO-CAN, TO-92, TO-220 and SOIC-8 packages are in active production as of the July 2026 addendum. The leaded LM35DMX and LM35DT are NRND, and the TO-220 LM35DP was discontinued long ago with a different pinout from the LM35DT that replaced it.