LM35 temperature sensor: pinout, grades, accuracy and the ADC trap

The letters after LM35 are the accuracy grade and the package. The grades are A and CA (the most accurate), plain and C, and D (the cheapest). A Z at the end is the TO-92, M the 8-pin SOIC and T the TO-220; an H is the TO-46 metal can. The TO-92 you find in shops is an LM35DZ, marked LM35 over DZ on two lines; the C and CA grades come in the same package as LM35CZ and LM35CAZ.

Only the C, CA and D grades are made in TO-92. The plain LM35 and the LM35A, the parts rated down to −55 °C, come only in the metal can, where pin 1 is +VS, pin 2 VOUT, pin 3 ground and the case is ground; match the leads to TI's drawing on page 3, as this page does not draw the can. The SOIC (LM35DM) and the TO-220 (LM35DT) exist only in the D grade, and the TO-220 puts ground in the middle.

The supply may go from −0.2 to 35 V and the output from −1 to 6 V, with 10 mA out of the output pin at most. TI recommends 4 to 30 V on the supply. On the bench the −0.2 V figure matters most: an LM35 wired in backwards puts the full supply the wrong way across it.

The thermal resistance is not a rating to stay under so much as the number that sets self-heating, below. The TO-92 in still air is 180 °C/W; the TO-220 is 90 °C/W, and 26 °C/W in moving air.

TI specifies accuracy as the error between the output and exactly 10 mV/°C times the case temperature. An LM35DZ may be 1.5 °C out at 25 °C and 2 °C out at 0 or 100 °C; the LM35CAZ in the same package is within 0.5 °C at 25 °C. Typical figures are under half the limits, so most parts read better than their grade, but only the limit is promised.

The grades also differ in range. With the output simply loaded to ground, the basic circuit reads from 2 °C up; the output can sink only about 1 µA, so it cannot follow the temperature further down. For readings below zero, TI's full-range circuit adds a resistor from the output to a negative supply, sized at −VS / 50 µA, and the output then goes to −550 mV at −55 °C.

The LM35 measures its own die, and the die warms by the power it dissipates times the thermal resistance. On 5 V with no load that is 5 V × 56 µA = 280 µW, or 0.05 °C in a TO-92 in still air, and 0.072 °C with the 80 µA maximum current. TI's own claim is under 0.1 °C.

Two things spoil that. A higher supply: at 30 V the same 56.2 µA heats it by 0.3 °C. And a load on the output: 1 mA drawn from it adds (5 − 0.25) V × 1 mA, and the rise becomes 0.93 °C, more than an LM35CA's whole error at 25 °C. Feed it into a high-impedance input and run it from 5 V.

The leads matter too. In the TO-92 the copper leads are the main heat path, so the die sits closer to the air temperature than to the surface it is glued to. TI suggests a bead of epoxy over the leads and wires so they sit at the surface temperature.

• Room, enclosure and heatsink temperature on a microcontroller's ADC, from 2 to 100 °C with the D grade, without calibration or lookup tables.
• Thermostats and fan controllers with a comparator: 10 mV/°C is easy to set a threshold against. The comparator lesson covers the circuit.
• Remote sensing over a twisted pair, with the damper in the example below.

It is the wrong part below 2 °C without a negative supply, and for anything that needs better than about 0.5 °C without calibration. For high temperatures or a cheaper sensor with more work, see the thermistor.

The LM35DZ runs from the Uno's 5 V pin, its output goes to A0, and the room is at 25 °C. The ATmega328P's ADC turns a voltage into a code from 0 to 1023: code = V_IN × 1024 / V_REF.

On a short lead straight into A0, the LM35 and a 100 nF across its supply are all you need: TI says it drives 50 pF on its own. A cable is a heavier capacitive load than that, and TI's fix is the damper in Figure 4. It carries no DC, so it does not change the reading.

The ADC itself is good to about 2 codes, Microchip's typical absolute accuracy, so averaging a few dozen readings helps with noise but cannot beat the sensor's grade. The ADCs and DACs lesson explains the steps, and the ATmega328P page covers the chip.

• Put 100 nF from +VS to ground right at the sensor. TI recommends it in noisy places, and a long supply lead is always a noisy place.
• Throw away the first reading after switching the ADC reference. Microchip says it may be wrong; the AREF pin also needs a moment to settle.
• Glue the sensor to what you are measuring and cover the leads, or it reads somewhere between the surface and the air.
• Need 0.1 °C resolution on a 5 V reference? Amplify the output: a non-inverting amplifier with a gain of 4 maps 0 to 100 °C onto 0 to 4 V.
• Keep it dry. TI warns that moisture corrodes the sensor and its leads, especially where condensation forms in the cold; coat the board and the leads.

The true equivalents are the better LM35 grades in the same TO-92. Other analog temperature sensors use a different scale or an offset, so they are not drop-in parts even when the pins match.

1. Power it from 5 V and put the meter between VOUT and ground on the 2 V or 200 mV DC range. At room temperature it should read about 10 mV per degree: 250 to 300 mV in a 25 to 30 °C room.
2. Pinch it between finger and thumb. The reading should climb by several tens of millivolts over half a minute and fall back when you let go.
3. Measure the supply current: about 56 µA, and no more than 80 µA for a D grade at 5 V. A part that reads near the supply voltage or draws milliamps is wired wrong or damaged.

Reading datasheets shows how to find these test conditions in TI's tables.

Buy the LM35DZ for a 0 to 100 °C job and the LM35CAZ when you cannot calibrate and need half a degree. Run it from 5 V, never load its output, and on an Arduino read it against the internal 1.1 V reference after measuring that reference once at the AREF pin.

Ask for the LM35 by its full part number (LM35DZ, LM35CZ or LM35CAZ): a bag labelled only LM35 does not tell you which grade you are getting. Check every new sensor against a known thermometer with the meter test above before trusting it. Our guide to buying electronic components in India compares ten online sellers and covers how to spot fakes.