When managing heat in embedded systems, the direct answer for sensor selection is straightforward: for monitoring junction or case temperatures under 150°C on a PCB, digital IC sensors (like the TI TMP117) or 10kΩ NTC thermistors are the default picks. Thermocouples are strictly overkill unless you are monitoring >200°C exhaust manifolds or high-power MOSFET tabs in industrial drives. However, picking the sensor is only 10% of the battle. The real engineering work lies in calculating the thermal path, sizing the heatsink based on actual wattage, and placing the sensor where it can actually trigger a shutdown before the silicon cooks itself.
The Thermal Path: Junction-to-Ambient Math (RθJA)
Before you can select a sensor, you must know what temperature you are trying to measure and where. Thermal resistance is modeled exactly like electrical resistance, where temperature is voltage, power dissipation is current, and thermal resistance (Rθ, measured in °C/W) is the resistor.
Let’s run a real-world calculation for a TO-220 packaged LM317 linear regulator dropping 12V to 5V at 1A. The power dissipated ($P_D$) is 7W. The datasheet lists the Junction-to-Ambient thermal resistance ($R\theta_{JA}$) with no heatsink at roughly 50°C/W. If your enclosure ambient ($T_A$) is 40°C:
$T_J = T_A + (P_D \times R\theta_{JA})$
$T_J = 40 + (7 \times 50) = 390°C$
The silicon will vaporize long before hitting 390°C. The internal thermal shutdown will trip at ~165°C, but relying on a protection circuit to act as your primary cooling strategy guarantees a rebooting, unstable system. You need a heatsink.
The total thermal path with a heatsink is the sum of its segments: Junction-to-Case ($R\theta_{JC}$), Case-to-Sink ($R\theta_{CS}$), and Sink-to-Ambient ($R\theta_{SA}$).
- $R\theta_{JC}$: ~2.0°C/W (from the LM317 datasheet).
- $R\theta_{CS}$: ~0.5°C/W (using a standard 0.5mm silicone thermal pad).
- $R\theta_{SA}$: The heatsink value we need to calculate.
We want to keep the junction ($T_J$) at or below 110°C, leaving a 15°C safety margin below the 125°C maximum rating. Using the same 40°C enclosure ambient:
$110 = 40 + 7 \times (2.0 + 0.5 + R\theta_{SA})$
$70 = 7 \times (2.5 + R\theta_{SA})$
$10 = 2.5 + R\theta_{SA}$
$R\theta_{SA} = 7.5°C/W$
You must select a heatsink rated at 7.5°C/W or lower. Furthermore, when reading the power derating curve in the component datasheet, note that above 25°C ambient, the maximum allowable power drops linearly. If the curve shows derating to 0W at 150°C, every degree your heatsink fails to shed pushes you closer to that cliff.
Matching Types of Temperature Sensor to Your Thermal Zone
Once the thermal budget is established, you must choose from the available types of temperature sensor to monitor the critical nodes. The right sensor depends entirely on the thermal zone (junction, case, or ambient) and the required interface.
| Sensor Type | Example Part Number | Accuracy & Range | Interface | Best Thermal Zone |
|---|---|---|---|---|
| Digital IC | TI TMP117 | ±0.1°C (-20 to 50°C) | I2C | Board ambient, precise enclosure monitoring |
| NTC Thermistor | Murata NCP18XH103F03RB (10kΩ 3950) | ±1% (0 to 50°C) | Analog (ADC) | Heatsink case temp, battery pack surface |
| RTD (PT100) | TE Connectivity PT100 Class A | ±0.15°C at 0°C | Analog (Bridge/ADC) | Industrial motor windings, high-noise environments |
| Thermocouple | Omega Type K (KQSS-18U) | ±2.2°C or 0.75% | Analog (Cold Junction IC) | Exhaust, >200°C power tabs, kiln interiors |
Failure Signatures: How Hot is Too Hot for Embedded Silicon?
How hot is too hot? For commercial-grade silicon like the ESP32 or standard CMOS logic, the absolute maximum junction temperature ($T_J$) is typically 125°C. However, the maximum rated ambient temperature ($T_A$) is usually 85°C. If your enclosure ambient hits 85°C, your junction will easily exceed 125°C under any real load.
Pushing silicon to its thermal limits doesn't always result in immediate catastrophic failure; it often manifests as insidious degradation:
- Timing Skew and Protocol Drops: Propagation delays in CMOS gates increase with temperature. An I2C or SPI bus running at 1MHz might throw CRC errors or drop acknowledgments at 100°C because the setup/hold times are violated as the silicon slows down.
- Electromigration: High current density combined with high temperatures causes metal atoms in the IC's internal traces to physically migrate. This increases the $R_{DS(on)}$ of internal MOSFETs over months, eventually leading to an open circuit.
- Capacitor Dielectric Breakdown: The Arrhenius equation dictates that the lifespan of electrolytic and polymer capacitors halves for every 10°C increase in operating temperature. A capacitor rated for 2,000 hours at 105°C will dry out and fail in just 250 hours if run at 135°C.
Heatsink Selection and Airflow: What Buys You Headroom?
Returning to our requirement for a 7.5°C/W heatsink, a standard choice is the Aavid Thermalloy 531202B02500G, which provides roughly 12°C/W in natural convection. This is insufficient for our 7.5°C/W target. You have two physical levers to pull to buy headroom:
- Upgrade the Extrusion: Move to a larger finned profile, like the Aavid 577102B04000G, which drops natural convection $R\theta_{SA}$ to about 6.5°C/W. This solves the problem passively but requires significant PCB real estate.
- Add Forced Airflow: Keep the smaller 12°C/W heatsink but add a 25mm fan (e.g., Sunon MF25101V2 pushing 2.5 CFM). Forced air at just 1 meter per second typically cuts a heatsink's thermal resistance by 50% to 70%. The 12°C/W sink drops to ~5.0°C/W, easily clearing the 7.5°C/W hurdle.
The Enclosure Penalty: Airflow is useless if the hot air is trapped. Sealing a project in an IP65 polycarbonate enclosure adds a massive thermal resistance layer to the system boundary. The enclosure itself acts as a final "heatsink" to the outside room. If you must use a sealed enclosure, you must thermally bond your internal heatsink to the enclosure wall using a thermal pad, effectively using the entire plastic/metal box as your $R\theta_{SA}$.
Decision Tree: Which Sensor and Cooling Strategy to Pick
Use this decision matrix to terminate your design process with concrete component selections based on your calculated power dissipation.
| Power Dissipation ($P_D$) | Thermal Environment | Cooling Strategy Pick | Sensor Pick & Placement |
|---|---|---|---|
| < 1.5W | Open PCB or vented enclosure | No heatsink. Rely on PCB copper pour (2oz, 1 sq in) as a spreader. | TI TMP117: Place I2C sensor near the component to monitor local board ambient. |
| 1.5W to 5W | Standard TO-220 / D2PAK package | Passive extruded aluminum (e.g., Aavid 531202B02500G) + silicone thermal pad. | 10kΩ NTC Thermistor: Epoxied directly to the metal tab of the component. |
| 5W to 15W | Sealed IP65 enclosure or high ambient (>50°C) | Active cooling. Extruded sink + 25mm/40mm 5V DC fan (e.g., Sunon MagLev). | 10kΩ NTC on case + TMP117 for enclosure ambient. Use MCU to PWM the fan based on the delta between the two. |
| > 15W | Industrial drives, high-power motor controllers | Liquid cold plate or massive fin-stack with chassis mounting. | Type K Thermocouple: Bolted to the IGBT/MOSFET baseplate, read via MAX31855 amplifier. |
For the vast majority of embedded makers and IoT engineers building ESP32 or Raspberry Pi power stages in the 2W to 5W range, the concrete default is the Murata 10kΩ 3950 NTC thermistor bonded to the case, paired with an Aavid passive extruded heatsink. It provides the fastest thermal response time, requires only a simple voltage divider and a 12-bit ADC to read, and costs pennies compared to digital alternatives. Calculate your RθSA, size the aluminum accordingly, and let the math dictate the shutdown threshold.
