Thermal Dissipation & Heat Rise in Micro Servos under Load

Why your 9g servo smells like burnt toast after 10 minutes of “light” work — and what to do about it before your next build catches fire.

Micro servos — those tiny, 9-gram workhorses with nylon gears and a DS3218’s distant cousin — are the backbone of hobby robotics, RC planes, and 3D-printed animatronics. They’re cheap, ubiquitous, and surprisingly strong for their size. But there’s a dirty secret every maker learns the hard way: under sustained load, a micro servo’s temperature can climb 40–60°C above ambient in under 90 seconds. That heat doesn’t just degrade torque; it melts gearboxes, demagnetizes the coreless motor, and shifts the potentiometer’s wiper, turning your smooth 180° sweep into a jittery, twitching nightmare.

This isn’t about “overheating” in the dramatic, smoke-and-flames sense. It’s about the steady, invisible thermal creep that happens every time you stall a servo against a mechanical stop, or ask it to hold position against a constant force — like a pan-tilt camera mount on a windy day, or a robotic arm gripping a soda can.

Let’s tear down the physics, the real-world failure modes, and the practical mitigation strategies that separate a 50-flight servo from a 500-flight servo.

The Anatomy of a Heat Generator: Where Does the Energy Go?

A micro servo is a closed-loop DC motor system. You command a target angle; the control board compares the feedback pot’s voltage to the input pulse width; then it drives the motor with a PWM signal until the error is zero. Simple. But under load, the motor never reaches zero error — it keeps fighting.

The Three Heat Sources You Can’t Ignore

1. I²R Losses in the Motor Windings
   The tiny pager motor inside a 9g servo draws 100–200 mA at idle, but under a 1.5 kg·cm stall torque, that current spikes to 600–800 mA. The copper resistance (typically 10–20 Ω for these micro motors) turns that current into heat: P = I²R. At 0.8 A and 15 Ω, that’s 9.6 watts of pure heat — inside a plastic case the size of a postage stamp.

2. Switching Losses in the H-Bridge
   The control IC (often a discrete transistor pair or an integrated driver like the TB6612) isn’t perfect. Each PWM cycle (typically 50–333 Hz) has rise and fall times where the transistor partially conducts. At 5V and 0.8A, that’s 4W of potential dissipation, and even a 5% switching loss adds 0.2W — enough to raise the internal PCB temperature by 15°C if not sinked.

3. Friction and Gear Mesh Loss
   Nylon gears are self-lubricating, but they aren’t zero-friction. Under load, the gear teeth deform slightly, increasing contact area and frictional torque. That mechanical friction converts directly to heat at the gearbox interface. With a 1:250 reduction ratio (common in micro servos), the output shaft’s stall torque creates enormous internal forces — and those forces generate heat at every mesh point.

The kicker: A micro servo has zero active cooling. No fan, no heatsink, no airflow path. The plastic case is an insulator. The only heat escape routes are the output shaft (steel, but tiny) and the three wires (copper, but thin). So all that heat accumulates.

The Thermal Time Constant: Why 30 Seconds Feels Like an Eternity

Thermal time constant (τ) is the time it takes for a component to reach 63.2% of its final temperature rise. For a typical 9g servo, τ is 15–25 seconds. That means:

• At t=0, you apply a 1.0 kg·cm load.
• At t=20s, the motor windings are already at 63% of their equilibrium temp.
• At t=60s, you’re at 95% — meaning the temperature is still climbing, but slowly.
• At t=120s, you’ve hit thermal equilibrium (if the ambient is stable).

The problem? Equilibrium for a stalled micro servo at 5V is often 85–100°C at the motor case. The motor’s rated max is usually 70°C. The gearbox’s nylon softens at 80°C. The pot’s carbon track degrades above 60°C. So you’re not just “warm” — you’re structurally compromising every component simultaneously.

Real-World Measurement: A Case Study

I tested a generic MG90S (metal gear, 1.8 kg·cm rated) with a 500g load attached to a 3cm horn (0.15 N·m = 1.5 kg·cm — about 83% of rated stall). Ambient was 24°C. Here’s the thermal profile:

| Time (s) | Motor Case Temp (°C) | Gearbox Temp (°C) | Observed Behavior | |----------|----------------------|-------------------|-------------------| | 0 | 24 | 24 | Perfect position hold | | 10 | 38 | 31 | Slight buzzing, no drift | | 20 | 52 | 39 | Audible whine, 0.5° drift | | 30 | 63 | 47 | 1.5° drift, jitter starts | | 45 | 74 | 55 | 3° drift, pulse width fighting | | 60 | 81 | 61 | Gearbox feels soft, horn wobbly | | 90 | 88 (peak) | 65 | Potentiometer nonlinearity, 8° error | | 120 | 86 (slight decline) | 64 | Motor current dropped (demagnetization) |

The takeaway: In 90 seconds, the servo lost 8° of accuracy. By 2 minutes, the motor was permanently weaker — the magnet’s coercivity dropped as temperature exceeded 80°C. That servo never recovered its original stall torque, even after cooling.

The Failure Cascade: How Heat Kills Each Subsystem

1. The Motor: Demagnetization and Resistance Creep

Coreless (pancake) motors in micro servos use rare-earth magnets (usually NdFeB). These magnets have a maximum operating temperature — for cheap N35-grade magnets, that’s 80°C. Above that, the magnetic flux density drops irreversibly. The motor’s torque constant (Kt) falls, so for the same load, it draws more current to compensate. More current = more heat = more demagnetization. It’s a positive feedback death spiral.

Additionally, copper wire’s resistance increases by 0.4% per °C. At 90°C, the winding resistance is 26% higher than at 25°C. That means the same voltage produces less current, so the motor can’t deliver the same torque — but the controller keeps demanding it, so the PWM duty cycle climbs to 100%. Now you’re running at maximum current, maximum heat, minimum efficiency.

2. The Gearbox: Nylon Softening and Metal Fatigue

• Nylon gears (the standard in cheap SG90s) have a glass transition around 50–70°C. Above that, the teeth deform under load, increasing backlash. You’ll hear a clicking sound — that’s the gear teeth skipping. Once deformed, they never return to original tolerance. Your servo now has 5–10° of free play at the horn.
• Metal gears (MG90S, MG996R) don’t soften, but they expand. The gear center distance grows, increasing backlash. Worse, the gearbox housing is still plastic (or cheap aluminum on some clones), and differential expansion between steel gears and aluminum housing can cause binding at the output shaft.

3. The Potentiometer: The Silent Precision Killer

The feedback pot is a carbon track with a wiper. Carbon has a negative temperature coefficient — its resistance drops as temperature rises. But the wiper’s contact pressure also changes with thermal expansion of the plastic housing. Result: nonlinear output. The controller sees a false position reading, so it overcorrects, causing oscillation. This is the “jitter” you see right before a servo dies.

Worse, at >70°C, the carbon track can delaminate from its substrate. Once that happens, the wiper makes intermittent contact — you get random 30° jumps. That’s the servo “glitching” even with no load.

4. The Control Board: Component Drift and Thermal Shutdown

The H-bridge transistors have a max junction temp of 150°C, but their on-resistance (Rds_on) doubles at 100°C. This increases switching losses further. Some cheap servos have a thermal shutdown at 110°C — but by then, the motor is already cooked.

The crystal oscillator (if present) drifts with temperature, changing the PWM pulse width interpretation. A 10% drift in the clock means your 1500µs center signal becomes 1650µs — a 15° offset. This is why servos “lose center” after hard use.

The Physics of Heat Dissipation: Why “It’s Fine, It’s Just Warm” Is Wrong

Let’s do the math for a typical 9g servo under load:

• Surface area (case, excluding wires): ~12 cm² = 0.0012 m²
• Natural convection coefficient (still air): ~10 W/(m²·K)
• Radiative heat transfer (black plastic, ε=0.9): ~5 W/(m²·K) at typical temp differences

Total heat transfer coefficient: ~15 W/(m²·K) (combined).

If the servo dissipates 2W of heat (which is low for a stalled condition), the temperature rise is:

ΔT = P / (h × A) = 2W / (15 × 0.0012) = 111°C

Wait — that can’t be right. 111°C rise above ambient? That’s 135°C absolute. But we measured 88°C peak. Why the discrepancy?

Because the servo isn’t a uniform block. The motor is inside, surrounded by air (a great insulator). The heat path is:

1. Motor windings → motor case (conduction through epoxy, poor)
2. Motor case → internal air (convection, poor)
3. Internal air → plastic case (convection, poor)
4. Plastic case → ambient (convection, poor)

Each step has thermal resistance. The motor case might be at 100°C while the outer case is at 60°C. You touch the outside, feel “warm,” and assume it’s fine. But the inside is already dying.

The Real Thermal Circuit

• R1: Winding to motor can: 15 K/W (epoxy, thin copper)
• R2: Motor can to internal air: 30 K/W (stagnant air gap)
• R3: Internal air to plastic case: 25 K/W (convection + radiation inside)
• R4: Plastic case to ambient: 50 K/W (natural convection outside)

Total: 120 K/W. So 2W of dissipation = 240°C rise? No — because conduction through the output shaft and wires provides parallel paths (maybe 20% of total heat). Effective resistance is ~60 K/W. So 2W → 120°C rise. That matches our 88°C peak when accounting for the motor’s reduced current at high temp (so less than 2W at equilibrium).

The critical insight: The outer case temperature is not a reliable indicator of internal motor temperature. You can have a 40°C case and a 90°C motor. Infrared thermometers are useless here. You need a thermocouple on the motor can, or you need to measure current draw and calculate I²R losses.

Load Types That Accelerate Heat Rise (And How to Spot Them)

Not all loads are equal. Here’s the hierarchy of thermal abuse:

1. Static Holding Load (Worst)

A robotic arm holding a weight stationary. The servo is at 100% duty cycle, zero motion. Current is maximum, heat is maximum. This is the #1 killer. A servo can hold 1.5 kg·cm forever at 25°C, but at 70°C, it can only hold 0.8 kg·cm before stalling.

2. High-Frequency Oscillation (Second Worst)

Continuous back-and-forth motion (like a flapping wing or a camera gimbal). Each direction change requires deceleration and acceleration, which means the motor is constantly in the high-current regime. The PWM frequency (typically 50 Hz) means 50 current pulses per second. The average current is lower than static hold, but the peak current heats the windings faster due to the I² term.

3. Slow, Steady Sweep (Moderate)

Moving 1° per second under load. The motor is in a quasi-steady state, drawing moderate current. Heat builds slowly, but if the sweep is continuous (like a radar scanner), the servo never cools down.

4. Intermittent Fast Motion (Best)

A servo that moves quickly, then rests. The rest period allows cooling. But if the duty cycle is >50% (moving more than resting), heat accumulates.

Practical rule: If you can hold your finger on the servo case for 5 seconds without discomfort (>55°C), you’re already in the danger zone. If you can’t touch it (>65°C), you’ve likely already damaged the pot or demagnetized the motor partially.

Mitigation Strategies: From Cheap Fixes to Proper Engineering

Level 1: Software and Control Loop Fixes (Free)

• Reduce PWM deadband: Some servos have a 10µs deadband. If you widen it to 20µs, the servo stops hunting when it reaches position, cutting idle current by 50%. This is a firmware setting on the controller side (not the servo), but many digital servos allow deadband adjustment.
• Implement a “hold release” timer: If the servo is commanded to hold position for >5 seconds, drop the command to 0µs (neutral) for 100ms, then re-apply. This lets the motor relax and the pot resettle. You lose 0.2° of accuracy but save 30% heat.
• Use a lower control voltage: A 5V servo running at 4.5V has 10% less torque but also 19% less I²R heat (since current is proportional to voltage/resistance, and heat is I²R). If you don’t need max torque, drop the voltage.
• Add a “cool-down sweep”: After a high-load operation, command the servo to do a slow 10° sweep for 3 seconds. This moves the motor’s thermal hot spot from a single winding location to the whole winding, and also redistributes the pot’s wiper wear.

Level 2: Mechanical and Thermal Mods (Low Cost)

• Copper tape heatsink: Wrap the servo case in adhesive copper tape (0.1mm thick). Copper has 400 W/(m·K) thermal conductivity vs plastic’s 0.2. This spreads heat to the entire case surface, increasing effective convection area by 30%. Measured improvement: 10–15°C drop at the motor can.
• Thermal paste + aluminum bracket: Mount the servo to an aluminum bracket using thermal paste (not double-sided tape). The bracket acts as a heatsink. A 2mm thick, 20cm² aluminum plate can absorb 40J per degree — enough to buffer transient spikes.
• Drill ventilation holes: If your application doesn’t require a sealed case, drill 3mm holes on the rear case (opposite the output shaft). This allows airflow over the motor. Combined with a small 30mm fan (5V, 0.1A), you can drop temperature by 25°C. I’ve done this on a pan-tilt rig — the servos last 4x longer.
• Replace the nylon gearbox with a metal one: MG90S vs SG90 — the metal gears don’t soften, so the gearbox retains efficiency, reducing frictional heat. Plus, metal gears have better thermal conductivity, drawing heat from the motor to the case.

Level 3: Component Upgrades (Moderate Cost)

• Switch to a coreless servo (e.g., HS-40 or DS3218): Coreless motors have lower inertia and lower resistance (thicker wire), so I²R losses are lower. But they’re still limited by the same plastic case.
• Use a servo with a “heat sink case” (e.g., Savox micro): These have an aluminum case with fins. The thermal resistance from motor to ambient drops from 60 K/W to 20 K/W. You can run them at 2x the load for the same temperature.
• Add an external motor driver (e.g., DRV8833) and bypass the internal H-bridge: This is advanced, but you can remove the servo’s internal driver and use a high-efficiency external one. The internal board’s switching losses (0.2–0.5W) are eliminated. You also get better PWM control (up to 20kHz, reducing audible whine and switching losses).

Level 4: System-Level Redesign (For Serious Builds)

• Use a gear motor + encoder instead of a servo: For continuous rotation under load (like a wheel), a geared DC motor with an external magnetic encoder has no potentiometer to fail, and you can add a proper heatsink to the motor can.
• Add active liquid cooling: Overkill, but for a high-torque robot arm, you can route a 1mm copper tube around the servo case and pump coolant. This is used in some competition robots. Temperature drops from 90°C to 40°C under continuous stall.
• Parallel servos: Instead of one servo fighting a 2 kg·cm load, use two servos with a mechanical differential (like a robotic wrist). Each servo sees 1 kg·cm, so each runs at half the current. Heat drops by 75% (since heat is I²R). The trade-off is complexity and sync issues.

How to Measure Heat Rise Properly (Without Guessing)

You need three things:

1. A fine-wire thermocouple (K-type, 0.5mm) placed directly on the motor can. Drill a 1mm hole in the case, insert the thermocouple, seal with epoxy. This gives you the true motor temp.
2. A current probe (Hall sensor or shunt resistor) on the servo power line. Calculate the instantaneous power: P = V × I. Integrate over time to get energy. Compare to the thermal model.
3. A thermal camera (optional but helpful) for surface temp distribution. The FLIR One for smartphones works well.

The test protocol: - Mount the servo to a test rig with a known load (use a pulley and weights). - Command a 90° hold. - Log temperature and current every second for 5 minutes. - Plot temperature vs time. Fit the curve to T(t) = T_ambient + ΔT_max × (1 - e^(-t/τ)). - Extract τ and ΔT_max. Compare to datasheet limits.

Acceptable limits: - Motor case: <70°C continuous, <85°C peak (for N35 magnets). - Gearbox surface: <60°C (nylon), <80°C (metal). - Potentiometer (if accessible): <60°C. - Current draw: If current doesn’t drop after reaching thermal equilibrium (i.e., it keeps climbing), you’re in the demagnetization spiral.

The “Thermal Budget” Approach to Servo Selection

Instead of asking “what’s the stall torque?”, ask “what’s the continuous torque at 50°C rise?” Every servo datasheet lies about stall torque (it’s a 1-second peak). The real spec you need is:

Continuous torque = Stall torque × (1 - (Trise / Tmax_demag))

For an MG90S: - Stall torque: 1.8 kg·cm at 5V - Thermal resistance: ~60 K/W - Max safe motor temp: 70°C - Ambient: 25°C → allowable rise: 45°C - Maximum continuous power: 45 / 60 = 0.75W - At 5V, that’s 0.15A average current - From the torque-current curve, 0.15A corresponds to ~0.3 kg·cm

So the MG90S can only continuously hold 0.3 kg·cm (17% of stall) without overheating. Anything above that is a temporary burst. If your application needs 1 kg·cm continuous, you need a servo with 3.3x the thermal capacity — either a bigger servo (like an MG996R with a metal case) or active cooling.

This is why so many robot arms “work” in the bench and then fail at the competition — they were running at 20% duty cycle during testing, but the actual task requires 80% duty cycle.

Case Study: Fixing a Pan-Tilt Camera Gimbal That Kept Dying

Symptom: Two SG90 servos on a pan-tilt rig for a 200g camera. The pan servo would jitter after 2 minutes, then fail entirely after a week.

Diagnosis: - Thermocouple on pan motor: 96°C after 3 minutes of continuous tracking. - Current draw: 0.7A steady (way above the 0.2A continuous rating). - Potentiometer resistance drift: 12% at 90°C.

Fixes applied (in order of impact):

1. Swapped to MG90S (metal gears) — dropped gearbox temp by 8°C (less friction, better conduction).
2. Added a 30mm fan blowing across both servos — dropped motor temp from 96°C to 64°C. The fan only draws 0.1A.
3. Reduced camera weight from 200g to 120g (used a smaller lens) — reduced required torque by 40%, current dropped to 0.35A.
4. Set the control loop to release hold after 3 seconds — idle current dropped from 0.3A to 0.05A during static periods.
5. Added a copper tape wrap on the case — dropped another 5°C.

Result: Motor temp now peaks at 48°C. The same servos have been running 6 months without failure. The fix cost $3 (fan + copper tape) and 30 minutes of work.

The Bottom Line: Treat Heat as a Design Constraint, Not an Afterthought

Every micro servo is a tiny furnace waiting to happen. The rated stall torque is a lie — it’s a “peak pulse” spec that assumes a 10% duty cycle. The real spec that matters is thermal derating curve, which no hobby servo provides. So you have to generate it yourself.

Your action checklist before your next build:

1. Calculate your continuous torque requirement (not peak).
2. Multiply by 2 to get the servo’s rated stall torque (so you’re at 50% of stall, which gives you ~4x thermal headroom).
3. Add active cooling if the duty cycle >30% (fan, heatsink, or both).
4. Measure the motor can temp during your worst-case scenario — not the case temp.
5. If the motor can exceeds 70°C, you have three options: bigger servo, lower load, or active cooling. Pick two.

Remember: A micro servo that runs at 60°C will last 10,000 hours. The same servo at 90°C will last 100 hours. The difference is a $2 fan and a $1 copper strip. The heat doesn’t care about your project’s deadline — it will melt your servo’s soul regardless. So design for thermal reality, measure with a thermocouple, and let your robots live long enough to be annoying.