LM317 Heat: Dissipation, Heatsink Sizing, and the 3 V Rule
The LM317 is a 1.5 A regulator that cannot deliver 1.5 A from 15 V without a heatsink bigger than the board. The heatsink size is set by one equation, and the headroom rule catches two common design mistakes.
Quick Answer
- Power: P = (Vin − Vout) × Iout + Vin × 5 mA. At 12→5 V and 0.5 A that is 3.56 W.
- Heatsink: θSA ≤ (125 − Ta)/P − θJC − θCS. For 3.56 W at 40 °C with a TO-220, that is 7.5 °C/W to stay at 125 °C, and under 1 °C/W to stay at 100 °C.
- Headroom: the part needs about 3 V in-to-out. A 5 V input cannot make 3.3 V.
- Practical: above ~1 W of dissipation, a buck converter is smaller, cooler and cheaper than regulator plus heatsink.
1. The 3 V rule, and the 5 V trap
The LM317 is an adjustable regulator with a 1.25 V reference and a minimum output of 1.25 V. Its pass element is a Darlington-like NPN, which means it needs headroom: the datasheet says the device requires up to 3 V between input and output to stay in regulation.
Vout = 3.3 V -> Vin >= 3.3 + 3.0 = 6.3 V Vout = 5.0 V -> Vin >= 8.0 V Vout = 12 V -> Vin >= 15 V
This is the first trap: a 5 V rail feeding an LM317 set to 3.3 V is outside the specified operating condition. It may produce something close to 3.3 V at light load and collapse under load or at low temperature, and the failure looks like a thermal problem because the part is also hot.
The second trap is temperature. Dropout rises at the extremes; the typical curve is about 2.2 V at 25 °C and near 2.5 V at −50 °C and 150 °C. Designing to exactly 3 V leaves no margin for a cold start.
Why the headroom is so large
The pass element is a Darlington: two base-emitter junctions in series, plus the current source that drives them. That is why the minimum differential is roughly 2 V of junction drop plus the drop across the current-setting resistor, and why it is specified as 3 V rather than the 0.3 V of a modern LDO.
The same Darlington is why the part is robust: it survives input transients and output shorts that would destroy a low-voltage LDO. The trade is headroom and dissipation — a deliberate choice in the original design, not a defect.
Minimum load
The LM317 is not a true low-dropout regulator, and it needs a minimum load current to keep the feedback loop in regulation. The datasheet specifies this, and it is typically a few milliamps. If the load can fall to zero — a relay that is off, a microcontroller in deep sleep — add a permanent bleed resistor at the output or the voltage will rise.
There is also an absolute limit on the differential: the device is rated for 40 V between input and output, and the input must stay below 40 V. A 36 V rail feeding a 1.25 V output is 34.75 V of differential and 30 W at 1 A — electrically allowed, thermally impossible.
2. Dissipation: the equation that decides everything
P = (Vin - Vout) x Iout + Vin x Iq Iq ~ 5 mA (bias current, flows to the output) 12 V in, 5 V out, 0.5 A: P = 7 x 0.5 + 0.06 = 3.56 W
The bias current is small but not negligible at high input voltage: 12 V × 5 mA is 60 mW, about 2 % of the total in this example and irrelevant at 1 A. It matters when the load is light and the input is high, which is also when a linear regulator is least efficient.
Note that the bias current flows out of the ADJUST pin path into the load, so it is not wasted in the same way as a fixed regulator's ground current — but it still comes from the input and still adds to dissipation.
Where the heat actually goes
In a TO-220 the die is mounted on the metal tab. Heat travels die → tab → (compound) → heatsink → air. Without a heatsink, the tab is the heatsink, and the number that matters is RθJA, not RθJC.
This is why "it has a metal tab, it will be fine" is not an argument. A TO-220 with no heatsink has a junction-to-ambient resistance near 23.5–38 °C/W depending on the board and airflow; at 3.5 W that is still a 90–130 K rise.
3. Heatsink sizing: four thermal resistances in series
Heat flows from junction to case, case to heatsink, heatsink to air:
Tj = Ta + P x (RthJC + RthCS + RthSA) RthJC junction-to-case, from the datasheet RthCS case-to-heatsink, ~0.5 C/W with compound RthSA heatsink-to-ambient, the part you choose RthSA <= (Tjmax - Ta)/P - RthJC - RthCS
Ambient is the variable people underestimate. The same 3.56 W design in a 5 °C/W heatsink gives a 116 °C junction at 40 °C, 131 °C at 55 °C, and 146 °C at 70 °C. A sealed enclosure at 55 °C ambient is not unusual for industrial products, and it is enough to make this design fail.
| Package | RθJC | RθJA (no heatsink) | Comment |
|---|---|---|---|
| TO-220 (KCS) | 15.9 °C/W | 23.5 °C/W | Best for heatsink mounting |
| TO-263 (KTT) | 36.5 °C/W | 38.0 °C/W | SMD, heatsink via copper |
| SOT-223 (DCY) | 43.2 °C/W | 66.8 °C/W | Low current only |
For 3.56 W in a TO-220, θSA must be 7.5 °C/W to stay at 125 °C in 40 °C ambient, and about 0.5 °C/W to stay at 100 °C. The first is a small clip-on heatsink; the second is a chassis-mount extrusion. This is the moment to ask whether a buck converter is a better answer.
Mounting details that change the number
- Thermal compound: a thin, even layer. Too much is worse than too little; the interface resistance is specified with a void-free joint.
- Insulator: a mica or sil-pad washer adds 0.5–2 °C/W and is required when the tab must not be at output potential.
- Mounting torque: follow the datasheet. Under-torque leaves voids; over-torque cracks the die.
- Airflow: natural convection gives the datasheet numbers; a fan changes them by a factor of two or more, which is why a fan-cooled test is not a product test.
4. Worked examples
| Case | P | Result |
|---|---|---|
| 12 → 5 V, 0.5 A, SOT-223 | 3.56 W | Tj ≈ 278 °C → max ≈ 0.18 A |
| 12 → 5 V, 0.5 A, TO-220 + 5 °C/W | 3.56 W | Tj ≈ 116 °C at 40 °C ambient |
| 15 → 1.25 V, 1 A, TO-220 + 2 °C/W | 13.8 W | Impossible; max ≈ 0.34 A |
| 9 → 5 V, 0.3 A, TO-220 + small sink | 1.24 W | Tj ≈ 63 °C, comfortable |
5. Efficiency: why the heatsink exists
A linear regulator's efficiency is approximately Vout/Vin:
5 V out from 12 V in: 42 % 5 V out from 9 V in: 56 % 12 V out from 15 V in: 80 % 3.3 V out from 6.3 V: 52 %
This is why the same regulator is perfectly reasonable for a 15→12 V rail at 0.5 A (1.5 W) and unreasonable for 12→5 V at 1 A (7 W). The package did not change; the voltage difference did.
The two-stage compromise
If the load needs low noise and the input is 12 V, a buck converter followed by an LM317 is often the best answer: the buck drops most of the voltage efficiently, and the linear stage cleans up the switching ripple with a small dissipation.
12 V -> buck 88 % -> 6.8 V -> LM317 -> 5 V at 0.5 A Buck loss: 1.65/0.88 - 1.65 = 0.23 W LDO loss: (6.8 - 5) x 0.5 = 0.90 W Total: 1.13 W instead of 3.56 W
The linear stage still needs a small heatsink, but a fraction of the size, and the output noise is set by the regulator rather than the switcher.
Reading the bench results correctly
Measure the tab, not the plastic. With a thermocouple on the tab, the junction is higher by P × RθJC — about 57 K at 3.56 W for a TO-220. A tab at 70 °C can mean a junction near 127 °C, which is already at the limit.
Also measure at the real ambient. A board on a bench at 23 °C with open air is not the same as a board in a closed box at 55 °C, and the difference is larger than any tolerance in the heatsink datasheet.
6. Failure modes worth recognising
- Thermal shutdown cycling: the output collapses every few seconds under load. The part is protecting itself; the design is over-dissipating.
- Dropout, not heat: output sags only at high current or cold. Input is below Vout + 3 V.
- Oscillation: missing output capacitor or a long input lead. The LM317 needs an output capacitor for stability, and an input bypass if the filter is far away.
- Adjust pin floating: output rises toward Vin. The device is protected, but the load is not.
- Wrong resistor ratio: Vout = 1.25 × (1 + R2/R1) + Iadj × R2. The 50 µA adjust current adds a small error that grows with R2.
Protection diodes
If the output has a large capacitor and the input can be shorted or removed while the output stays charged, add a reverse diode from output to input. Otherwise the capacitor discharges through the regulator's internal junction, which is a path the datasheet warns against.
The same applies to the ADJ pin when a bypass capacitor is fitted: a diode from ADJ to output prevents the capacitor from discharging into the adjust pin during a fast output collapse. These are two cheap diodes that prevent a failure mode that looks like a random dead regulator.
7. When to stop using a linear regulator
| Situation | LM317 fits? | Better |
|---|---|---|
| Vin − Vout ≈ 3 V, I ≤ 0.5 A | Yes | LM317 with small heatsink |
| 12 → 5 V, 1 A | No (7 W) | Buck converter |
| 5 → 3.3 V | No (headroom) | LDO such as AMS1117 |
| Low-noise analog rail from a switching preregulator | Yes | Two-stage: buck + LM317 |
| Sealed enclosure, 60 °C ambient | Only under ~0.5 W | Switching regulator |
8. Verifying the design on the bench
| Check | Pass | If it fails |
|---|---|---|
| Vin at full load | ≥ Vout + 3 V | Output sag; reduce load or raise Vin |
| Heatsink temperature, 30 min | Stable, < 90 °C | Bigger heatsink or lower current |
| Vout accuracy | Within 2 % at 25 °C | Check R1/R2 and adjust-pin current |
| Stability at load steps | No sustained oscillation | Output capacitor and layout |
| Cold start | Regulates at lowest operating temperature | Headroom margin too small |
Our LM317DCYR is stocked in SOT-223 for low-current use; for heatsink mounting, choose the TO-220 package and size θSA from the equation above. More linear regulators are listed in the LDO and linear regulator collection, and our traceability process covers date-code and lot checks on incoming regulators.
For a lower-current alternative in the same family, the LM317L in SOIC-8 is limited to 100 mA and has a different thermal path — check the package before substituting by name.
9. FAQ
Can the LM317 run from 5 V to make 3.3 V?
Not within specification. It needs about 3 V of headroom, so 3.3 V output requires at least 6.3 V input. From 5 V it will drop out under load or when cold.
How hot is too hot for the package?
Design for 100 °C junction and treat 125 °C as the absolute operating limit. The case temperature will be lower than the junction by P × RθJC, which is why measuring the tab is more useful than touching the plastic.
Do I need a heatsink at 0.5 A?
That depends on the voltage difference, not the current. From 9 V to 5 V at 0.5 A (2 W) a small clip-on is enough; from 12 V to 5 V at 1 A (7 W) no practical heatsink saves the design.
Why does the output rise when the adjust pin is open?
The regulator regulates the 1.25 V between output and adjust. With the pin floating, the feedback loop is broken and the output rises toward the input. The part survives; the load may not.
Is a larger output capacitor enough for stability?
Usually yes, but the datasheet gives a range. Too small can oscillate, and extremely large low-ESR capacitors can also upset some variants. Follow the recommended value and place it close to the output pin.
What replaces an LM317 in a 12 V system?
A buck converter for the main rail, optionally followed by a linear regulator for a quiet analog section. The two-stage approach keeps efficiency high and noise low.
Why does the LM317 need an output capacitor?
Without one, the feedback loop can oscillate, especially with long input leads. The datasheet gives a recommended value; the capacitor should be close to the output pin, and a small ADJ bypass capacitor improves ripple rejection.
Can I parallel two LM317s for more current?
Not directly. Like other linear regulators, they do not share current evenly. Ballast resistors or a dedicated current-sharing scheme are needed, and a switching regulator is usually the simpler answer.
10. Sources
- Texas Instruments, LM317 3-Pin Adjustable Regulator datasheet — 1.5 A, 1.25–37 V, 3 V headroom, RθJA/RθJC per package, thermal overload protection.
- Texas Instruments, Linear Regulator Design Guide for LDOs (SLVA118A) — thermal and layout guidance.
- Texas Instruments, ESR, Stability, and the LDO Regulator (SLVA115) — capacitor and stability conditions.
Sizing a linear regulator?
Send the input voltage, output voltage, load current and enclosure ambient. We will compute dissipation and the heatsink you actually need.
