Heat, junction temperature and dropout of an AMS1117, LM1117, 7805, LM317 or a low-power LDO such as the MCP1700, from your input voltage and load. Next to the result: the more efficient parts for the same job.
A linear regulator turns the whole voltage difference into heat. These parts waste less, need less headroom or draw less at rest. The ones that suit your values are marked; the estimates use typical datasheet figures.
Now (AMS1117): 125 mA from the input and 1.1 W of heat.
Converts the voltage difference instead of burning it. AP63203 (3.3 V) and AP63205 (5 V): 3.8–32 V in, 2 A, 22 µA quiescent, up to 88 % efficient even at 5 mA. TPS62160: 3–17 V, 1 A, 17 µA. Cheap modules use the LM2596 (150 kHz, 5 mA of its own, 73 % from 12 to 5 V at 3 A) or the MP1584, which MPS no longer recommends for new designs.
AP6320x: about 41 mA from the input and 99 mW of heat.
Switching modules with the 78xx pin order (input, ground, output) that fit the TO-220 footprint and need no heatsink. R-78E5.0-0.5: 7–28 V in, 500 mA, 82–92 % efficient at full load. TSR 1-2450: 6.5–36 V, 1 A, up to 94 %. There are 3.3 V versions as well.
R-78E3.3-0.5: about 45 mA from the input and 0.15 W of heat.
ESP32 modules run from 3.0 to 3.6 V, and 3.6 V is also their absolute maximum. A LiFePO4 cell spends most of its discharge around 3.2–3.3 V, so it can feed the module directly, with no losses and no quiescent current. A standard charge, though, ends at 3.65 V, above that maximum: use a charger that stops below 3.6 V and a cut-off that disconnects the module at 3.0 V. Never with a Li-ion cell: 4.2 V is far too much.
Only when the supply always stays within the module’s 3.0–3.6 V, such as a LiFePO4 cell charged below 3.6 V and cut off at 3.0 V.
Holds 3.3 V while a Li-ion cell goes from 4.2 to 3.0 V: it steps down when the cell is full and up when it is low, so you use the whole battery. TPS63802: 1.3–5.5 V in, 2 A at 3.3 V from 2.3 V, 11 µA quiescent. TPS63020: 1.8–5.5 V, 2 A at 3.3 V above 2.5 V, 25 µA.
Not for these values: 1.3–5.5 V input, up to 2 A.
Dropout in the hundreds of millivolts instead of about 1 V. TLV755P: 150 mV typical at 500 mA, input up to 5.5 V, 25 µA. AP2112K: 250 mV at 600 mA, up to 6 V, 55 µA. From a Li-ion cell they hold 3.3 V down to about 3.45–3.55 V.
Not for these values: input 1.45–5.5 V (TLV755P) or 2.5–6 V (AP2112K), up to 600 mA.
For battery nodes that sleep most of the time. TPS62740: 360 nA quiescent and up to 90 % efficient at 10 µA, 2.2–5.5 V in, 300 mA, output 1.8–3.3 V. TPS62840: 60 nA, 1.8–6.5 V, 750 mA. When the battery gets close to the output they pass it straight through (100 % mode).
Not for these values: input 1.8–6.5 V, output up to 3.6 V, up to 750 mA.
The same heat as any linear regulator, but 1–4 µA at rest instead of about 5 mA. MCP1700: 1.6 µA, 250 mA, up to 6 V. HT7333: 4 µA, 250 mA, up to 12 V. XC6206: 1 µA, 200 mA, up to 6 V. They are not enough for an ESP32 with Wi-Fi: Espressif requires a supply of at least 500 mA. Use them for sensors and microcontrollers without radio, or pick a 500 mA LDO.
Not for these values: input 2.3–6 V (MCP1700) or up to 12 V (HT7333), up to 250 mA.
Only 1 °C of margin: fine on the bench, but in a closed box or in summer it reaches the limit. Aim for at least 25 °C.
| Heat in the regulator | 1.1 W: 1.04 W from the voltage drop and 60 mW for its own consumption |
|---|---|
| Junction temperature | 124 °C = 25 °C + 1.1 W × 90 °C/W |
| Lowest input that still regulates | 4.4 V at 500 mA: 3.3 V plus 1.1 V of dropout. Your input goes down to 11.4 V. |
| Own consumption | 5 mA, that is 120 mAh a day |
| Peak current | 500 mA of 900 mA maximum |
| Input voltage | 12 V of 15 V maximum |
| Heat it can take | 1.11 W at 25 °C ambient with 90 °C/W |
To cut the heat, feed it a lower voltage (5 V instead of 12 V), or let a switching converter do most of the drop and the regulator only the last part. A resistor in series with the input also moves part of the heat out of the regulator, as long as the input at the peak current stays above the minimum shown in the results.
The copper on the board is the heatsink of a SOT-223 or SOT-23 part. TI measured the LM1117 in SOT-223 at 136 °C/W on a pad of about 8 mm² and at 66 °C/W with 6.5 cm² of copper: connect the tab to a large copper area, with vias to the other layers.
Wi-Fi draws its peaks in short bursts. Espressif asks for a supply that can deliver at least 500 mA, and a bulk capacitor close to the module helps the regulator through them.
Fit the capacitors the datasheet asks for: some regulators need them to stay stable. The AMS1117 wants 22 µF of tantalum on the output and the LM1117 at least 10 µF of tantalum, while the MCP1700 is stable with 1 µF ceramic.
A linear regulator adds no switching noise. That is why one is often kept after a buck converter for ADCs, audio and radio, with the buck doing the large drop.
A linear regulator carries the whole load current and drops the difference between input and output across itself. All of that power becomes heat, plus the small current the regulator uses to run (its quiescent current).
P = (Vin − Vout) × Iload + Vin × Iq Tj = Ta + P × θJA η = (Vout × Iload) ÷ (Vin × (Iload + Iq))
Example: an AMS1117-3.3 from 12 V feeding an ESP32 with Wi-Fi on (120 mA). P = 8.7 V × 0.12 A + 12 V × 5 mA = 1.1 W. In SOT-223, at 90 °C/W, the junction reaches 25 + 99 = 124 °C, right at the 125 °C limit. The efficiency is 26 %: three quarters of the power turns into heat.
A regulator needs some voltage across it to work. Below the output voltage plus the dropout, the output follows the input down. Bipolar parts such as the AMS1117 and the 7805 need about 1.1 V and 2 V; CMOS LDOs such as the MCP1700 need tens to a few hundred millivolts, rising with the current. The check uses the peak current and the lowest input voltage, because that is where the regulator fails first.
Vin,min = Vout + Vdropout(Ipeak)
Example: an AMS1117-3.3 needs 3.3 + 1.1 = 4.4 V. From USB (4.75 V or more) it works; from a Li-ion cell, which goes from 4.2 V down to 3.0 V, it never regulates properly.
Tj = Ta + P × (θJC + θCS + θSA) θSA ≤ (Tjmax − 25 °C − Ta) ÷ P − θJC − θCS
In free air a 7805 in TO-220 has θJA = 50 °C/W: at 25 °C ambient it reaches 125 °C with 2 W. On a heatsink the heat flows through the case (θJC, 5 °C/W), the contact (θCS, about 1 °C/W with thermal paste, assumed here) and the heatsink (θSA, from its datasheet). The calculator sizes the heatsink to keep 25 °C of margin below the limit.
| Part | Package | Dropout | Quiescent | Max current | Max input | θJA |
|---|---|---|---|---|---|---|
| AMS1117 | SOT-223 | 1.1 V / 800 mA | 5 mA | 900 mA | 15 V | 90 °C/W |
| LM1117 | SOT-223 | 1.2 V / 800 mA | 5 mA | 800 mA | 15 V | 61.6 °C/W |
| 78xx (L7805…) | TO-220 | 2 V / 1 A | 8 mA | 1.5 A | 35 V | 50 °C/W |
| 78Lxx (78L05…) | TO-92 | 1.7 V / 40 mA | 3.8 mA | 100 mA | 30 V | 156 °C/W |
| LM317 | TO-220 | 2 V / 1 A | 5 mA | 1.5 A | 40 V (input − output) | 50 °C/W |
| MCP1700 | SOT-23 | 178 mV / 250 mA | 1.6 µA | 250 mA | 6 V | 336 °C/W |
| MCP1700 | TO-92 | 178 mV / 250 mA | 1.6 µA | 250 mA | 6 V | 160 °C/W |
| HT73xx (HT7333…) | SOT-89 | 90 mV / 40 mA | 4 µA | 250 mA | 12 V | 200 °C/W |
| XC6206 | SOT-23 | 250 mV / 100 mA | 1 µA | 200 mA | 6 V | 200 °C/W |
| AP2112K | SOT-23-5 | 250 mV / 600 mA | 55 µA | 600 mA | 6 V | 184 °C/W |
| TLV755P | SOT-23-5 | 150 mV / 500 mA | 25 µA | 500 mA | 5.5 V | 231 °C/W |
Typical values at 25 °C from the makers’ datasheets (AMS, Texas Instruments, STMicroelectronics, Microchip, Holtek, Torex, Diodes). The guaranteed maximum dropout is higher: 1.3 V for the AMS1117. Each maker measures θJA on its own test board, so the figures are only roughly comparable.
Small packages shed their heat through the copper they are soldered to. TI measured the LM1117 in SOT-223 at 136 °C/W with a pad of about 8 mm² and at 66 °C/W with 6.5 cm² (one square inch) of copper; AMS gives 55 to 80 °C/W for the AMS1117 depending on the copper area. The datasheet figure gives the order of magnitude, and the copper on your board decides the rest.
A linear regulator is cheap, small and quiet: its output carries no switching ripple, which is why it is still used for ADCs, audio and radio, often after a switching converter. It makes sense when the voltage difference is small or the current is low. With a large drop (12 V to 3.3 V) or hundreds of milliamps, a buck converter wastes far less. On a battery that sleeps most of the time, what counts is the current drawn at rest: 1.6 µA for an MCP1700, about 5 mA for an AMS1117.
P = (Vin − Vout) × Iload + Vin × Iq. From 12 V to 3.3 V at 120 mA that is 8.7 V × 0.12 A = 1.04 W, plus 12 V × 5 mA = 0.06 W for an AMS1117: 1.1 W in total. Multiply by θJA and add the ambient temperature to get the junction temperature.
Because it burns the whole voltage difference. From 12 V at 120 mA it dissipates 1.1 W, and in SOT-223 at 90 °C/W that means about 100 °C above ambient. From 5 V the same load gives 0.23 W and about 21 °C above ambient. Lower the input voltage or use a buck converter.
No. It needs about 1.1 V of headroom, so at least 4.4 V at the input, while a Li-ion cell goes from 4.2 V down to 3.0 V. Use a low-dropout CMOS LDO (the TLV755P has 150 mV at 500 mA, so it regulates down to about 3.45 V) or, to use the whole battery, a buck-boost converter such as the TPS63802.
From 12 V to 5 V at 500 mA it dissipates 7 V × 0.5 A + 12 V × 8 mA = 3.6 W. To stay 25 °C below the 125 °C limit at 25 °C ambient, θSA must be at most 75 ÷ 3.6 − 5 − 1 = 15 °C/W. Without a heatsink it would reach 25 + 3.6 × 50 = 205 °C, and the thermal protection would cut in.
Yes. The Recom R-78E and Traco TSR 1 modules have the same pin order (input, ground, output), fit the TO-220 footprint and need no heatsink. Check the input range (R-78E5.0-0.5: 7–28 V, 500 mA) and fit the capacitors that the module’s datasheet asks for.
From a LiFePO4 cell, yes, with care: ESP32 modules accept 3.0 to 3.6 V (3.6 V is also their absolute maximum) and the cell sits around 3.2–3.3 V for most of its discharge. A standard charge ends at 3.65 V, so use a charger that stops below 3.6 V, and cut the module off at 3.0 V. From a Li-ion cell, no: at 4.2 V it is far above the maximum.
Yes: when the voltage drop is small (5 V to 3.3 V at a low current), when the circuit is sensitive to noise (ADC, audio, radio), or when size and cost matter more than efficiency. A common combination is a buck converter close to the target voltage followed by a low-noise LDO.