These four parts share one SOT-23 footprint and one pin order, and the part number tells you nothing about what you are getting. The sequence runs 2302, 2303, 2304, 2312 — N-channel, P-channel, N-channel, N-channel, at 20V, 30V, 30V and 20V.
The real split is by voltage class. The 20V pair publishes on-resistance down to 2.5V of gate drive; the 30V pair starts at 4.5V. The same line divides the gate window, at ±12V below and ±20V above.
From what we see across Shenzhen reorders (2025–2026), most selection mistakes in this family come from reading the current rating and missing the drive row behind it.
At a 4.5V gate the four span 33mΩ to 180mΩ — a 5.5× range that no front-page number reveals.
| Parameter | NCE2302 | NCE2303 | NCE2304 | NCE2312 |
|---|---|---|---|---|
| Channel | N | P | N | N |
| VDS | 20V | -30V | 30V | 20V |
| VGS | ±12V | ±20V | ±20V | ±12V |
| ID @ TA = 25°C | 2.9A | -2.0A | 3.6A | 4.5A |
| ID @ TA = 70°C | not published | not published | not published | 3.6A |
| IDM pulsed | 10A | -10A | 15A | 13.5A |
| RDS(on) @ 2.5V drive | 59mΩ max | not published | not published | 40mΩ max |
| RDS(on) @ 4.5V drive | 45mΩ max | 180mΩ max | 73mΩ max | 33mΩ max |
| RDS(on) @ 10V drive | not published | 130mΩ max | 58mΩ max | not published |
| VGS(th) min / typ / max | 0.5V / 0.85V / 1.2V | -1.0V / -1.6V / -2.5V | 1.2V / 1.5V / 2.2V | 0.5V / 0.65V / 1.2V |
| Total gate charge | 4.0nC @ 4.5V | 8.5nC @ -10V | 4.0nC @ 10V | 15nC max @ 4.5V |
| Ciss | 300pF @ 10V | 226pF @ -15V | 230pF @ 15V | 500pF @ 8V |
| Crss | ~80pF | 28pF | 17pF | 140pF |
| Body diode current | 2.9A | not published | 1.6A | 1A |
| Body diode VSD | 0.75V typ @ 2.9A | -1.2V max @ -2.0A | 1.2V max @ 2.7A | 1.2V max @ 1A |
| PD (t ≤ 10 sec) | 1.0W | 1.0W | 1.7W | 1.25W |
| RθJA (same condition) | 125°C/W | 125°C/W | 73.5°C/W | 100°C/W |
| Switching times | 10 / 50 ns typ, 15 / 85 ns max | 9 / 9 / 18 / 6 ns typ | 10 / 50 / 10 / 20 ns | 40 / 40 / 108 / 56 ns max |
| Package marking | 2302 | 2303 | 2304 | 2312 |
Only one gap is left in the table. Three of the four publish a continuous body diode current and all four publish a thermal resistance — the NCE2303 is the only part that rates its diode on voltage alone.
All four publish a row at 4.5V of gate drive, which makes it the one fair comparison point in the family. Everything above 30mΩ in this chart is the cost of a higher voltage rating or of P-channel silicon, not of a worse part.
Polarity runs N, P, N, N across four consecutive-looking numbers, and voltage runs 20, 30, 30, 20. Nothing about 2303 versus 2304 predicts that one is P-channel and the other is N-channel.
What is worse, the two share a footprint and a pin order. A part placed wrongly will not show a pin error on a bench, because there is none to show.
Polarity lines up with voltage class, not with the part number. Both 20V parts are N-channel and both publish drive rows down to 2.5V.
The single P-channel part sits in the 30V row, and it is the only member of the four that needs its gate pulled to the rail to switch off.
Both 20V parts carry a ±12V gate window and publish on-resistance at 2.5V and 4.5V. Both 30V parts carry ±20V and publish at 4.5V and 10V.
That second difference is the one that costs designs. If your gate rail is 3.3V and your load rail is 24V, the part you want is in the 30V row — and neither 30V part publishes a row at your drive voltage.
The trade is not arbitrary. A wider gate window comes with a thicker gate oxide, and a thicker oxide needs more voltage to reach the same channel enhancement. You get the 20V rating or you get the 2.5V characterization, and the family does not offer both.
Three of the four publish a continuous body diode current, and the ratio against each part's own channel rating runs from 100% down to 22%. Nothing about the channel rating predicts it.
The channel ratings of the three rated parts span 2.9A to 4.5A, but their diode ratios span a factor of 4.5. Reading the channel number and assuming the diode follows is how a freewheeling path gets undersized.
All three diode figures sit under the same ten-second qualifier as the power rating. The NCE2303 is the outlier in the other direction — it publishes a forward voltage of -1.2V at -2.0A and no continuous current at all.
The four gate charge figures are measured to different gate endpoints and different drain voltages. The NCE2303 is taken to -10V, the NCE2304 to 10V, the NCE2312 to 4.5V, and the NCE2302 to 4.5V at a different drain bias again.
Gate charge scales with the swing and with the drain voltage being switched. Lining those numbers up in a column and picking the smallest is a comparison of test conditions, not of silicon.
If you need one figure to compare across the four, use the input capacitance — it is at least measured at a single stated bias per part.
Choose NCE2312 when your gate rail is 2.5V or 3.3V and your load rail is 20V or below. It is the only part here publishing a guaranteed 40mΩ at 2.5V, and it carries the highest current of the four at 4.5A.
Choose NCE2302 when you are re-sourcing an existing 20V design. Its 4.5V row of 45mΩ is beaten by the NCE2312, and its 2.5V row of 59mΩ is beaten too — but the footprint and pinout are already on your board.
Choose NCE2304 when your rail is 24V and your gate driver reaches 4.5V. It carries the lowest on-resistance of the two 30V parts at 58mΩ and the best thermal path at 73.5°C/W, and it is the only one of the four whose body diode is rated on a channel-sized basis.
Choose NCE2303 when the switch has to sit above the load. It is the only P-channel part in the family, and its -2.5V maximum threshold gives more off-state margin than a tighter logic-level part when the gate cannot reach the rail.
Choose a part outside this family when the driver is fixed and the rail is above 20V. Neither 30V part publishes a row below 4.5V, and the AO3400A or AO3401A cover that gap from the same footprint.
A: Yes — SOT-23, pin 1 gate, pin 2 source, pin 3 drain, on all four. That is what makes the family convenient and what makes it easy to fit the wrong part. The NCE2303 is the one that differs, and it differs in polarity rather than in pin order.
A: Either 20V part, and preferably the NCE2312. Both 20V parts publish a 2.5V row, so 3.3V sits inside their characterized range. The NCE2312 gives 40mΩ at 2.5V and 33mΩ at 4.5V, and carries more current than the NCE2302 at a higher power rating.
A: Because gate oxide thickness and voltage rating trade against each other. Both 30V parts carry a ±20V gate window against the 20V parts' ±12V, and a thicker gate oxide needs more voltage to reach the same enhancement. The family gives you the higher voltage or the lower drive voltage, not both.
A: No — they are measured to different gate endpoints. The NCE2303 is taken to -10V, the NCE2304 to 10V and the NCE2312 to 4.5V. Gate charge scales with the swing, so the column compares test conditions. Use input capacitance if you need a cross-part figure.
A: Three of the four — the NCE2302 at 2.9A, the NCE2304 at 1.6A and the NCE2312 at 1A. The NCE2303 publishes a forward voltage but no continuous current rating. The ratio against each part's own channel runs from 100% down to 22%, so there is no pattern to infer from the channel number.
A: No — they are opposite polarity. The NCE2303 is P-channel and the NCE2304 is N-channel, with identical pads and pin numbers. Their body diodes point in opposite directions, so a swap will not show a pin error and will show up as conduction through the diode instead.
A: The NCE2304 at 73.5°C/W, against 100°C/W for the NCE2312 and 125°C/W for both the NCE2302 and the NCE2303. All four figures are ten-second measurements on an FR4 board rather than steady-state ones.
A: Set a loss budget first, then stop. A 2A load that can accept 100mV of drop needs 50mΩ or less, and at a 4.5V gate only the two 20V parts here clear it — 45mΩ for the NCE2302 and 33mΩ for the NCE2312.
Going below the budget buys conduction loss back at the price of a larger die, more capacitance and a higher unit cost.





