Why Your SOP-8 MOSFET Gets Hot — and When to Move to TO-252

Where Does the Heat Actually Come From?

Three loss mechanisms, and only one gets the attention it deserves:

• Conduction loss (I²R). The one everyone knows. At 10A, a 53mΩ part burns 5.3W; a 5.5mΩ part burns 0.55W. Ten times the resistance, ten times the heat — at the same current.
• Switching loss. Every edge the FET spends in the linear region dumps power into the die. Slow gate drive + high Q_g = free heat. A 70nC part driven by a GPIO through a resistor stays in that zone for most of the edge.
• Partial enhancement (the quiet one). Drive a part specified at 4.5V with 3.3V and it conducts — at 2–3× its rated R_DS(on). The circuit works, the heat is just always there. No fault, no warning, just a warm board.

Rule of thumb: if the load current times itself times the resistance comes out above the package's power budget, the design is wrong regardless of the part's datasheet rating.

The Package Is the Limit — Not the Die

The numbers are why the SOP-8's 6A rating is a "datasheet rating" and the TO-252's 100A rating is a "capability." 4A through 28mΩ is 0.45W — half the NCE9926's 1.25W budget.

4A through 53mΩ is 0.85W — already a 42°C rise on the NCE9435's 50°C/W. The die is fine; the package is sweating.

The TO-252's tab changes the physics. It's soldered to a copper pour, which is soldered to the ground plane — the heat leaves through the board, not the air. That's how 110W becomes physically possible.

The Three Heat Traps We See Most

Trap 1: driving a 4.5V-specified part with 3.3V. The P-Channel 9435 family is specified from -4.5V. At -3.3V the channel conducts at 2–3× the resistance — and the "unexplained heat" is just the missing gate headroom.

In our experience, this is the single most common cause of warm P-Channel switches in battery products.

Trap 2: an MCU pin driving a power-stage gate. A 70nC gate charge through a GPIO is a linear-region ride. The FET turns on, the load runs, and the switching loss roasts the die.

The part isn't wrong — the driver is missing. If Q_g is above ~20nC, budget a real gate driver.

Trap 3: a remarked die. In Shenzhen (2025–2026), the 9435 family is one of the most frequently remarked sockets. The counterfeits pass visual checks and even V_GS(th) spot tests.

Measure R_DS(on) at logic level and a smaller die fails by 2–3×. The board runs hot, fails thermally, and nobody suspects the part.

When to Move to TO-252 (and When Not To)

Move to TO-252 when: the load is above ~4A continuous, the board has a copper pour to spare, and a gate driver exists or is budgeted. A TO-252 N-Channel at 10A burns 0.55W — the same 10A that pushes an SOP-8 past its budget.

Don't move when: the switch is high-side and simple (a P-Channel SOP-8 needs no driver), the current is under 2–4A, or the layout can't take a tab pour. A TO-252 without copper is a TO-252 without its advantage — the tab does nothing if it floats.

The 60-Second Loss Check

Before you blame the MOSFET, run this on a napkin:

1. I²R. Load current squared, times the R_DS(on) at your actual gate voltage. Compare against the package P_D.
2. Gate drive. Is the gate at the voltage the part is specified at — not the voltage that "kind of turns it on"?
3. Package. Is the thermal path real — pins on copper, tab soldered, no floating island?

Most "hot MOSFET" tickets die at step one. If I²R fits the budget and the board still burns, check steps two and three — that's where the free heat lives.