The TO-247-4L package is the four-lead version of the TO-247 outline, and the fourth pin is a Kelvin source that gives the gate driver its own return path to the die, out of the power current. Published same-die comparisons put the gain anywhere between 8% and 71% lower switching loss. Which end of that range you land on depends almost entirely on di/dt. Here is the mechanism, the published numbers, and the arithmetic that decides whether it pays.

Key takeaways

• The fourth pin carries gate current only. It is the driver’s reference terminal, not a second power connection.
• Published same-die results span 8% lower switching loss on silicon superjunction to 71% lower turn-on loss on 1200 V SiC. The spread tracks di/dt.
• Do not subtract Eon from the three-lead and four-lead datasheets of the same die. Vendors characterise the two at different gate resistance and different current.
• Pin order reverses and the pitch changes from 5.44 mm to 5.08 mm. There is no drop-in path from a three-lead footprint.
• An external gate-source capacitor is permitted on a Kelvin-source package and is explicitly not recommended on TO-247-3.

What the Kelvin source pin actually does

The TO-247-4L package adds a driver source pin bonded to the source metallisation inside the package that carries only gate current. The gate driver returns to that pin instead of to the power source lead, so the voltage developed across the source lead inductance by drain current no longer subtracts from the applied gate-source voltage.

That subtraction is the whole problem. In a three-lead package the source lead sits in both the power loop and the gate loop. Any di/dt in the drain current develops a voltage across it, opposing the driver during turn-on and adding to it during turn-off.

The numbers stay small until they are not. Wolfspeed’s PCB layout note puts a 15 mm long, 3 mm wide shared return trace at about 8 nH and shows a 250 A/µs edge across it costing 2 V of effective gate drive. Power Master quotes 8 nH to 10 nH of common source inductance for the TO-247-3L package itself, and Wolfspeed’s SiC design guidance uses 12 nH for the same package.

Take the middle of that band at SiC slew rates. At 1000 A/µs through 10 nH the induced drop is 10 V. Against a +15 V drive with a threshold near 2.5 V, that removes most of the overdrive at the moment the channel is trying to open. The crossover period lengthens and the loss lands in turn-on energy.

[IMAGE 1: Side-by-side gate loop schematic showing the shared source inductance in TO-247-3 and the separated driver source path in TO-247-4L | alt: “TO-247-4L package Kelvin source pin removing source inductance from the gate driver loop”]

TO-247-4L pinout and outline: what changes from three leads

The pin order does not simply gain a lead, it reverses. On Wolfspeed’s TO-247-3 drawing the sequence is gate, drain, source. On its TO-247-4L drawing the sequence is drain, source, driver source, gate, with the tab as a fifth drain node.

That rearrangement is deliberate. Placing the source between gate and drain reduces the drain-to-gate coupling that drives gate oscillation at turn-off under high dv/dt, which is why Infineon describes the four-pin arrangement as easier to lay out rather than harder despite the extra lead. Power Master gives the same reason, stating that the pinout differs from TO-247-3L to preserve creepage from the high-voltage drain lead.

Attribute	TO-247-3 (Wolfspeed PRD-05964)	TO-247-4L (Wolfspeed PRD-05966)
Pin 1	Gate	Drain
Pin 2	Drain	Source
Pin 3	Source	Driver source (Kelvin)
Pin 4	Not present	Gate
Tab	Drain	Drain
Lead pitch	5.44 mm BSC	5.08 mm BSC
Body length D	20.75 to 21.05 mm	23.30 to 23.63 mm
Body width E	15.75 to 16.13 mm	15.75 to 16.13 mm
Lead length L	19.73 to 20.48 mm	17.31 to 17.82 mm
Drain-to-source creepage	Not stated on the datasheet	8 mm (datasheet feature list)
RθJC, same 1200 V 16 mΩ die	0.27 °C/W	0.23 °C/W
TJ range	−40 to +175 °C	−55 to +175 °C
Mounting torque	1 N·m (M3 or 6-32)	1 N·m (M3 or 6-32)

Table 1. Wolfspeed C3M0016120D against C3M0016120K: the same 1200 V 16 mΩ die in both outlines. Datasheet revisions Rev. 02 (September 2024) and Rev. 5 (November 2025).

The mechanical envelope also grows. On Wolfspeed’s own drawings the four-lead body runs 23.30 mm to 23.63 mm against 20.75 mm to 21.05 mm for three leads, and the pitch drops from 5.44 mm to 5.08 mm to fit four leads into a similar width. Lead length falls from a 19.73 mm to 20.48 mm range down to 17.31 mm to 17.82 mm. Nothing about a three-lead footprint accepts this part.

Creepage moves in your favour. Wolfspeed specifies 8 mm of creepage between drain and source on the TO-247-4L device and states in its layout note that TO-247-4 and TO-267-7 both offer more creepage than TO-247-3.

[IMAGE 2: Pin maps of TO-247-3 (G, D, S) and TO-247-4L (D, S, driver source, G) with pitch dimensions marked | alt: “TO-247-4L package pinout and 5.08 mm lead pitch compared with the TO-247-3 footprint”]

What the Kelvin pin is worth: five published results

The table below collects the same-die comparisons published by manufacturers that state their test conditions. Read the technology column before the percentage column.

Source	Technology	Matched pair	Stated conditions	Published result
Infineon press release, 2013	600 V Si superjunction (CoolMOS C7)	Same die, TO-247 vs TO-247-4	CCM PFC, 1.2 kW	Switching loss up to 8% lower; 5 W less device dissipation; +0.4 percentage points full-load efficiency
Infineon CoolMOS TO-247 4pin page (current)	Si superjunction	Same die, 4-pin vs 3-pin	Not stated on the page	3x lower switching losses claimed; +0.6 percentage points full-load efficiency; enables stepping from a 45 mΩ to a 60 mΩ die
Toshiba DTMOSIV-H article	600 V Si superjunction	TK62N60X (TO-247) vs TK62Z60X (TO-247-4L)	Measured turn-on waveform, gate-source voltage probed at the die	Turn-on loss 19% lower
Wolfspeed design guidance	1200 V SiC	Same product, TO-247-3 vs TO-247-4	30 A IDS; 12 nH source inductance in the three-lead case	Total switching loss approx. 430 µJ vs approx. 150 µJ
Power Master AN-CM2307 Rev. 1, June 2023	1200 V 80 mΩ SiC	PCW120N80M1 (TO-247-3L) vs PCZ120N80M1 (TO-247-4L)	1200 V; VGS −3/+18 V; RG 2 Ω; SiC freewheeling diode; ID 25 A	Eon 384 → 110 µJ (−71%); Eoff 88 → 63 µJ (−28%). In a 5 kW 40 kHz boost stage, total device loss −18%
Wolfspeed PRD-06752 Fig. 3	650 V SiC	C3M0060065D (TO-247-3) vs C3M0060065K (TO-247-4)	400 V; RG,ext 2.5 Ω; VGS −4/+15 V; body diode; TJ 25 °C; 0 to 25 A	Loss curve published across the sweep; no single numeric delta stated

Table 2. Every manufacturer-published same-die TO-247-4L comparison with stated conditions, as of August 2026. Percentages are not comparable across rows because the conditions differ.

Two things fall out. Silicon superjunction devices see single-digit to roughly 20% improvement. Silicon carbide sees 30% to 71%. That is not vendor spin, it is di/dt: a superjunction MOSFET switching a few hundred amps per microsecond develops a volt or two across the source lead, and a SiC device switching several thousand develops enough to matter.

The second is that device percentages and converter percentages are different quantities. Power Master’s 71% turn-on improvement at the device becomes 18% of total device loss once conduction loss enters a 5 kW boost stage. Infineon’s 8% switching-loss claim becomes 0.4 percentage points of full-load efficiency in a 1.2 kW CCM PFC. Budget against the system number.

The measurement trap

Do not compute the benefit by subtracting datasheet numbers. Vendors characterise the two package variants under different conditions, and the difference flatters the four-lead part.

Wolfspeed’s C3M0016120D, the TO-247-3 device, is specified at 75 A with 5 Ω of external gate resistance. The C3M0016120K, the same 1200 V 16 mΩ die in TO-247-4, is specified at 80 A with 2.5 Ω. Different current, different drive impedance. Subtracting those columns overstates the package contribution and hides the gate resistance change.

There is a second trap in the lab. A three-lead device on a bench fixture that senses gate-source voltage close to the package will not show the loss the same device shows on a production board with 15 mm of shared return. The clean method is what Power Master and Wolfspeed both used: one board per package, with supply voltage, gate drive voltage, gate resistance, freewheeling diode and junction temperature held fixed while drain current is swept.

Gate drive rules that only apply to the TO-247-4L package

Some layout guidance changes when the Kelvin pin exists, and one rule reverses outright.

• An external gate-source capacitor is permitted on a Kelvin-source package. Wolfspeed states in its layout application note that a gate-source capacitor applies only to packages with a KS pin and that it does not recommend one on TO-247-3. If you carried a damping capacitor across from a three-lead design without checking, check now.
• A gate-source resistor of roughly 10 kΩ belongs next to the device on either package. Its job is to discharge the gate if the MOSFET becomes disconnected from the driver, which otherwise permits false turn-on.
• Route the driver return to the Kelvin pin and nowhere else. The pin is sized for gate current. Tying it to the power source lead at the device restores the shared inductance and gives back the entire benefit while leaving the cost.
• When paralleling four-lead devices, place the driver IC at the centre of the two gate terminals rather than at the centre of the two packages, keep the gate and Kelvin source trace lengths equal, and add resistance in each source return path to limit circulating current caused by unequal source inductances.
• Keep the gate loop off the power loop. Wolfspeed’s worked case is blunt: 0.01 mm² of overlap between drain and gate traces on FR4 gives about 38 pF, and at 800 V and 100 kHz that costs 1.2 W in the gate loop alone.
• Silicon carbide still wants a negative turn-off bias. The Kelvin pin reduces the induced disturbance at the gate, it does not remove the need for a hard off state.

Worked example: does it pay at your switching frequency

Take Power Master’s measured pair at 25 A. Turn-on energy falls from 384 µJ to 110 µJ and turn-off from 88 µJ to 63 µJ, so the package saves 299 µJ per switching event.

Multiply by switching frequency. At 40 kHz that is 12.0 W per device, at 100 kHz it is 29.9 W, at 20 kHz it is 6.0 W. Through a junction-to-case path of 0.23 °C/W and a case-to-sink path of, say, 0.5 °C/W, the 40 kHz figure is roughly 9 °C of junction temperature you no longer have to remove.

Now run it at low current. The same sweep starts at 5 A, where the packages sit close together because di/dt is low and the induced gate voltage is a fraction of a volt. That end of the curve does not justify a new footprint.

The decision rule is one multiplication: per-event energy delta times switching frequency. If the answer is under about 1 W per device, the four-lead package buys nothing you can see on a thermal camera, and a three-lead part with a well-separated return trace will do.

Sourcing: the one-letter decoder

Four-lead and three-lead siblings usually differ by a single character in the part number, and the character is not the same across vendors. The decoder below is assembled from manufacturers’ own naming documents and from matched part-number pairs.

Vendor	3-lead marker	4-lead marker	Example parts	Basis
Wolfspeed	D	K	C3M0016120D / C3M0016120K (same die)	PRD-06752 Figure 1 documents D = TO-247-3, K = TO-247-4
Toshiba	N	Z	TK62N60X / TK62Z60X (same die)	Part-naming page documents N = TO-247, Z = TO-247-4L
Power Master	W	Z	PCW120N80M1 / PCZ120N80M1 (same die)	AN-CM2307 identifies the pair by package
Infineon	IPW / IMW	IPZ / IMZ	IPW65R041CFDFKSA1 (3-pin) and IPZ65R045C7XKSA1 (4-pin); different die	Distributor listings state pin count and package for each
onsemi	NTHL	NTH4L	NTHL020N120SC1 (TO-247-3L) and NTH4L013N120M3S (TO-247-4L); different die	onsemi SiC MOSFET product table

Table 3. Part-number markers that distinguish the four-lead variant from the three-lead one. Where the example parts are not the same die, the row documents the naming convention only.

Two consequences. Searching a distributor by package name misses parts whose only difference is that letter, so search the base number instead. And a cross-reference tool matching on voltage and on-resistance will hand you the wrong lead count without comment.

Lifecycle, cost and second sourcing

The four-lead variant is no longer a premium part. Infineon’s IPZ65R045C7XKSA1, a 700 V 45 mΩ CoolMOS C7 in TO-247-4, was listed at RS Components at £6.07 in single quantity with 324 units in stock at the time of writing.

At the fast end of the range the pressure now runs the other way. DigiKey lists Wolfspeed’s three-lead C3M0016120D as obsolete and no longer manufactured, while the four-lead C3M0016120K remains active with a November 2025 datasheet revision. onsemi’s newer 1200 V M3S and M3P silicon carbide parts appear in TO-247-4L and D2PAK-7L, with TO-247-3L carrying the older M1 generation.

For second sourcing, compare outline drawings before electrical parameters. onsemi issues its four-lead drawing separately as TO-247-4LD, CASE 340CJ. Qualify a named drawing at a named revision, and buy through authorized distribution.

[IMAGE 3: Decision flowchart from device technology and switching frequency through the watts-saved test to TO-247-4L or TO-247-3 | alt: “Decision path for choosing the TO-247-4L package with a Kelvin source pin over TO-247-3”]

Frequently asked questions

What is the Kelvin source pin on a TO-247-4L package?

It is a fourth lead bonded to the source metallisation on the die and used only as the gate driver’s return. Because it carries no power current, no voltage develops across it from drain di/dt, so the driver sees the gate-source voltage it actually applied. That is the entire mechanism behind the lower switching losses.

Is TO-247-4 the same as TO-247-4L?

In practice yes. Wolfspeed and Infineon write TO-247-4; onsemi, Toshiba and Power Master write TO-247-4L; onsemi’s own case drawing says TO-247-4LD. All describe the four-lead Kelvin-source variant of the TO-247 outline. Body dimensions still vary between vendors, so treat the names as interchangeable in conversation but not in a drawing.

Can I use a TO-247-4L device in a TO-247-3 footprint?

No. The pin order reverses, the count changes and the pitch drops from 5.44 mm to 5.08 mm. The four-lead body is also about 2.5 mm longer. Retrofitting needs a new footprint and a rerouted gate loop, which is why it is worth laying out for four leads at the start of a hard-switched SiC design.

Can I connect the Kelvin source pin to the power source pin?

Only at the driver, and only if you have no alternative. Bonding the two at the device puts the source lead inductance back into the gate loop and removes the benefit you paid for. Leaving the pin open is worse, because the driver then has no defined return.

How much switching loss does a Kelvin source pin actually save?

Between about 8% and 71%, depending on device technology and current. Silicon superjunction parts sit at the low end and silicon carbide at the high end. Take the per-event energy difference from a same-conditions measurement, multiply by your switching frequency, and judge the answer in watts rather than percentages.

Specify it, or stay at three leads

Choose TO-247-4L when the device is silicon carbide and hard-switched, when the per-event energy saving times your switching frequency exceeds roughly a watt per device, when you are already fighting turn-off gate ringing, or when the part you want is released only in four leads. Above 1200 V and below about 20 mΩ that last condition is increasingly the deciding one.

Stay at three leads when the switch is an IGBT or a rectifier, when a silicon superjunction MOSFET runs below roughly 20 A and 20 kHz, or when board length is the binding constraint. In those cases route the gate return separately from the power source and you recover much of what the fourth pin would have given you.

Before release: pull the specific outline drawing and revision into your library rather than a generic four-lead footprint, confirm any gate-source capacitor carried over from an older design is intended for a Kelvin package, and measure per-event switching energy on the board you will ship rather than trusting a datasheet subtraction.

Internal links to place

• [INTERNAL LINK: TO-247-3 package outline and dimensions -> three-lead TO-247 reference]
• [INTERNAL LINK: SiC gate driver design and negative bias -> gate drive for silicon carbide MOSFETs]
• [INTERNAL LINK: measuring switching losses with a double pulse test -> power device characterization]
• [INTERNAL LINK: paralleling power MOSFETs for current sharing -> multi-device layout]
• [INTERNAL LINK: creepage and clearance for power converters -> IEC 60664-1 insulation coordination]