Reading a GPU VRM Power Stage as a Shopping List

One VRM phase is six passive positions, and only one of them is hard to buy. Input bulk, high-frequency decoupling, bootstrap, gate resistor, feedback divider, current sense. In Walsin Technology (Walsin) data the current sense position is the narrow one: of 10,428 thick film low-ohm current sense part numbers, exactly two go below 0.010 Ω.

---

Venatech Limited, an authorised Walsin distributor in Hong Kong, holds Walsin’s full parametric export — 203 files, 359,135 distinct part numbers — and 28 Walsin datasheets. This post is written from those files. Every part number is real and every count is of part numbers, not pieces.

Venatech holds no system vendor’s bill of materials. Nothing here is an NVIDIA, GB200 or other platform BOM, and we have no approval from any system vendor. This is a generic multiphase buck power stage, read as a procurement problem.

What are the six passive positions in one VRM phase?

Six, and only one of them is hard to buy. Strip a synchronous buck phase down and the passive count per phase is small: input bulk, high-frequency decoupling, bootstrap, gate resistor, feedback divider, current sense. The quantity is large because the phase count is large, not because each phase is complicated.

Position	What it is doing	What decides the choice
Input bulk	Holds the input rail up through the switching transient	Capacitance under DC bias, ripple current, height
High-frequency decoupling	Supplies charge the bulk cap is too slow to deliver	Case size and loop inductance, not capacitance
Bootstrap	Floats the high-side gate drive supply	Rated voltage margin and dielectric stability
Gate resistor	Damps the gate loop and sets switching edge rate	Low ohms, tolerance is loose, pulse power is not
Feedback divider	Sets the output voltage the controller regulates to	Ratio accuracy and temperature coefficient
Current sense	Reports phase current for balancing and limiting	Absolute resistance, TCR, Kelvin connection

Five of the six are commodity positions you can fill from any broad catalogue. The sixth is not.

Which position actually decides whether the phase regulates?

Current sense, because it is the only position whose error becomes a current error directly. A divider error shifts the output voltage by a known ratio and can be trimmed. A sense resistor that drifts with temperature mis-reports phase current, and in a multiphase converter that mis-balances the phases — the hottest phase reads low, takes more current, and gets hotter.

That is why the specification to read is not the resistance. It is the temperature coefficient, and the resistance range the manufacturer will actually hold it over.

How far down in resistance does Walsin’s thick film current sense line go?

To 0.010 Ω, for all practical purposes. Measured across all 10,428 rows of Thick_Film_Low_Ohm_Current_Sense_Chip_Resistor__1of1.xlsx, parsing the Resistance column:

Resistance band	Thick film part numbers	Best TCR offered in that band
Below 0.010 Ω	2	≦±200 ppm/°C
0.010 to 0.049 Ω	846	≦±100 ppm/°C, on 26 of them
0.050 to 0.099 Ω	817	≦±100 ppm/°C, on 347 of them
0.100 to 0.999 Ω	8,763	≦±100 ppm/°C, on 1,128 of them

Two part numbers out of 10,428 go below 0.010 Ω, and no thick film row anywhere in the file is better than ≦±100 ppm/°C. We have not seen that counted anywhere. The metal plate line is a different instrument entirely: 1,188 part numbers, resistance down to 0.0005 Ω — twenty times lower — and 612 of those 1,188 rows hold ≦±50 ppm/°C, a grade the thick film line does not reach at any resistance.

At 0402 the contrast is stark enough to settle the argument on one line:

Part number	Construction	Resistance	Tolerance	TCR	Rated power	Match / mismatch
WW04DR005FTL	Metal plate	0.005 Ω	±1%	≦±100 ppm/°C	1/3 W	Reaches 5 mΩ at 0402
WW04CR025FTL	Thick film	0.025 Ω	±1%	0 to +350 ppm/°C	1/8 W	Mismatch — five times the resistance, a one-sided TCR band, under half the power
WW06RR005FTL	Metal plate	0.005 Ω	±1%	≦±70 ppm/°C	1/3 W	Mismatch — 0603 footprint, tighter TCR
WW12XR010FTL	Thick film	0.010 Ω	±1%	≦±2100 ppm/°C	1/4 W	Mismatch — 1206 footprint, TCR 21 times wider than the 0402 metal part
WW25MR003FTL	Metal plate	0.003 Ω	±1%	≦±50 ppm/°C	1 W	Mismatch — 2512 footprint, lowest TCR grade

WW04CR025FTL and WW04DR005FTL are the same case size and the same ±1% tolerance. One is a current sense resistor you can balance phases with; the other is not, and the only field that says so is TCR.

Why is thick film TCR so much worse at low ohms?

Because at a few milliohms the resistive element is a thin, short film and the termination metal is a significant fraction of the total resistance. Termination metal has its own, much larger temperature coefficient, so as the designed resistance falls the termination’s contribution dominates. Walsin’s own export shows the effect as a gradient: in the 0.100 to 0.999 Ω band no row is worse than ≦±800 ppm/°C, while in the 0.010 to 0.049 Ω band 174 rows carry ≦±2100 ppm/°C and 122 carry ≦±1500 ppm/°C.

A metal plate part moves the element to a bulk alloy strip and brings the Kelvin connection out to four terminals, which is why it holds ≦±50 ppm/°C where thick film cannot.

Beyond Walsin, Stackpole current sense resistors are a line Venatech supplies, and they are where we go when a design needs a four-terminal part below 1 mΩ. Venatech is an authorised distributor for Walsin Technology; Stackpole is a line we supply, and we do not describe it as an authorised franchise.

Can I pick a feedback divider from the parametric export?

Not on tolerance, and this is worth knowing before you filter. Walsin’s WF thin film precision export states ±0.01% on every single row of all seven non-empty case-size files — 52,808 rows from 0402 to 2512, with no other tolerance value present anywhere in them. The 0201 file, by contrast, carries three tolerances across its 5,506 rows: ±0.1%, ±0.5% and ±1%.

What the TCR column does let you do is pick on the specification that matters for a divider. For a ratio, the two resistors’ TCRs track each other only if you buy them from the same TCR grade: WF04Z1002TTL is 10 kΩ at ≦±5 ppm/°C and WF04B1001TTL is 1 kΩ at ≦±3 ppm/°C, both 0402, both 1/10 W. Mixing a ≦±5 part with a ≦±50 part gives you a ratio that drifts even though both parts are “precision”.

Viking precision thin film is also a line Venatech supplies and is the other route to a matched divider pair. The same wording applies: a line we supply, not an authorised franchise.

What does the shopping list look like with real part numbers?

Six positions, eight real Walsin part numbers, one generic 12 V to 0.9 V phase. The mismatch column is the part of the table to read.

Position	Part number	Key specification	Match / mismatch against the position
Input bulk	1206B226K100	22 µF, X7R, 10 Vdc, 1206, T 1.60+0.30/−0.10 mm	Mismatch on height — 1.60 mm nominal with a +0.30 mm band; check your keep-out
Input bulk, lower profile	0805X106K6R3	10 µF, X5R, 6.3 Vdc, 0805	Mismatch on voltage — 6.3 V is no margin on a 12 V rail; a 5 V or 3.3 V intermediate rail only
HF decoupling	0402B104K160	100 nF, X7R, 16 Vdc, 0402, T 0.50±0.05 mm	Match — the tight ±0.05 mm thickness band
Bootstrap	0402B104K250	100 nF, X7R, 25 Vdc, 0402	Match — 25 V gives margin above a 12 V bootstrap swing
Bootstrap, automotive	MT15B104K250	100 nF, X7R, 25 Vdc, 0402, AEC-Q200 compliant	Mismatch — same electricals, different grade and lead time
Gate resistor	WR04W2R20FTL	2.20 Ω, ±1%, TCR −200 to +400 ppm/°C, 1/16 W	Mismatch on TCR — irrelevant here; a gate resistor does not need TCR
Feedback divider, top	WF04Z1002TTL	10 kΩ, 0402, TCR ≦±5 ppm/°C, 1/10 W	Match — buy the pair from one TCR grade
Current sense	WW04DR005FTL	0.005 Ω, ±1%, TCR ≦±100 ppm/°C, 1/3 W, metal plate	Mismatch on TCR grade — the 2512 metal part reaches ≦±50 ppm/°C if you have the area

Note what the gate resistor row is doing. Its TCR is the loosest number in the table and it does not matter, because nothing downstream reads that resistance. The sense resistor’s TCR is the tightest and it decides phase balance. The same column means two different things in two rows, and a filter cannot know that.

Confirm every one of these against both datasheets and qualify them in your own circuit. We will not tell you a part fits; we will show you where it differs.

For the construction difference behind the sense-resistor table, see Low-ohm current sense: thick film vs metal. For the thin-film side, see Thin film vs thick film chip resistors. The full range is on the Walsin chip resistor range page.

What can Venatech add that a selector cannot?

We read the columns before we trust them. The WF tolerance finding above is an example: a selector filtering on that field would hand you 52,808 parts at ±0.01% and none of them confirmed. We hold the 28 datasheets and will send the one covering your part number, and where we hold none — and we hold none for the WF thin film line, none for inductors, MOV, disc capacitors, RF or antennas — we say so and point you at Walsin’s own site rather than guessing.

We also hold a dated Hong Kong stock snapshot, as at 23 September 2026, covering 1,125 part numbers, so you can see which of the eight positions above is a stock pull and which is a production slot.

The short version

A VRM phase is six passive positions and five of them are commodities. The current sense position is the one that is genuinely narrow: 2 of Walsin’s 10,428 thick film low-ohm part numbers go below 0.010 Ω, no thick film row beats ≦±100 ppm/°C, and the metal plate line reaches 0.0005 Ω with 612 of 1,188 rows at ≦±50 ppm/°C. Pick the sense resistor on TCR and construction, buy divider resistors from a single TCR grade, and do not filter a divider on the export’s tolerance column until Walsin confirms it. Qualification stays with you.

Send us the passive lines from one power stage — or paste them into the free cross-reference tool — and we will come back with Walsin candidates field by field, differences stated, plus which of them was physically in Hong Kong on 23 September 2026. Use the cross-reference tool for the list and ready stock for what is on the shelf.

---

Frequently asked questions

---