Passives for AI Server Power Stages: A Selection Checklist

Only 20 of the 146 Walsin Technology inductors that publish a saturation current have a DCR value that is physically possible for their package size. The other 126 would dissipate between roughly 2 W and 655 W in a 4 mm or 6 mm chip inductor at rated current — not achievable — which means their DCR column is very likely in milliohms mislabelled as ohms. This checklist works through inductors, current-sense resistors and bulk MLCC for an AI server power stage using only the parts that check out.

---

Venatech Limited, an authorised Walsin distributor in Hong Kong, holds Walsin Technology (Walsin) full inductor, chip resistor and MLCC parametric exports. Nothing here is a datasheet claim for the inductor line — Walsin publishes no inductor datasheet PDF at all, so every inductor figure below comes from the parametric export alone and is flagged accordingly. Every part number is real; every count is of part numbers, not pieces.

How many Walsin inductors publish a saturation current at all?

146 of 5,410 real inductor part numbers (excluding the Test123999 QA row). The other 5,264 publish rated current only, which tells you the thermal limit, not the point at which the core starts to lose inductance under load — the number a buck converter design actually needs. All 146 sit in three series: CSME (89), CSMEA (53) and MIP-H (4).

Series	Parts with saturation current	Package
CSME	89	4040, 5050, 6060
CSMEA	53	4040, 5050, 6060
MIP-H	4	1608(0603), 2012(0805)

Of those, how many have a DCR you can actually trust?

20. We ran a basic plausibility check on all 146: multiply the published rated current by itself and by the published DCR, and see whether the result is a power figure that a chip inductor of that package size could physically dissipate. 126 of the 146 fail by a wide margin — the worst case is a 6060-package CSME part rated 6 A against a stated 18.2 Ω DCR, which works out to 655 W in a 6 mm body.

Series / package	Example rating	Stated DCR	Implied power at rated current	Match / mismatch
CSME, 6060 package	1.0 µH, 6 A	18.2 Ω	655 W	Mismatch — physically impossible
CSME, 4040 package	1.0 µH, 3.2 A	27 Ω	276 W	Mismatch — physically impossible
CSMEA, 5050 package	1.0 µH, 4.9 A	13 Ω	312 W	Mismatch — physically impossible
CSME prefix family, 4040 package, plausible subset	1.0 µH, 5.8 A	0.028 Ω	0.94 W	Match — plausible
MIP2012H2R2MBE (MIP-H, 2012/0805)	2.2 µH, 900 mA	0.17 Ω	0.14 W	Match — plausible

Of the 20 plausible parts, 16 share a part-number prefix one character different from the other 130 CSME/CSMEA rows, even though the Series column still reads “CSME Series” for all of them. The part-number prefix, not the Series field, is what actually separates the trustworthy rows from the rest in this series.

Is the implausible DCR confined to the saturation-current subset?

No — it runs wider. Across the whole inductor export, 4,998 rows publish both a numeric rated current and a numeric DCR. 173 of those 4,998 describe a power dissipation above 5 W in a package no larger than 6 mm square, which is not achievable. The 173 span four series, not just the two above:

Series	Rows failing the plausibility check
CSME	73
CSMEA	53
WLFI	30
WQCW	17

The WLFI and WQCW failures are a different shape from CSME and CSMEA. Dividing their DCR by 1,000 brings most rows under 2.6 W, but the WQCW series also publishes rated currents of 50 A and 80 A on 0402 and 0603 packages — WQCW1608Z0JR47PB at 80 A, 0603(1608) — which stays implausible on current alone no matter what unit the DCR is in.

Which inductor passes both checks for a multiphase buck?

Within the 14-part plausible family at the 4040 package size, inductance runs 0.24 µH to 47 µH with saturation current from 0.7 A to 14 A:

Inductance	DCR	Rated current	Saturation current	Match / mismatch
0.24 µH	0.017 Ω	6 A	14 A	Match — high-current, low-inductance phase
1.0 µH	0.028 Ω	5.8 A	8.7 A	Match — common buck inductance band
4.7 µH	0.115 Ω	2.85 A	4 A	Mismatch on current — too low for a main VRM phase

None of these reach the 20–60 A per phase a CPU or GPU VRM commonly runs. Within what Walsin publishes with a confirmed saturation current, the ceiling in this family is 14 A at 0.24 µH. For higher per-phase current you are back to rated-current-only parts — see Choosing an inductor for a buck converter and Rated current vs saturation current in chip inductors for how to size against rated current alone when saturation current is not published.

What covers low-ohm current sensing on the 12V or 48V rail?

Metal plate current-sense resistors publish a materially tighter TCR than thick film at the same resistance and case size, which matters for a sense resistor feeding a current-mode controller:

Part number	Technology	Case	Resistance	TCR	Rated power	Match / mismatch
WW12XR010FTL	Thick film	1206(3216)	0.010 Ω	≦±2100	1/4 W	Baseline
WW25MR001FTL	Metal plate	2512(6432)	0.001 Ω	≦±70	1W	Mismatch on TCR — 30× tighter
WW25AR0L5FTL	Metal plate	2512(6432)	0.0005 Ω	≦±70	3W	Mismatch on resistance and power — lowest ohm value in the line

0.0005 Ω at 3 W gives a lower burden voltage at high rail current than any thick film part in the catalogue, at the cost of the larger 2512 footprint. For the full technology trade-off, see Low-ohm current sense: thick film vs metal.

Which MLCC give bulk capacitance at the point of load?

49 General Purpose MLCC parts reach 47 µF or more. At 0805 and 1206 — a smaller footprint than the 1210 parts already covered for AI server power supplies in our North America post — X5R and X6S dielectrics both reach 100 µF, but not on the same voltage:

Part number	Case	Dielectric	Capacitance	Voltage	L × W × T (mm)	Match / mismatch
0805X107M4R0	0805	X5R	100 µF	4 Vdc	2.00×1.25×1.25	Baseline
0805S107M4R0	0805	X6S	100 µF	4 Vdc	2.00×1.25×1.25	Match on footprint and rating
0805X476M6R3	0805	X5R	47 µF	6.3 Vdc	2.00×1.25×1.25	Mismatch on capacitance — half the capacitance for 60% more voltage headroom

X6S holds its capacitance better under DC bias than X5R at the same rated voltage — see MLCC DC bias and voltage derating before committing to a case size at the point of load. For the larger-case, higher-voltage side of an AI server power supply, see Walsin Passive Components for North America. The full MLCC range is on the MLCC product page.

The short version

Saturation current exists on only 146 of 5,410 Walsin inductors, and of those, only 20 have a DCR figure that survives a basic physics check. Treat any CSME, CSMEA, WLFI or WQCW DCR above roughly 1 Ω, or any WQCW/WLFI rated current above 10 A on a 0402 or 0603 package, as unconfirmed until Walsin states the units in writing. For current sensing, metal plate parts clear a 30×-tighter TCR than thick film at the same resistance. For bulk capacitance, 0805 and 1206 General Purpose MLCC reach 100 µF in the same footprint at two different dielectrics.

Send us your power-stage BOM and we will check which Walsin inductor, sense resistor and MLCC lines actually have confirmed data behind them, line by line. Or give us one incumbent part number and ask us to cross it.

Check it against the catalogue yourself. Cross-reference accepts one part number or a whole BOM column and prints the mismatches as plainly as the matches — qualification stays with you, which is the honest division of labour. Ready stock covers the ceramic capacitor lines held in Hong Kong.

---

Frequently asked questions

<!– wp:rank-math/faq-block {"questions":[{"id":"faq-question-01","title":"How many Walsin inductors publish a saturation current?","content":"146 of 5,410 real part numbers, across the CSME (89), CSMEA (53) and MIP-H (4) series. The remaining 5,264 publish rated current only, which is a thermal limit, not the saturation point a buck converter design needs.","visible":true},{"id":"faq-question-02","title":"Why might Walsin's CSME and CSMEA DCR figures be wrong?","content":"126 of the 146 inductors with a published saturation current show a DCR that would mean dissipating between roughly 2 W and 655 W at rated current in a 4 mm or 6 mm package, which is not physically possible. Dividing those DCR values by 1,000 brings every one back under 1.2 W, consistent with a milliohm figure recorded as ohms. This is Venatech's own measurement and has not been confirmed by Walsin.","visible":true},{"id":"faq-question-03","title":"Does the DCR problem affect any inductor series besides CSME and CSMEA?","content":"Yes. Across the 4,998 inductor rows that publish both a rated current and a DCR, 173 describe an implausible power dissipation, split across CSME (73), CSMEA (53), WLFI (30) and WQCW (17). The WQCW series also publishes rated currents of 50 A and 80 A on 0402 and 0603 packages, which stays implausible regardless of the DCR unit.","visible":true},{"id":"faq-question-04","title":"Which Walsin inductor family has both a trustworthy DCR and a published saturation current?","content":"The 14-part plausible family sharing one distinct part-number prefix, all in the 4040 package, inductance 0.24 µH to 47 µH, saturation current 0.7 A to 14 A. The Series column still labels them \"CSME Series,\" so filtering on Series alone will not separate them from the 126 flagged rows.","visible":true},{"id":"faq-question-05","title":"What is the difference between thick film and metal plate current-sense resistors?","content":"Metal plate parts publish a TCR around ±50 to ±75 ppm/°C against roughly ±2100 ppm/°C for thick film at the same case size, and reach lower resistance values — down to 0.0005 Ω at 2512 — at the cost of a larger footprint for the lowest values.","visible":true},{"id":"faq-question-06","title":"Can I get 100 µF in an 0805 case size from Walsin?","content":"Yes. 0805X107M4R0 (X5R) and 0805S107M4R0 (X6S) both publish 100 µF at 4 Vdc in the same 2.00 × 1.25 × 1.25 mm body. Voltage headroom above 4 Vdc drops the available capacitance in the same footprint — 47 µF at 6.3 Vdc on the X5R dielectric.”,”visible”:true},{“id”:”faq-question-07″,”title”:”Does Venatech hold a datasheet for Walsin’s inductor line?”,”content”:”No. Venatech holds no inductor datasheet PDF at all. Every inductor specification in this post comes from Walsin’s parametric export, and every uncertain figure is flagged for direct confirmation with Walsin.”,”visible”:true}],”listStyle”:””,”titleWrapper”:”h3″,”titleCssClasses”:””,”contentCssClasses”:””,”sizeSlug”:””} –>

---