Bottom line first: liquid cooling does not simply add a heat spreader to a PCB — it moves the whole board from an air environment into a chemical one. Cold-plate and immersion systems ask completely different things of the laminate: the first mainly tests thermal cycling and high-current carrying, the second tests the long-term compatibility of material and coolant. Buyers and engineers who keep selecting boards by air-cooled habit are most likely to fail the 1000-hour reliability test — at which point mass production is already imminent.
Two liquid-cooling routes, two different board requirements
Cold plate: the PCB is still dry. A cold plate is bonded to high-power devices through a thermal interface material, so the board itself never touches coolant. The effect concentrates in three places. Heat flux: chip-level hot spots can exceed 100 W/cm², creating a large gradient between the area near the cold plate and the far end, which raises board warpage and solder joint stress. High current: feeding those devices usually pushes copper into the 3~6oz thick-copper range. Thermal cycling: on-off operation produces −40~125 °C cycles that keep working on barrel copper and pads.
Immersion (single or two-phase): the whole board goes into the medium. A board immersed long term in a dielectric coolant — fluorinated liquid or synthetic hydrocarbon — meets problems that did not exist in air: plasticiser and residual monomer leaching from the resin, swelling of the laminate interface as it absorbs liquid, loss of coating adhesion, and the long-term insulation effect of fluid decomposition products. Two-phase immersion adds the large temperature swing of boiling and condensation, which is harsher on thermal expansion matching.
Material and test: put "immersed or not" into the first requirement
- Moisture absorption and CAF resistance: conductive anodic filament is the main fear in immersion and high humidity. Prefer laminates with moisture absorption <0.15%, high resin content and tight glass-to-resin bonding; keep interlayer dielectric thickness at ≥0.10 mm to lengthen the ionic migration path.
- Dielectric compatibility soak test: do not rely on the laminate TDS alone. Ask the supplier for compatibility data with the intended coolant — Tg, Td, Dk/Df and volume resistivity before and after 1000 h of immersion (IPC-TM-650 2.5.5.9 for surface insulation resistance and 2.6.2.1 for moisture absorption are useful references); a rise in conductivity is a direct signal of leaching.
- Thermal cycling: for two-phase immersion, cycle −40~125 °C for 1000 cycles and then section, looking for barrel copper cracks and copper separation; cold-plate boards can start at 500 cycles.
- Thick copper and high layer counts: power delivery and shielding push stackup and copper weight up together. AI server carrier boards in the industry have reached 24~32 layers, with thick copper and high-current design in parallel, which stresses both lamination cycles and plating uniformity.
Executable advice for purchasing and negotiation
- Stage the roadmap: cold plate remains the mainstream near term — mature, compatible with existing architectures, cheap to retrofit — while immersion ramps in very high power density cabinets (about >100 kW per cabinet). Run a two-stage path of cold-plate compatible, then immersion ready: meet cold-plate demand on current laminates while qualifying immersion materials in small volume, rather than betting everything at once.
- Put reliability into the technical agreement: list CAF, cycle count, soak duration and the permitted drift of parameters before and after as acceptance items for incoming material and first article. Otherwise, if the soak test fails, responsibility and re-design cost are very hard to assign.
- Ask for data, not verbal assurance: request material-to-coolant compatibility reports, CAF test records and thermal cycling sections, and fold them into supplier evaluation; a supplier that offers only a catalogue goes on the watch list.
- Lot traceability and test budget: liquid-cooled boards mostly serve data centres and need traceable lots. More test items raise unit price and lead time, so reliability testing and sample cycles should be quoted as separate lines rather than added afterwards.
Direction and pace
Rising liquid-cooling penetration is a passive consequence of rising per-cabinet power density, not a technical choice made by the PCB industry. It affects procurement along three lines. Materials migrate from general-purpose FR-4 to low-absorption, high-Tg, CAF-resistant laminates, and pricing power shifts toward the high-end suppliers. Process runs thick copper and high layer counts in parallel, raising difficulty and yield risk. Acceptance gains test items, delivery cycles stretch, and pure price comparison becomes steadily less workable.
The practical response is to give liquid-cooled products their own price and lead-time baseline in the annual framework, and to make "is the material compatibility data complete" an entry condition. Raise large-format, high-layer and thick-copper demand early as well: Shenzhen Lingxin Technology's capability for AI compute covers 1–44 layers, finished copper up to 13oz (18oz outer base copper maximum) and a maximum size of 610×1100 mm, with lead times as fast as 24-48 hours, actual timing depending on product complexity; send stackup and Gerber to info@ipcba.com.cn for engineering and quotation review, and write the coolant, soak duration, CAF threshold and cycle count into the RFQ so the acceptance basis is fixed before tooling.
Send your Gerber files and stackup requirements; a 24-hour expedite channel is available.