Portuguese
High-Conductivity Thermal Pads for AI Servers: When Does 16 W/m·K Really Make a Difference?
The cold plate is torqued to the drawing. The new pad is rated at 16 W/m·K instead of the 8 W/m·K grade used in the previous build. Yet the difference between coolant inlet temperature and GPU hotspot temperature barely changes. The immediate conclusion is often that the higher-grade pad was oversold—usually before anyone checks whether the interface was thin enough, flat enough, and loaded hard enough for the additional conductivity to matter.
That is the central trap in selecting high-conductivity thermal pads for AI servers. Moving from roughly 6 to 12 W/m·K often produces a clear improvement because bulk resistance still represents a meaningful part of the thermal path. Moving from 12 to 16 W/m·K, or into the 15–18 W/m·K range, is different. The higher number creates value only when the mechanical interface allows it to do so.
The stakes are significant in an AI server. NVIDIA rates the H100 SXM module at up to 700 W. For a simplified 700 W accelerator example using a 60 × 60 mm interface area, interface resistance translates into temperature rise as follows:
Q = heat load in watts
R″ = area-normalized interface resistance in °C·in²/W
A = interface area in in²
A 60 × 60 mm area is approximately 5.6 in². At 700 W, every 0.01 °C·in²/W of interface resistance represents about 1.25 °C of temperature rise. An interface measuring 0.23 °C·in²/W therefore contributes close to 29 °C before the cold plate, coolant loop, and facility water conditions are considered.
This is why thermal-pad selection can materially affect an AI accelerator while making little practical difference in a low-power device. It is also why one thermal-pad grade should not automatically be applied to every component in a server. Accelerator packages, CPUs, M.2 SSD controllers, optical transceiver cages, and VRM stages can all require gap-filling materials, but their heat flux, pressure limits, electrical requirements, gap tolerances, and contamination risks are different.
Where Each Conductivity Range Usually Fits
| Conductivity band | Typical server applications | Main advantage | Main trade-off |
|---|---|---|---|
| 1–5 W/m·K | SSD controllers, optical cages, VRM stages, and low-power NICs | Soft, economical, and tolerant of stack-up variation | Bulk resistance can dominate when the gap becomes thick or heat flux rises |
| 6–12 W/m·K | CPU lids, DIMMs, and mid-power accelerators | Broad compression window and relatively straightforward assembly | The system still needs suitable pressure and bond-line control to realize the conductivity benefit |
| 15–18 W/m·K | Accelerator packages under cold plates, HBM stacks, and other high-heat-flux interfaces | Lower bulk resistance at a controlled bond line | Higher cost and potentially narrower mechanical and tolerance windows |
Commodity thermal pads commonly fall between 1 and 8 W/m·K. The 15–18 W/m·K range sits near the top of the silicone-pad ladder. Beyond that point, the material class often changes rather than simply advancing to another pad grade. In-plane graphite, carbon-fiber materials, and liquid metal have different electrical, directional, assembly, and reliability characteristics.
Phase-change material is not simply the next conductivity step either. ZIITEK’s TIC®800 phase-change series is listed from 0.95 to 9.6 W/m·K, but its value comes primarily from forming a thin joint of approximately 25–75 µm. In other words, the question “What material has the highest thermal conductivity?” cannot be answered with one number. The relevant questions are conductivity direction, bond-line thickness, contact behavior, electrical properties, and material format.
What a High-Conductivity Pad Actually Changes
The total interface resistance contains three terms in series:
- Contact resistance from the first surface into the pad
- Bulk resistance through the pad
- Contact resistance from the pad into the second surface
The conductivity value on the datasheet directly controls only the bulk term:
BLT = bond-line thickness in mm
k = through-plane thermal conductivity in W/(m·K)
R″ = area-normalized resistance in °C·in²/W
The conversion factor of 1.55 is required because 1 mm divided by 1 W/(m·K) equals approximately 1.55 °C·in²/W. For example, the calculated bulk resistance of a 0.25 mm bond line at 16 W/m·K is:
For the following comparison, the two contact terms are combined into a fixed 0.10 °C·in²/W. This is a first-order calculation rather than measured assembly data, but it shows how conductivity and bond-line thickness interact.
| Bond line | 6 W/m·K | 12 W/m·K | 16 W/m·K | Resistance reduction, 12 → 16 |
|---|---|---|---|---|
| 0.25 mm | 0.165 | 0.132 | 0.124 | 6.1% |
| 0.50 mm | 0.229 | 0.165 | 0.148 | 9.8% |
| 1.00 mm | 0.358 | 0.229 | 0.197 | 14.1% |
| 1.50 mm | 0.488 | 0.294 | 0.245 | 16.5% |
All resistance values are in °C·in²/W and assume combined contact resistance of 0.10 °C·in²/W.
How to interpret the calculation
At a 0.50 mm bond line, moving from 12 to 16 W/m·K reduces the calculated total interface resistance by roughly 10%. In the simplified 700 W, 60 × 60 mm example, that corresponds to approximately 2 °C.
At a 1.50 mm bond line, the same conductivity change is worth roughly 6 °C because bulk resistance represents a larger share of the joint. The higher-conductivity pad therefore shows a larger numerical benefit in a thicker interface.
However, reducing a 0.50 mm bond line by 0.20 mm at constant conductivity can be worth about 3 °C in the same example, provided the mechanical design permits it. This leads to an important design rule:
On relatively thin interfaces, improving bond-line control and contact quality may be more effective than buying a higher conductivity grade. On thicker interfaces, conductivity becomes a stronger lever.
A higher W/m·K rating is therefore not automatically better. It is valuable when bulk resistance is significant and the assembly provides adequate pressure and contact. If contact resistance dominates, the additional filler and cost may not produce a meaningful junction-temperature reduction.
Step 1: Read the Datasheet in the Right Order
Many selection errors occur before samples are tested because the datasheet is read from the headline conductivity downward. For an AI-server application, use the following order instead.
1. Conductivity and test method
Confirm that the quoted value is through-plane conductivity and identify the test method. ASTM D5470 covers thermal transmission measurements for thermally conductive electrical-insulation materials. A measured value and a value inferred from filler loading are not directly comparable, so request the method behind the figure.
2. Thermal impedance at a stated pressure and thickness
Thermal impedance is useful only when the test pressure, sample thickness, and compression condition are known. A value measured at 50 psi does not predict performance in an assembly that can retain only 15 psi.
3. Available thicknesses and tolerance
Depending on the material family, pads may be available from approximately 0.25 mm to 12 mm. Thickness tolerance is often given as a percentage. A ±10% tolerance on a 1 mm pad equals ±0.1 mm, which may be comparable to the intended compression allowance.
4. Hardness and compression behavior
Review Shore OO hardness according to ASTM D2240, but do not use hardness alone to predict assembly force. The deflection-versus-force or compression-versus-pressure curve is more useful because it shows whether the actual retention system can close the gap without overloading the package.
Conductivity direction must also be verified. Graphite and carbon-fiber materials may conduct heat much more effectively in-plane than through-plane. Carbon-fiber grades may also be electrically conductive. They should not be treated as direct substitutes for electrically insulating silicone pads.
Step 2: Match Hardness and Compression to Retention Force
Achieving 15–18 W/m·K generally requires a high loading of thermally conductive filler in the silicone matrix. As filler volume increases, modulus can also increase unless the formulation is specifically engineered to remain soft. A harder pad compresses less under the same force, potentially leaving a thicker effective bond line and returning part of the thermal benefit that the higher conductivity was intended to provide.
Three values define the usable mechanical window:
- Compression at assembly: server pads are commonly designed around approximately 10–30% compression, but the correct target must come from the product data and the specific assembly.
- Available force: calculate pressure over the complete pad area. A 60 × 60 mm pad requires approximately 500 N at 20 psi and 250 N at 10 psi.
- Stack-up tolerance: include component height, PCB bow, cold-plate flatness, standoff tolerance, and pad thickness tolerance.
A pad must work at the minimum and maximum assembled gap, not only at nominal dimensions.
Two high-conductivity examples with different mechanical priorities
| Product | Quoted conductivity | Relevant characteristics | Application emphasis |
|---|---|---|---|
| TIF®800SE | 15.0 W/m·K, ASTM D5470 | Shore OO 35; listed in 0.75–5.00 mm thicknesses | High-conductivity gap filling where the retention system can support the material’s mechanical behavior |
| TIF®800TU | 16.0 W/m·K | Very soft, gel-like formulation | HBM and optical assemblies where package loading from a stiff pad is a concern |
TIF®800TU is the highest grade ZIITEK quotes against D5470 in its current product ladder. Other suppliers offer grades rated at 17–18 W/m·K. When comparing them, review the underlying test records side by side: use the same method, through-plane direction, pressure, thickness, and conditioning wherever possible.
A customer reference associated with the TIF®800TU product reports that an AI accelerator HBM interface using the 16 W/m·K pad at 3 mm and 20 psi measured 5–8 °C cooler than the same interface using an ordinary 8 W/m·K pad. This is a customer-reported result under those stated conditions, not a guaranteed temperature reduction for other assemblies.
Step 3: Verify the Electrical, Flame, and Contamination Envelope
A server thermal pad often crosses electrically active components. Electrical insulation is therefore a qualification gate rather than a secondary datasheet item.
Electrical insulation
Check breakdown strength per unit thickness, volume resistivity, and dielectric constant where the interface is close to high-speed signals. ZIITEK’s 15–16 W/m·K TIF®800-series silicone pads are quoted at ≥5500 V/mm according to ASTM D149.
Evaluate insulation at the thinnest assembled bond line, not at nominal pad thickness. A practical design approach is to maintain at least a 2× margin over the highest working voltage and then account for thickness tolerance and other system-level derating requirements. The margin may be generous for a 48 V system but requires more careful analysis around a 400 V DC bus. In higher-voltage applications, an insulating carrier film may need to become part of the qualified stack.
Flame rating
UL 94 V-0 is commonly required for materials used inside server chassis. Where certification is mandatory, the rating should be traceable to the relevant certification record rather than accepted solely as a brochure statement.
Siloxane and outgassing control
Standard silicone pads can release low-molecular-weight siloxanes. This may be acceptable around many power components but can create problems near optical-module facets and connectors. For interfaces close to optics, specify a low-volatility or non-silicone material and request applicable outgassing data.
Step 4: Lock the Assembly Parameters
A high-conductivity material cannot compensate for an uncontrolled assembly. The following parameters should be included in the mechanical drawing or assembly work instruction.
- Pressure and distribution: 15–30 psi is a common working band for server pads, but the product-specific compression curve must govern the final value. Uniformity is as important as average pressure.
- Flatness and parallelism: local low spots create contact-resistance gradients. Cold-plate construction and manufacturing process must therefore be reflected in the flatness specification.
- Gap control: use standoffs or hard stops to define the minimum bond line rather than allowing torque repeatability and pad stiffness to set the final gap.
- Torque sequence: tighten in a cross-pattern and in stages to the specified value to reduce cold-plate tilt and uneven loading.
- Rework procedure: replace pads that have taken a compression set. Treat the pad as a consumable and define the surface-cleaning process before installing the replacement.
A single central fastener can bow a cold plate and reduce pressure at the perimeter. Likewise, a nominally correct torque does not guarantee the correct interface pressure if the standoffs, fasteners, springs, or plate stiffness change. Qualification should therefore evaluate the complete retention system, not the pad in isolation.
Step 5: Define Qualification and Acceptance Criteria
Acceptance limits should be written before samples are approved. Testing only the material at time zero can miss compression set, migration, binder degradation, or resistance drift that appears during aging and thermal cycling.
| Test | Condition to agree | What it evaluates | Example acceptance approach |
|---|---|---|---|
| High-temperature aging | 150 °C for 1,000 h | Binder degradation and oil loss | Interface-resistance rise ≤25%; no cracking |
| Damp heat | 85 °C / 85% RH for 1,000 h | Moisture-related degradation and bleed | Resistance rise ≤25%; no delamination |
| Thermal cycling | −40 to 125 °C; cycle count and dwell to be agreed | Migration and CTE-driven interface changes | No uncontrolled or accelerating resistance increase |
| Compression set | 25% deflection at 70 °C for 24 h | Permanent deformation and loss of contact force | Set within the agreed limit; required force retained |
| Incoming inspection | Thickness, hardness, and density by lot | Supplier or process drift | Results within approved datasheet and drawing tolerances |
ZIITEK publishes 150 °C/1,000 h aging, 85 °C/85% RH for 1,000 h, and −40 to 125 °C cycling conditions for its 800-series grades. These conditions illustrate the level of detail that should be agreed with a supplier. The actual acceptance limits must come from the system thermal budget.
For example, a 25% resistance increase has different consequences for an interface beginning at 0.15 °C·in²/W than for one beginning at 0.30 °C·in²/W. Percentage change alone is not enough; translate it into junction-temperature margin at the intended heat load.
Where practical, measure interface resistance or junction temperature during thermal cycling at fixed power and ambient conditions. End-point measurements can hide the trend. A steadily increasing temperature or resistance trace is often more informative than a single post-test result.
When to Use Paste, Phase-Change Material, Gap Filler, or a Pad
Thermal pads are strongest in applications with a measurable gap, reasonably uniform geometry, electrical-insulation requirements, field-service needs, and enough clamping force to reach the intended compression. Outside this window, another thermal-interface format may perform better.
| Material format | Typical bond line | Force requirement | Rework | Where it fits in a server |
|---|---|---|---|---|
| Thermal paste or grease | 25–100 µm | High clamp | Possible but messy | Bare die or IHS-to-cold-plate interfaces requiring very low resistance |
| Phase-change material | 25–75 µm | Moderate clamp | Moderate | High-heat-flux packages where pump-out resistance is preferred over grease |
| Dispensed gap filler | 0.1–3 mm | Near zero | Difficult | Variable gaps, limited clamping hardware, and automated dispensing |
| Thermal pad | 0.25–12 mm | Moderate and uniform | Easy | Uniform gaps requiring insulation, clean handling, or field service |
Pump-out in grease-based joints is driven by mechanical cycling. Differences in thermal expansion cause the interface to open and close, gradually moving material away from the active area. A solid pad is more resistant to this migration, but it introduces other concerns, including contact resistance at low pressure and permanent compression set.
Liquid metal belongs to a different risk class rather than representing the next thermal-pad grade. Its electrical conductivity, material compatibility, containment, manufacturing, and service requirements require separate evaluation.
Failure Modes to Design Out
| Failure mode | Typical field signature | Design response |
|---|---|---|
| Pump-out or migration | Hotspot temperature rises at constant load; interface material appears around the perimeter | Control the bond line, provide uniform pressure, and consider a cured or solid format |
| Compression set | Contact force decays after initial thermal cycles and the gap begins to reopen | Specify compression-set limits and use springs or retention features that maintain load |
| Silicone bleed | Oil film, fogging near optics, or contaminated connectors | Use a low-volatility or non-silicone pad and request relevant outgassing data |
| Over-compression | Package damage, deformed standoffs, or tearing at pad edges | Add hard stops, define force limits, and select a softer grade where required |
| Insufficient wetting | High resistance at local low spots or a thermal gradient across the package | Set flatness and parallelism requirements, increase appropriate force, or use a more conformable pad |
Several apparent material failures are actually drawing or assembly failures. Over-compression may result from unspecified torque or standoff height. Poor wetting may result from missing flatness requirements. Both are generally less expensive to correct in the design and process documents than by repeatedly changing materials.
One-Page Selection Checklist
- Measure the minimum, nominal, and maximum gap on real assemblies rather than relying only on CAD.
- Define the required bond line at target compression and calculate the force available from the retention hardware.
- Confirm through-plane conductivity and its test method, preferably with ASTM D5470 data.
- Compare thermal impedance at the intended pressure, thickness, and compression.
- Review both Shore OO hardness and the deflection-versus-force curve.
- Verify breakdown strength, volume resistivity, flame rating, and outgassing requirements.
- Add cold-plate flatness and parallelism requirements to the drawing.
- Document torque, tightening sequence, standoff height, and hard-stop dimensions.
- Agree on aging, damp-heat, thermal-cycling, and compression-set acceptance limits.
- Define incoming inspection and treat the pad as a consumable during rework.
If several of these items remain undefined, selecting the conductivity grade is premature. The most expensive material in the catalog cannot compensate for an uncontrolled gap, inadequate retention force, or poor surface contact.
Frequently Asked Questions
Is a 16 W/m·K thermal pad always better than an 8 W/m·K pad?
No. It has lower calculated bulk resistance at the same bond line, but total assembly performance also depends on contact resistance, compression, flatness, pressure distribution, and actual pad thickness. A softer 8 W/m·K pad can outperform a poorly compressed 16 W/m·K pad in a mechanically constrained joint.
When is upgrading from 12 to 16 W/m·K most worthwhile?
The upgrade becomes more valuable as heat load and bond-line thickness increase, provided the assembly maintains good contact. In the first-order model used here, the resistance reduction grows from about 6.1% at 0.25 mm to about 16.5% at 1.50 mm.
Should the thickest available pad be used to guarantee contact?
No. Excess thickness increases bulk thermal resistance and may require more compression force. Choose thickness from the measured gap window, target compression, tolerance stack, and available retention force.
Can the same pad be used for an HBM stack and an optical module?
Not automatically. HBM applications may prioritize high heat-flux performance and low package stress, while optical assemblies can impose stricter requirements for low volatility, siloxane control, and limited mechanical loading.
What should be compared when two suppliers quote different conductivity values?
Confirm the test method, heat-flow direction, sample thickness, pressure, compression state, and conditioning. Then compare measured thermal impedance under conditions close to the intended assembly. Headline conductivity values measured under different conditions are not sufficient for selection.
Final Recommendation
A 16 W/m·K thermal pad is not a universal upgrade. It is an application-specific tool for high-power interfaces where bulk resistance is important and the mechanical design can create a controlled, uniform joint.
For AI accelerators, HBM stacks, and other high-heat-flux components, begin with the real gap window and retention-force budget. Then evaluate conductivity, impedance, hardness, electrical insulation, flame rating, contamination risk, and long-term resistance stability as one system. If the assembly cannot deliver the pressure, flatness, and bond-line control required by the material, a softer grade or a different interface format may provide the better thermal solution.
Before issuing an RFQ, provide the stack drawing, measured gap range, contact area, heat load, voltage requirement, available force, and qualification conditions. Those inputs determine whether the interface can use a 16 W/m·K pad effectively—or whether another grade or material format is the more reliable choice.