Member Center
    Exit
    qr Code Url

    Scan qrcode to view mobile website

    image English
    • image English
    • image 日本語
    • image Français
    Home /News /THERMAL PASTE /Thermal Paste Selection for Server Thermal Management /

    Thermal Paste Selection for Server Thermal Management

    author: CHACE / Tousen Thermal Management Engineering Team
    2026-11-18
    {当前产品的产品关键词轮巡使用}

    Thermal Paste in a Server: The Joint You Actually Control Is TIM2

    A two-socket 1U node running a 300 W processor per socket is not a hard thermal problem in terms of temperature; it is a hard problem in terms of duration and repeatability. The processor reports a case temperature to be held for six years, twenty-four hours a day, and the heatsink is removed and refitted several times during that life. Heat flux at the lid is roughly 17 W/cm2 for a 300 W part on a 45 x 45 mm integrated heat spreader, which is modest, but the joint has to be the same thickness on unit one thousand as on unit one. That is where server thermal management differs from every consumer device: nobody is chasing a burst, everybody is chasing consistency across a fleet.

    Three interfaces, and only one of them is yours

    A packaged server processor already contains a thermal path inside the package, and confusing it with the assembly joint is the most common source of wrong specifications. The die-to-lid joint is applied by the processor vendor, usually as solder or a pre-applied metal or polymer film, and it is not a purchasing decision for the server builder. The lid-to-heatsink joint is the one created on the line, and that is the only interface a thermal paste choice actually controls.

    Interface Flux at rated power Who sets it Serviceable
    Die to integrated heat spreader (TIM1) 40–60 W/cm2 Processor vendor No
    Lid to heatsink base (TIM2) 15–20 W/cm2 Server assembly line Yes
    Heatsink base to chassis airflow Below 1 W/cm2 Mechanical design Not a joint

    Because TIM1 is fixed, the only lever left at assembly is the bond line of the thermal interface material at TIM2, plus the flatness of two surfaces that both carry tolerances. A paste specification that ignores this split is optimised for the wrong interface.

    High pressure is the advantage a server has

    Unlike a phone or a tablet, a server heatsink is bolted down through a load plate at 30 to 100 psi on the lid. That pressure is what makes a thermal interface material of 10 to 20 µm realistic with a non-curing paste, and it is also what makes a compressible pad unnecessary on this joint — an advantage no consumer device has. At 17 W/cm2 the difference between joint options is large enough to be seen in fan power, not only in junction temperature.

    Option Bond line Rise at 17 W/cm2 Verdict for a server
    Thermal paste, 0.03 °C·cm2/W 10–20 µm 0.5 K Best fit at high clamp load
    Thermal paste, 0.080 °C·cm2/W 20–40 µm 1.4 K Fit when the lid is less flat
    Thermal paste, 0.201 °C·cm2/W 30–60 µm 3.4 K Acceptable in low-TDP nodes
    Gap filler pad, 0.5 mm at 5 W/m·K 500 µm 17 K Wrong joint for a bolted lid
    Liquid metal 10–20 µm 0.4 K Gain too small for the risk
    Note: liquid metal appears because it is the first question engineers ask. On a bolted lid the paste already reaches 0.5 K, and the remaining 0.1 K does not repay the handling restrictions; the same holds for any thermal interface material chosen by headline conductivity alone.

    Five things a datasheet does not describe

    The lid is not flat. A spreader can be concave by 20 to 50 µm across its face after assembly, and the thermal paste is expected to fill that dish. Specifying a 10 µm bond line against a 40 µm concavity produces a joint that is thin at the edges and thick at the centre, and the average is what shows up in the test data.

    Multi-die packages are uneven by design. A processor built from several chiplets presents a lid whose local temperature varies by 10 K or more across the package. The paste joint does not level that out; it only has to avoid adding a second, uncontrolled gradient on top of it, which is the whole ambition of server thermal management on a multi-die package.

    Six years at low amplitude. A server does not cycle like a phone. It runs at high load continuously and changes between power states many times per hour, which drives slow pump-out rather than fast — the ageing mode that governs server thermal management. The relevant property is oil separation and creep over thousands of hours, not the viscosity measured in one second.

    Refurbishment. A node is opened for processor replacement, for cleaning or for resale, and each opening destroys the joint. A non-curing paste allows a clean re-paste; a cured or phase-changed film does not, and a pad compressed twice reads worse than a fresh print.

    Airflow is the real limit. In a 1U chassis the heatsink is fed by 40 mm fans at high static pressure, and a 0.5 K gain on the thermal paste joint is worth about the same as a small fin-pitch change.

    Selecting the grade for a bolted lid

    The TSAD Series from TOUSEN is a solvent-free, non-curing, thixotropic gray paste covering 1.2–6.5 W/m·K, with measured impedance of 0.03 to 0.201 °C·cm2/W at 30 psi, specific gravity 2.5–2.9 g/cm3 and a 12-month shelf life. For a bolted server lid the selection question for a thermal interface material is narrower than in a phone: pick the grade whose bond line the load plate can hold flat, then confirm oil separation over the service life.

    Grade W/m·K Impedance @ 30 psi (°C·cm2/W) Density (g/cm3) Typical server position
    TSAD60 1.2 0.201 2.5 Low-TDP nodes, auxiliary components
    TSAD66 3.5 0.080 2.7 NIC and retimer lids, PSU modules
    TSAD66-P 4.0 0.070 2.6 Lids with visible concavity
    TSAD83 PLUS 5.2 0.050 2.6 Mainstream 150–250 W processors
    TSAD88 PLUS 5.2 0.030 2.9 High-TDP processor, thin joint target
    TSAD90 6.5 0.030 2.7 300 W class processor, 1U heatsink
    TSAD100 6.5 0.030 2.8 Cold plate for accelerators at 500 W and above

    Dose in milligrams, because the joint is only 10 to 20 µm

    At 2.7 g/cm3 a 15 µm bond line needs 4 mg per cm2, so a 20 cm2 lid takes about 80 mg. At that scale a 30 percent over-dose of thermal paste is a 5 µm penalty spread over the whole lid, and a syringe with a defined shot size is a better process control than an operator judging a cross pattern.

    Print when the line builds thousands of units

    Stencil printing the paste at 0.1 mm with 80 to 90 percent coverage gives a joint whose thickness is set by the screen and the clamp, not by the operator, and it is the only method that makes unit one thousand equal to unit one. Dispensing remains valid for low-volume builds and refurbishment cells, where volume is small and geometry varies.

    Where paste is the wrong part

    Around voltage regulator modules with large gaps, under mezzanine cards and between a chassis and a cold plate flange, a gap filler or a pad is correct; those joints exist to bridge 0.5 to 2 mm and to absorb tolerance. A paste there would simply flow out of the gap, a mistake that recurs in server thermal management because the gaps differ from node to node.

    Qualification for a fleet, not for a sample

    Because server thermal management is judged across a fleet, run the node at rated TDP for 1,000 hours and log lid and inlet temperatures at fixed fan speed; the joint that reads best on day one is not always the one that reads best on day forty. Add 5,000 power-state cycles of ten minutes each, a 150 °C gravimetric oil separation test, 500 hours at 60 °C and 90 % RH for storage and transport, and an impedance re-measurement of the paste joint on a disassembled unit. Measure lid flatness before assembly on a sample from each lot, because a change in concavity will change the bond line more than a change of grade will.

    Assembly rules that survive an audit

    • Clean both surfaces with isopropyl alcohol and verify visually; a residue under 15 µm is a hot spot no heatsink can average out.
    • Torque the load plate in the drawing's diagonal sequence in two passes, and do not re-torque after the joint has settled.
    • Use a single, dated syringe per node or a printed deposit; topping up a partly used syringe changes the solids content.
    • Never reuse paste removed from a lid, and never mix two grades to "improve" the result.
    • Store cartridges closed below 30 °C, away from flame; the product is non-hazardous, but an open cartridge loses solvent balance over months.

    Conclusion

    Server thermal management is less about a peak temperature and more about a thermal paste joint that is identical across a fleet and stable across six years. The processor's internal TIM1 is fixed by the vendor; the only interface the assembly line owns is TIM2, and at 30 to 100 psi it can be held to 10 to 20 µm with a non-curing thermal paste. Choose the grade by the bond line the load plate can actually hold, dose it in milligrams, print it where volume justifies a stencil, and qualify it over a thousand hours at rated power rather than over a single thermal image. Between 0.03 and 0.201 °C·cm2/W at 17 W/cm2 there are almost 3 K, which in a 1U node is the difference between a fan curve that holds and one that ramps.

    Impedance data, grade table and packaging options for the thermal paste described here are published on the TSAD Series thermal paste product page, and samples for stencil trials on your own lid can be requested there.

    Share:

    Thermal Paste Selection for Smartphone Thermal Management

    Related Article

    image
    Smartphone thermal management is a burst problem: at 25 W/cm2 a thermal paste joint adds 0.75 K where a 0.5 mm pad adds 12.5 K, and the joint is the one layer assembly still controls.
    Thermal Paste Selection for Smartphone Thermal Management
    2026-11-17
    image
    Tablet thermal management is a passive problem: thermal paste at 8.5 W/cm2 adds 0.3 K where a 0.5 mm gap filler adds 8.5 K, and the bond line remains the one layer the assembly process still controls.
    Thermal Paste Selection for Tablet Thermal Management
    2026-11-13
    image
    Laptop thermal management lives in a 20 to 50 micron interface: thermal paste at 35 W/cm2 adds under 2 K, while a 0.3 mm gap filler adds 17 K to the same joint between die and cold plate.
    Thermal Paste Selection for Laptop Thermal Management
    2026-11-12
    image
    Thermally conductive paste suits flat ceramic faces and low clamp pressure in thermoelectric cooler thermal management: a 30 micrometre joint adds 0.3 K at 10 W/cm2 where a 0.5 mm pad costs about 6 K.
    Thermally Conductive Paste Selection for Peltier Cooling Interfaces
    2026-11-11
    SiteMap

    Home

    Products

    Application

    News

    Download

    HOW CAN I HELP YOU

    TEL YANAGI: +86 18566122282 WhatsApp: +81 80 7029 9037

    Email: ZDLIU@ITOUSEN.COM SALES@ITOUSEN.COM  Europe@itousen.com

    If you have any questions related to thermal management, including material selection and the design of thermal management systems and solutions, please feel free to contact us. Together, we can ensure the stable and reliable operation of our equipment and machines!

    Copyright © 2021 King Theme v2. Powered by web Jinggong Network Security No. 32058302002032

    Terms Of Use Privacy Cookies Terms & Conditions

    CONTACT US

    • You can only upload 1 attachments at most
    • Supported formats: PDF,JPG,PNG,EXCEL,WORD,STP,IGS,DWG,DXF,PDF,STEP
    • The size of the uploaded file should not exceed 10MB
    (384096)
    0