Company News

MIWAKEE launches high-conductivity TIM line for AI accelerators

2026-11-18
Key takeaways
  • New 10–15 W/m·K silicone pad family targets AI GPU accelerators
  • Low-CTE formulation accommodates tall tolerance stack-ups
  • UL94 V-0 rated, RoHS and REACH compliant

Data center GPU accelerators now routinely exceed 700 W per board, pushing thermal interface material (TIM) requirements well beyond the 5–8 W/m·K range that served the last decade of server CPU design. MIWAKEE has responded with a new high-conductivity silicone pad family spanning 10, 12 and 15 W/m·K, purpose-built for the thermal demands of AI accelerators.

Why 15 W/m·K is the new baseline

At the junction-to-case level of a modern AI accelerator, a 1 °C reduction in die temperature can extend device lifetime by several months under continuous load. Our engineering team found that the 15 W/m·K grade, when paired with a low-CTE formulation and a 0.3 mm bond line, delivers a 4–6 °C drop versus the 8 W/m·K grade that was common in 2023–24 GPU reference designs.

The new line is available in two CTE options: a standard formulation (250 ppm/°C) for general high-power modules, and a low-CTE variant (80 ppm/°C) for stack-ups where thermal expansion mismatch between silicon, substrate and heatsink creates a risk of pad extrusion over time.

Grade Thermal conductivity (W/m·K) Thickness range (mm) Max operating temp CTE (ppm/°C)
15 W 15.0 0.3 – 5.0 200 °C 250 / 80
12 W 12.0 0.5 – 10.0 200 °C 250
10 W 10.0 0.5 – 10.0 200 °C 250

What this means for your design

For AI accelerator board designers evaluating the new family, we recommend starting with a 0.3–0.5 mm 15 W/m·K grade in the primary die-to-heatsink interface, then stepping down to 10 or 12 W/m·K for secondary interfaces where the thermal budget is less critical. Our application engineers can model your specific stack-up and recommend the optimal grade and bond-line thickness.

Samples of the new line are available through the Request a Sample page. We expect first mass-production shipments to begin in Q1 2027.

AI ServerHigh-conductivityNew Product
Share