Published August 25, 2026

Artificial intelligence (AI) and machine learning (ML) at the edge are becoming increasingly common. Today, edge inference models routinely run at 15 trillion operations per second (TOPS) to 40TOPS while consuming 15W to 25W of power. These models often operate under load in sealed, IP67-rated enclosures designed to resist dust ingress and temporary water immersion, even at high ambient temperatures. In applications such as security, autonomous robotics, drones, and emerging data harvesting, these edge models and other compute activities run nearly continuously without access to a cooling fan. This can be thought of as desktop-class heat flux concentrated inside a compact sealed enclosure.

These edge compute systems are often sealed either to prevent environmental ingress, as in EV chargers, industrial vision, 5G small cells, and automotive compute, or to ensure sterility and reduce maintenance, as in medical and laboratory environments. As edge computing devices become more powerful and compact, additional applications for compute technology are possible, but the increased thermal flux at such high-power densities presents a challenge that must be addressed.

This blog explores the thermal challenges associated with sealed, compute-dense electronics and examines the materials and design strategies that help manage heat when active cooling is not an option.

Why Sealed and Dense Edge Compute Breaks Classical Thermal Intuition

Within an extremely dense, sealed compute device, natural convection becomes weak and stratified. In these enclosures, conduction is the dominant form of heat transfer. The primary thermal pathway, starting at the die, travels through the die package, the thermal interface material (TIM), to the heat spreader—if used—the enclosure body, and then to the ambient environment (Figure 1). This poses a challenge at the power densities of modern edge compute hardware. For instance, trapped air is an effective insulator, with a thermal conductivity of approximately 0.026W/mK. For the same 1mm bond-line thickness, a trapped air gap has roughly 150 times the thermal resistance of a 4W/mK pad and roughly 40 times that of a 1W/mK grease. With the need to use the sealed enclosure wall as a radiator, the TIM can become the dominant thermal resistor in the stack. This can result in extremely high semiconductor junction temperatures, leading to issues such as thermal throttling, reduced life, or even hard failures. Junction temperature depends on many factors, such as power dissipation, duty cycle, ambient temperature, nearby heat sources, transient duration, and junction-to-ambient thermal resistance. When the operating conditions and external environment are fixed, reducing the thermal resistance of each component in the thermal stack may help lower the die's operating junction temperature.

Figure 1: Thermal management keeps electronic devices cool, enhancing performance, reliability, and longevity. (Source: Würth Elektronik) >>Click to read the entire article at Mouser

Where the Heat Path Actually Fails

Though optimizing each element of the thermal stack is critical in edge compute applications, there are particularly challenging areas to consider. At high peak temperatures, likely during bursty inference loads,  thermal grease can pump out and dry. For thermal pads, under long-term clamping loads, these pads can compress over time, resulting in a drift in thermal conductivity. After thermal cycling and vibration, coefficient of thermal expansion (CTE)-driven TIM delamination can occur.

In each of these cases, increased thermal resistance and higher peak temperatures of the die, package, and circuit board can lead to outcomes that are consistent with common electronics reliability heuristics. For many electrolytic capacitors, an approximately 10°C increase in operating temperature can significantly reduce lifetime, while Coffin-Manson-type fatigue models are commonly used to relate repeated temperature cycling and strain to solder-joint fatigue. In summary, overheating and high-temperature peaks result in reduced system performance, increased failure rates, and a significantly shorter working lifespan for compute hardware.

Grease pump-out is primarily a material-flow and mechanical-displacement problem. Thermal cycling and CTE mismatch can gradually move grease out of the bond line, while aging at elevated temperature can cause separation or dry-out. The resulting voids may be filled with air, whose thermal conductivity is typically tens or even hundreds of times lower than that of common gap-filling materials. Phase-change materials provide one alternative. They remain solid during assembly and handling, then soften and wet the mating surfaces above their transition temperature. Because their flow is more controlled and they resolidify as the assembly cools, they can offer greater resistance to pump-out than conventional grease. Their suitability still depends on operating temperature, clamping pressure, bond-line thickness, orientation, and containment.

Long-term pad performance is primarily a mechanical-aging concern. A gap-filling pad should therefore be selected for the expected gap, clamping pressure, compression range, compression set, and stress-relaxation behavior. Selecting a soft pad based only on its initial thermal conductivity may overlook how its contact pressure and ability to follow dimensional changes will evolve over the life of the assembly.

CTE-driven interfacial separation is generally addressed by providing sufficient compliance within the thermal stack. A conformable pad or gap filler can accommodate gap variation and differential movement between components during thermal cycling. Thin adhesive thermal-transfer tapes may accommodate less gap variation or relative movement than thicker gap fillers, although their performance depends on thickness, modulus, adhesive properties, bond strength, and the magnitude of the expected CTE mismatch.

>>Click to read the entire article at Mouser

About the Author

Jean-Jacques (JJ) DeLisle attended the Rochester Institute of Technology, where he graduated with a BS and MS degree in Electrical Engineering. While studying, JJ pursued RF/microwave research, wrote for the university magazine, and was a member of the first improvisational comedy troupe @ RIT. Before completing his degree, JJ contracted as an IC layout and automated test design engineer for Synaptics Inc. After 6 years of original research—developing and characterizing intra-coaxial antennas and wireless sensor technology—JJ left RIT with several submitted technical papers and a US patent.

Würth Elektronik eiSos GmbH & Co. KG is a manufacturer of electronic and electromechanical components for the electronics industry and a technology enabler for future-oriented electronics solutions. Würth Elektronik is part of the Würth Group, a world leader in assembly and fastening technology. The product portfolio of Würth Elektronik eiSos includes EMC components, inductors, transformers, RF components, varistors, capacitors, resistors, crystals, oscillators, power modules, wireless power transfer, optoelectronics, sensors, radio modules, connectors, REDCUBE terminals, switches and buttons, assembly technology, and thermal management.