Industry Insights

From 22 W to E1.S Airflow: How an Enterprise SSD Thermal Budget Is Allocated

著者:JH SemiconductorSSD thermal design / thermal resistance chain / thermal interface material

Thermal capability is a property of the drive and the chassis airflow together. This piece walks the series thermal resistance from junction to ambient, the TIM compromise, and why the shape of the throttling policy matters more than the size of the spreader.

Form Factor Sets the Budget First

Thermal capability is not a property of the drive alone; it emerges from the drive and the chassis airflow together. J750 and J770 use a 15mm U.2/U.3 form factor, with product-definition/EVT design specifications of 18 W typical, 22 W maximum and 6 W idle. Heat originates mainly at the controller and the NAND packages, conducts into the metal case, and leaves through front-to-back chassis airflow. Put the same power into a thin form factor such as E1.S and the available dissipation area drops sharply. That is one practical reason FLEX 9 distinguishes E3.S (J100E, J130E) from E1.S (J100S, J130S): different thermal budgets support different performance points.

Only the Middle of the Resistance Chain Is Yours to Improve

Junction to ambient is a series thermal resistance: die junction to package case, case through the thermal interface material to the heat-spreading structure, and structure to airflow. The first term is fixed by the package; engineering effort goes into the other two. Thermal interface material selection is a compromise between conductivity, tolerance absorption and long-term aging — thin, rigid materials give lower resistance but demand flat assembly, while thick compliant materials forgive tolerance at the cost of resistance. The spreader itself trades fin density against airflow impedance, and fins that are too dense are counterproductive in a low-flow 1U chassis.

Throttling Policy Matters More Than the Heatsink

Every design has operating conditions beyond its dissipation capability; what distinguishes designs is how they retreat. The crude approach cuts hard at a temperature threshold and produces a performance cliff. The better approach reduces NAND channel concurrency or command issue rate in graded steps with hysteresis, so bandwidth degrades smoothly. Hysteresis matters especially: oscillation around a threshold makes bandwidth flip between two levels repeatedly, which disturbs upper-layer scheduling more than a stable drop to the lower level would. In parallel, state has to be reported honestly through NVMe temperature thresholds and SMART logs so the orchestration layer can migrate load in advance rather than react to an alarm.

Validation Belongs in the System

Bare-drive temperature-rise data proves the spreader works; it does not prove the product is usable in the target chassis. Validation requires thermal imaging and junction temperature readback in a real machine at full slot population, covering worst-case slot position, maximum ambient and fan degradation, with composite temperature and throttle level logged continuously through those conditions to confirm that threshold actions match the design and that slot-to-slot spread stays bounded. Temperature distributions observed during system bring-up frequently differ from bench conclusions by a whole throttle step.

Thermal Design Is Not Won at 22 W

The maximum power on a spec sheet is only an entry condition. What determines the operational experience is what happens past it: whether the drive quietly loses speed, or hands its state up the stack honestly enough for someone else to act on.

キーワード
SSD thermal designthermal resistance chainthermal interface materialthrottling hysteresisE1.SE3.Ssystem level validation