AM625: AM625 Elevated junction temperature in 70 °C ambient

Device/SoC: AM6254ATCGGAALW
Use case: HMI (graphics UI), without active cooling
Form factor: Custom SMARC-size SoM + carrier (EVM-SK2-like schematic, same PMIC & active parts)
Operating point: CPU @ 1 GHz (extended industrial range)
Goal: Stable operation at 70C ambient
SW stack: Yocto-built Linux 6.1.83, Qt 6.6.3 (no vendor SDK changes beyond board enablement)
Thermal readout: Linux thermal zones (cat /sys/class/thermal/thermal_zone*/temp) — die sensors

Symptoms / Measurements

All temps are junction temps from on-die sensors (°C). Ambient measured with calibrated probe.

Notes:

• Reducing CPU core clock and even stopping cores did not meaningfully reduce reported die temperature.
• GPU/DDR/display active as per normal HMI workload.

Questions to TI

1. How can we effectively reduce the operating temperature on AM6254 in this configuration (kernel 6.1.83, 1 GHz OPP, HMI workload)?
   • Recommended DVFS/cooling-device settings, trip points, or kernel config for AM62x?
   • Any guidance on which thermal zones correspond to CPU vs. GPU vs. SoC/PMIC sensors on AM62x so we can target the right hotspots?
2. Why does the die temperature increase further inside the enclosure (70C ambient in a closed box results in ~102C reported die temp), even with a full-module heatsink?
   • We’d appreciate an explanation specific to AM62x package/PMIC/display heat paths and enclosure effects (conduction paths, recirculation, LCD backlight, PMIC proximity, etc.).

What we’ve already tried

• Lowered CPU frequency and disabled cores → negligible change in die temp readings under the same ambient.
• Bench vs. enclosed comparisons (data above).
• Basic heatsink that spans the SMARC module top surface.

Environment Details

• Schematic: Based on EVM-SK-AM62 (PMIC and all other active components match reference).
• Thermal readouts:

• for z in /sys/class/thermal/thermal_zone*; do
•     echo -n "$(basename $z): "; cat $z/type; cat $z/temp
• done

• CPUFreq/OPP info:

• cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
• cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq

• Loads while measuring: Qt HMI running , display active;

Requested TI Guidance

• Recommended thermal design power (TDP) budgeting and junction-to-ambient guidance for AM6254 at 1 GHz in 70C ambient, including:
  • Heatsink selection (theta_sa targets), TIM recommendations, and whether heat spreader + case conduction is advisable for SMARC.
  • Any AM62x-specific DVFS tables, cooling-device bindings, or passive cooling policies we should enable in DT/Kconfig.
  • Clarification on thermal sensor calibration and accuracy in Linux for AM62x (do we need any eFuse/trim enablement in DT/driver?).
  • Mapping of thermal_zoneN → sensor/function on AM6254 (CPU, GPU, DDR, SoC, etc.).
• Confirmation that reported temps (~86–102C at 70C ambient under HMI load) are expected / out-of-family, and suggested target junction limits for our grade.

Thanks in advance for guidance on AM6254 thermal best practices for HMI at high ambient.