When an HMI panel goes into a machine that will be manufactured and supported for ten years, the processor choice is the least interesting decision and the most consequential one. The datasheet tells you what the chip can do on launch day; it says nothing about whether you can still buy it in 2033 or rebuild the image after a kernel CVE.
Start from the silicon longevity commitment
We design around platforms where the vendor has published a multi-year availability commitment - Rockchip's RK3568 and RK3588, and Allwinner's long-life industrial lines are examples. A consumer-grade SoC with no published EOL plan is a hidden liability on a long-life machine.
Budget for BSP maintenance, not just bring-up
The board works on day one. The question is who keeps the Linux BSP patched through five kernel releases. Contract a maintenance plan with reproducible toolchains so a build from 2026 can be reproduced bit-for-bit in 2032.
- Published silicon longevity (10+ years)
- Maintained BSP: kernel CVE patching and quarterly releases
- Formal PCN process and buffer stock against forecast
- Managed last-time-buy window
A discontinued panel that still fits the enclosure is fixable. A discontinued processor with no drop-in replacement is a redesign.
Make the mechanical interface boring
Standard cut-outs, standard connector positions and a documented panel timing library remove the two variables that cause most HMI redesigns: the bezel and the display driver. Keep those fixed and the electronics become swappable.