nrf54lm20a: flash algorithm faults (IPSR=3) and 'pyocd reset' is a no-op

[Divad27]

Summary

With target nrf54lm20a (Seeed XIAO nRF54LM20A, on-board CMSIS-DAP), pyOCD cannot
program the device and pyocd reset has no effect. Reading and mass-erase work fine,
so the probe connection itself is healthy.

Tested with pyOCD 0.44.1 and 0.45, on 5 boards, Windows 11.

Command	Result
pyocd list	✅ works
pyocd rtt	✅ works (RTT streaming is reliable)
pyocd erase --mass	✅ works (needed once on fresh boards — they ship with APPROTECT set)
pyocd flash / load	❌ fails inside the flash algorithm with a fault (IPSR = 3 / HardFault); the old image stays on the device
pyocd reset	❌ no effect — the target keeps running, accompanied by NRF54LM20A is not in a secure state

For contrast, the nRF54L15 (target nrf54l) programs flawlessly with pyOCD on the
same host and the same kind of on-board probe, so this appears specific to the
nRF54LM20A target support (RRAM instead of the usual flash controller).

Steps to reproduce

pyocd erase --mass -t nrf54lm20a -u <SERIAL>     # fresh boards: clears APPROTECT
pyocd flash -t nrf54lm20a -u <SERIAL> zephyr.hex # faults in the flash algo
pyocd reset -t nrf54lm20a -u <SERIAL>            # returns, target unaffected

Expected

flash programs the image; reset restarts the target.

Workaround we use

OpenOCD with Seeed's support/openocd.cfg, using the vendor nrf54lm20a-load command
plus verify_image, each command passed as its own -c (chaining them with semicolons
swallows the load/verify output):

openocd -f boards/seeed/xiao_nrf54lm20a/support/openocd.cfg \
  -c "adapter serial <SERIAL>" -c init -c "reset halt" \
  -c "nrf54lm20a-load app.hex" -c "verify_image app.hex" \
  -c "reset run" -c shutdown

We still use pyOCD for list, rtt and erase --mass — those are solid.

Environment

• pyOCD 0.44.1 and 0.45 (both affected)
• Windows 11, Python 3.12
• Seeed XIAO nRF54LM20A (non-Sense), on-board CMSIS-DAP, 5 units — all behave identically

Note

Possibly related to #1985 (pyocd load fails for nRF54L15), but the signature differs
(there: skipped bytes on nRF54L15; here: a fault inside the algorithm on nRF54LM20A) and
nRF54L15 works fine for us — so I am filing this separately rather than piling on.

Happy to run any diagnostic command that would help narrow this down; the hardware is
on my desk.