Compact ESP32-S3 clones (Zero/SuperMini) may overheat and fail after a normal CSI session

[matiapa]

Summary

Three ESP32-S3-Zero boards (compact/coin-sized clones, 4MB flash) ran a normal
CSI streaming session (WiFi connected, edge_tier=2, ~20-30 min, node_ids
1-3) without any errors. They ran warm/hot to the touch throughout, which
wasn't flagged as unusual at the time. After being unplugged and later
reconnected (to a different, known-good USB power source), none of the three
boards powered on again, and all three heated up immediately when powered —
consistent with permanent regulator/component damage rather than a transient
fault.

Suspected root cause

This firmware runs the WiFi radio with modem sleep disabled
(WIFI_PS_NONE, needed for continuous CSI capture) plus a full edge
DSP pipeline on Core 1 (edge_tier=2) — sustained high current draw with
no duty-cycling. Full-size dev boards (DevKitC-1, XIAO) have enough
copper pour / thermal mass around the regulator to handle this
indefinitely. Coin-sized clones (ESP32-S3-Zero, SuperMini) have minimal
PCB copper and often budget voltage regulators, and may not be rated for
sustained full-load operation without a heatsink or airflow. All three
boards in this report were the same model, run side by side in the same
session, which is consistent with a shared thermal/duty-cycle cause
rather than three independent unit failures.

What we already documented (#1288)

Added an explicit thermal-risk warning to:

• firmware/esp32-csi-node/README.md (Hardware Requirements section)
• CLAUDE.md (Supported Hardware table)
• ruview-hardware-setup skill (so an AI agent flashing on someone's
  behalf asks about board form factor and warns proactively)

Possible follow-up (not done in #1288)

• A Kconfig / NVS-configurable duty-cycled mode (periodic
  esp_wifi_stop()/light-sleep windows) for compact boards, trading
  some CSI throughput for sustainably lower average power.
• A boot-time runtime check or periodic internal temperature read (if
  the SoC's internal temp sensor is exposed) that logs a warning if the
  chip die temperature climbs unusually high, to catch this before
  hardware damage rather than after.
• Explicit non-recommendation of ESP32-S3-Zero/SuperMini for edge_tier=2
  continuous deployments, only for short/tier-0-1 test sessions, pending
  real thermal characterization.

Environment

• Firmware: esp32-csi-node v0.7.0, 4MB build (sdkconfig.defaults.4mb)
• Board: Waveshare ESP32-S3-Zero (compact/coin form factor)
• Failure mode: no continuity fault (open, not shorted) measured at rest;
  ~0.04V on VBUS under USB power immediately after failure; later,
  boards drew current and heated immediately even from a known-good
  phone-charger USB source