ESP32-DevKitC-32E: stock deep sleep and light sleep with a timer wake-up
Demand rank: ESP32 is #1 (38 % of respondents, JetBrains State of Embedded
2025). This is depth work on a part already modelled, not a new part. The
tested system is Espressif's ESP32-DevKitC-32E with its ESP32-WROOM-32E
module. Deep sleep with a timer wake-up is the standard pattern for a
battery-powered ESP32 sensor node (Arduino-ESP32 example
DeepSleep/TimerWakeUp).
Vendor status (checked live 2026-11-09): Espressif's current ESP32-DevKitC V4 guide lists the exact ESP32-DevKitC-32E / ESP32-WROOM-32E combination. Its current module catalog lists ESP32-WROOM-32E-N4 and purchase links for DigiKey, Mouser, Amazon US, and Robu India. Espressif's longevity commitment lists ESP32 for at least 15 years from 2016 (through 2031 at minimum). These current vendor listings establish that the tested board/module remains available; the longevity page is a supply commitment, not an NRND status field.
Result
| sketch | before (lane 193e168d6) |
after |
|---|---|---|
deep_sleep_timer.cpp |
fault at 0x3FF5_D054 after sleeping |
LW_DS_OK, 2,027,761 steps |
light_sleep_timer.cpp |
not run (same rtc_sleep_init fault path) |
LW_LS_OK, 702,155 steps |
Deep sleep, after (after/deep_sleep_timer.uart.log):
boot=1 cause=0 reset=1 power-on: ESP_SLEEP_WAKEUP_UNDEFINED, ESP_RST_POWERON
sleeping
boot=2 cause=4 reset=8 ESP_SLEEP_WAKEUP_TIMER, ESP_RST_DEEPSLEEP
sleeping
boot=3 cause=4 reset=8
LW_DS_OK
bootCount is an RTC_DATA_ATTR variable. It counts 1, 2, 3, so RTC slow
memory survives the wake-up reset while IRAM and DRAM are reloaded.
What changed
- Radio power-down windows.
rtc_sleep_initclears FORCE_PU bits inFE2_TX_INTERP_CTRL,FE_GEN_CTRL,NRXPD_CTRLandBBPD_CTRLbefore every sleep. FE20x3FF4_5000, FE0x3FF4_6000, NRX0x3FF5_C000and BB0x3FF5_D000are now plain register stubs. No radio is modelled. - RTC_CNTL sleep (
peripherals/esp32/rtc_cntl.rs). A write ofSTATE0.SLEEP_ENwithTIMER_TRIGinWAKEUP_ENAmoves the RTC slow counter to theSLP_TIMER0/1alarm and setsWAKEUP_CAUSE = TIMER_TRIG. WithDIG_PWC.DG_WRAP_PD_ENset (deep sleep), both cores takeDEEPSLEEP_RESETand a wake latch is raised. Otherwise (light sleep)INT_RAW.SLP_WAKEUPis set, sortc_sleep_start's poll returns.INT_CLRnow clearsINT_RAW. - Deep-sleep reboot (
Machine::capture_esp32_deep_sleep_boot). Thelabwired testESP32 path records the IRAM/DRAM segment bytes, entry PC and stacks after its fast-boot setup. On the wake latch the machine writes those bytes back, resets both cores (APP_CPU halted until PRO_CPU releases it) and starts at the entry. RTC_CNTL, RTC slow and RTC fast memory are not touched. This is what the 2nd-stage bootloader does on silicon: it skips the RTC segments afterDEEPSLEEP_RESET. - ROM
rtc_get_reset_reason(0x4000_81d4) now decodesRTC_CNTL_RESET_STATEper core, as the BROM does. Before, it always returned POWERON_RESET.
The intermediate build in before/deep_sleep_timer.txt is a negative
control: with item 4 missing, the sketch wakes and counts but prints
LW_DS_FAIL.
Not covered
- Wake sources other than the RTC timer (EXT0/EXT1 GPIO, touch, ULP, UART). With only those enabled the chip sleeps for ever.
- Sleep duration in CPU time. Only the RTC slow counter jumps to the alarm;
the cycle counter and
esp_timerdo not advance across the sleep. - Power figures, the deep-sleep wake stub, and RTC GPIO hold.
- The wasm/playground path and
--rom-boot. They do not capture a reboot state, so a deep sleep there halts both cores. - ESP32-S2/S3/C3/C6. Their sleep goes through other register blocks (and PMU on the C6) and is not modelled.
Reproduce
From the repository root, build both sketches with the pinned PlatformIO project, then run the two acceptance scripts:
pio run --project-dir docs/receipts/2026-11-09-esp32-deep-sleep/sketch
labwired test --script docs/receipts/2026-11-09-esp32-deep-sleep/sketch/deep_sleep_timer.test.yaml --output-dir out/deep-sleep
labwired test --script docs/receipts/2026-11-09-esp32-deep-sleep/sketch/light_sleep_timer.test.yaml --output-dir out/light-sleep
The project pins Espressif PlatformIO platform 8.0.0; the recorded build used
Arduino-ESP32 3.3.12. Both scripts use the exact
configs/systems/esp32-devkitc-32e-gpio-smoke.yaml board system.