Esp32s3 Rlcd42

Waveshare ESP32-S3-RLCD-4.2

A 4.2" reflective panel with no backlight, an audio front end, four I2C slaves and an 18650, on an N16R8 module — and four free GPIOs on the header when you are done.

Two things shape almost every task here. First, the panel is 1 bpp and reflective: a full frame is 15,000 bytes, it holds its image with the SoC asleep, and it is invisible in the dark. Second, the board has no UART bridge chip and no user LED — the single Type-C socket is the S3's native USB, so if the console flags are wrong there is no fallback route and no LED to blink. A board that "does nothing" is usually a board that is running fine and cannot tell you.

• reference/board-hardware.md — the board: the vendor's "Interface Introduction" GPIO table transcribed and corrected, the full 2×8 header with the four trap positions, the display/audio/storage/power sections, console and recovery, flash layout, and a symptom → cause → fix table.
• reference/esp32s3-soc.md — the silicon: which GPIOs do not exist, strapping semantics, ADC1/ADC2 and touch maps, power-up glitches, memory map and PSRAM addressing, deep sleep and RTC wake pins.
• reference/esp32-family.md — the rest of the family, for "should this be a different ESP32?" questions.
• reference/recipes.md — code that builds: platformio.ini, USB console, ST7305 bring-up and HPM/LPM, the I2C bus with SHTC3 and PCF85063A drivers, battery, SDMMC, the audio pin/parameter reference, deep sleep.
• template/ — a project that builds clean, in two variants, plus a scaffold script. See template/README.md.

Confirm the board first

Waveshare sells several 4.2" ESP32 boards whose pin maps do not port. Check two things:

If the vendor table you are reading mentions touch (TP_SDA, TP_INT) or a QMI8658C IMU, that is not evidence this board has them — it does not, and the schematic shows those connector pins unconnected. See rule 2.

Orientation

Rules that prevent the expensive mistakes

Each of these produces a failure that looks like something else.

1. -DARDUINO_USB_MODE=1 and -DARDUINO_USB_CDC_ON_BOOT=1, or nothing is ever printed. There is no UART bridge chip on this board — the one Type-C socket is the S3's native USB. Without the flags Serial goes to UART0 on GPIO43/44, which reaches a header nobody has a cable on. The board enumerates, flashes, runs, and looks dead. There is no LED to check instead. Then wait for the host to enumerate before printing, or the boot banner is lost: uint32_t t0 = millis(); while (!Serial && millis()-t0 < 2000) delay(10);

2. The wiki's "Interface Introduction" table has wrong rows; the schematic is the authority. It lists TP_INT (GPIO7), TP_SDA/TP_SCL (13/14), TP_RESET (GPIO42) and a QMI8658C column. This board has no touch panel and no IMU. The touch signals are real pins on the LCD's 21-way FPC connector — TP_RESET, TP_INT, TP_SDA, TP_SCL on pins 16-19 — and all four are marked unconnected; the footprint merely supports a touch variant of the panel that this board does not fit. Search the schematic for nets GPIO7 and GPIO42 and they appear only on the module symbol.

   So GPIO7, GPIO42, GPIO47 and GPIO48 are genuinely free — but none of them reaches the expansion header, so using one means soldering to a module pad. Note also that the schematic PDF carries its own copy of this table in which those cells are blank: the two vendor copies disagree with each other. board-hardware.md §2.2 has the full diff.

3. The TF slot is SDMMC, not SPI — there is no CS pin, and its absence is not an omission. The card's CD/D3 has a 10 K pull-up and reaches GPIO17 only through R7, which is not populated, so the card comes up in native SD mode; D1 and D2 are unconnected at the socket, and card-detect is tied to ground. Waveshare's own BSP is CustomSDPort(name, clk = 38, cmd = 21, d0 = 39, width = 1) over esp_vfs_fat_sdmmc_mount(). SD.h on a SPI bus cannot work whatever CS you invent. In Arduino use SD_MMC.setPins(38, 21, 39) before SD_MMC.begin() — called after, setPins() returns true, changes nothing, and begin() falls back to the S3 defaults, which overlap the octal PSRAM pins of rule 4 and boot-loop the board. Pass mode1bit = true: 4-bit mode cannot work at any speed.

4. board_build.arduino.memory_type = qio_opi, or the 8 MB PSRAM is never mapped. esp32-s3-devkitc-1 is an N8 with no PSRAM in PlatformIO's database and defaults to qio_qspi. Nothing reports the mismatch: psramFound() simply returns false and large allocations fail later, somewhere else. Check psramFound() first, always. GPIO35, 36 and 37 are that PSRAM — driving one corrupts the bus the CPU reads its cached data from, and the board panics with Cache disabled but cached memory region accessed, pointing anywhere but the pin you touched.

5. SPI.begin(11, -1, 12, -1) before u8g2.begin(), or the panel stays blank with no error. U8g2's hardware-SPI constructor takes CS, DC and RST but never assigns SCLK and MOSI — it inherits whatever the global SPI object has, and the ESP32-S3 defaults are not this board's pins. Claiming the bus first makes U8g2's own SPI.begin() a no-op. This is the single most common "the display does not work" cause, and it produces no message anywhere.

6. GPIO45 and GPIO46 are strapping pins that this board also uses for audio. GPIO45 is VDD_SPI: high at reset selects a 1.8 V flash rail and the module stops reading its own flash — a board that looks bricked and is fine as soon as the pull comes off. It is also I2S LRCK. GPIO46 is a boot/ROM-log strap and also the speaker amplifier enable; the board's own R49 10 K pull-down holds it low through reset, which is exactly what the strap wants. The ESP drives both itself, so ordinary use is safe; the trap is adding an external pull-up, or wiring a peripheral that idles either line high.

7. Wire.begin() with no arguments lands on GPIO8/9 — the I2S data and bit clock. The S3 Arduino variant defaults Wire to SDA 8, SCL 9. The symptom is not an I2C error, it is a scan that finds nothing while the codecs are sitting right there. Always Wire.begin(13, 14).

8. The I2C bus is not yours. Four slaves already answer at 0x18, 0x40/0x42, 0x51 and 0x70, and the bus already has its pull-ups. An external device on the header's SDA/SCL must avoid those five addresses. Scan before you assume — the ES7210 is at 0x40 on most batches and 0x42 on some.

9. setPowerSave(1) is not the ST7305 low-power mode. It sends 0x28, display-off, and the image disappears. LPM is sendF("c", 0x39) and HPM is sendF("c", 0x38), both followed by a settling delay; U8g2 wraps neither. Writing while in LPM is allowed but the update can take a full refresh period (~1 s) to appear — so a fast redraw is HPM → draw → sendBuffer() → LPM.

10. I2S DOUT is GPIO8 and DIN is GPIO10, from the host's point of view. The vendor table names them from the codec's side (DSDIN into the codec, ASDOUT out of it), which reads backwards. The schematic is unambiguous: GPIO8 lands on ES8311 pin 9 DSDIN, and ES7210 pin 11 SDOUT1 reaches GPIO10 through R42 (51 R). The Zephyr pinctrl has them genuinely swapped and its i2s0 node is never enabled, so that source is untested — do not copy it. Getting this wrong gives you a silent speaker and a microphone that captures nothing, with both chips answering normally on I2C — which is what one user reported hitting before fixing it.

11. GPIO46 low means silence, whatever else is correct. The NS4150 amplifier's CTRL is active high and held low by a 10 K pull-down, so it is muted until firmware raises it. First thing to check when the ES8311 answers on I2C and nothing comes out. Raise it before initialising the codec.

12. Four of the sixteen header positions are traps, not expansion. GP19/GP20 are the native USB D−/D+: using them removes the console, the flashing route and USB-Serial-JTAG at once, and there is no second port. GP0 is the BOOT button — anything holding it low at reset puts the board in download mode, which presents as "my firmware stopped running". GP18 is the KEY button with the button still fitted in parallel. SDA/SCL are the shared bus of rule 8.

13. Set 12 dB ADC attenuation on GPIO4 or the battery reads a stuck maximum. A full 18650 through the 1/3 divider presents ~1.40 V, well above the S3 ADC's default ~0.95 V full scale. analogSetPinAttenuation(4, ADC_11db), then multiply the reading by 3. GPIO4 is ADC1_CH3, so it keeps working with Wi-Fi up — ADC2 does not.

14. The RTC ships with no cell, so its oscillator-stop flag is set out of the box. Bit 7 of the PCF85063A's Seconds register means "the time is not trustworthy", and on a new board it will be. That is not a fault: it is an empty PH1.0 holder. Treat it as "sync from NTP now". Only rechargeable cells belong in that holder — the board charges it, and a primary cell there can vent.

15. There is no EN/RST button — CHIP_PU carries only an RC network, no switch. The manual download sequence is hold BOOT → tap PWR (or replug USB) → release BOOT → upload, not the BOOT/EN dance printed in every other ESP32 tutorial. PWR is Key3 on a power-latch IC: long press off, single click on, and firmware cannot read it.

16. Do not use huge_app.csv. Its app slot is 3 MB and its coredump sits at 0x3F0000, so everything above 4 MB of a 16 MB part is silently stranded. default_16MB.csv is the board's sane default (dual OTA, 6.25 MB per slot, 3.375 MB SPIFFS); drop to a single-app table if you want the 6 MB back.

When the task is the display

Start from template/src/display.cpp — it is a working bring-up for this exact panel.

The mechanism worth carrying: this is a 1-bit-per-pixel reflective panel with its own self-refresh. The host writes a 15,000-byte frame into the controller's memory; the controller then holds and refreshes that image on its own, at 32 Hz in HPM or 1 Hz in LPM, with no further host involvement. So:

• U8g2 is the right library, and the full-frame _F_ buffer is affordable — 15 KB in internal SRAM, no firstPage()/nextPage() loop, no PSRAM needed. The constructor is U8G2_ST7305_300X400_F_4W_HW_SPI(U8G2_R1, 40, 5, 41). Ask for ^2.36.18, not ^2.36.19. Waveshare's docs say "v2.36.19 or later" and upstream's ChangeLog does list ST7305 300×400 under 2.36.19 — but the PlatformIO registry's newest U8g2 is 2.36.18, and that package already carries the constructors (verified by compiling against it). Asking for ^2.36.19 in lib_deps simply fails to resolve. U8G2_R1 gives the 400×300 landscape the board is read in; U8G2_R3 if it comes out upside down in your enclosure.
• Redrawing on a timer is usually wrong. The image costs nothing to hold, only to write. Redraw when the content changes. A dashboard that updates once a minute is a reasonable default; ESPHome's own config for this board ships update_interval: 1min.
• ~45 host frames/s at 24 MHz is the ceiling, and the panel still self-refreshes at 32 Hz, so writing faster than that changes nothing you can see.
• It is invisible in the dark and that is not a fault. No backlight pin exists. "The display is dim" is answered with ambient light, not with code.
• Do not port the Zephyr numbers. That port drives the same panel as an ST7306 with width = 312 and start-column = 204. Those are not arbitrary — the ST7305 addresses memory in 12-pixel blocks, so 300 pads to 312 — but they belong to that driver; feeding them to an ST7305 driver gives a shifted image. Follow Waveshare's ST7305 route.
• Waveshare ships a working driver of their own, 10_U8G2_Test/ST7305_U8g2.cpp in github.com/waveshareteam/ESP32-S3-RLCD-4.2, alongside ESP-IDF, ESPHome, LVGL and XiaoZhi examples and prebuilt firmware. Read it before hand-writing an init sequence.

When the task is choosing a pin

Count first. With the panel, the audio, the I2C bus, the card and the battery all wired — which is the board as sold — the free list is:

Then the ones you buy by giving something up: 38/21/39 cost you the card, 8/9/10/16/45/46 cost you audio, 11/12/40/41/5 cost you the panel (6 is the panel's TE line and is taken even though nothing reads it), 19/20 cost you the only USB port. 35/36/37 are never available (rule 4).

GPIO7, 42, 47 and 48 are connected to nothing (rule 2) but do not appear on the header, so they are reachable only by soldering to a module pad. Count them for a respin, not for a jumper wire.

If four is not enough, the answer is an I2C expander on the header's SDA/SCL at an address outside 0x18/0x40/0x42/0x51/0x70 — not one of the PSRAM pins.

When the task is "the board does not work"

Diagnose in this order; each step only depends on the ones before it.

1. Does anything print? If not, this is rule 1 nine times out of ten — check the two USB build flags before anything else. The port is /dev/cu.usbmodem*, /dev/ttyACM* or a COM port, VID 303A. There is no second port and no LED to fall back on.
2. Does the ROM banner (ESP-ROM:esp32s3-...) appear? It comes out before any of your configuration applies. No banner ⇒ cable, port, or the board is held in download mode by something on GP0 (rule 12).
3. Does psramFound() return true? If not, stop: rule 4. Everything downstream is a consequence.
4. Is anything wired to GPIO35/36/37, GP0, GP19/GP20 or GPIO45? Unplug it (rules 4, 6, 12).
5. Does it boot-loop? Read esp_reset_reason(). ESP_RST_BROWNOUT ⇒ thin cable or a tired 18650, and check the WRN LED for a reversed cell. Cache disabled but cached memory region accessed ⇒ rule 4.
6. Panel blank? In order: SPI.begin() before u8g2.begin() (rule 5), then U8g2 version ≥ 2.36.18, then setPowerSave (rule 9), then ambient light.
7. I2C scan empty? Rule 7 — the pins. If some addresses answer and one does not, it is that chip, not the bus.
8. Audio silent? GPIO46 first (rule 11), then the DOUT/DIN direction (rule 10).
9. Card will not mount? Rule 3, in that order: SDMMC not SPI, setPins() before begin(), mode1bit = true, then FAT32, then another card.
10. Still lost? Flash --minimal: it touches only the toolchain, the flashing route, the console and the PSRAM mapping, so it separates "wrong build flags" from "wrong peripheral code" without guessing.

template/src/board_report.cpp prints the PSRAM state, flash size, reset reason, button levels and battery voltage, which answers most of steps 3-5 in one paste.

Starting a new project

Do not hand-assemble one — the platformio.ini has five settings that are wrong by default and silent when wrong. Scaffold from template/:

~/.claude/skills/esp32s3-rlcd42/template/variants/new-project.sh <target-dir> [--full|--minimal]
cd <target-dir> && pio run

• --minimal — USB CDC console heartbeat, PSRAM check, button levels. 342,508 B flash, 24,256 B RAM. Flash this first on a board you have not used before: on a board with no LED it is the only way to separate "wrong USB flags" from "wrong peripheral code".
• --full (default) — board self-test: report, I2C scan, SHTC3, PCF85063A, battery, and a dashboard on the panel with KEY to redraw. 372,904 B flash, 39,928 B RAM.

Both build clean with zero warnings at the toolchain versions in the orientation table. Nothing is generated and no paths are embedded, so copying template/ by hand works identically. template/README.md maps files to subsystems.

The TF card and the audio codecs are deliberately not in the template — see reference/recipes.md §6 and §7, which say plainly which parts are untested.

When the user already has a project, prefer bringing their platformio.ini and pin header in line with the template over rewriting their sketch.

ESP-IDF is a reasonable alternative and every pin rule above applies unchanged; for audio, Espressif's esp_codec_dev component ships the ES8311 and ES7210 drivers, and ESPHome (framework esp-idf) is the path with the most published evidence behind it for this board.

Flashing

pio run -t upload -t monitor

One Type-C socket, native USB, no bridge chip. It enumerates as /dev/cu.usbmodem* (macOS), /dev/ttyACM* (Linux) or a COM port (Windows), VID 303A.

When upload fails with Wrong boot mode detected or No serial data received, check GP0 before the cable (rule 12). Then the manual sequence — note that there is no EN/RST button on this board:

hold BOOT → tap PWR (or replug the USB cable) → release BOOT → upload

Bad firmware cannot brick this board. The ROM bootloader is mask ROM and always answers that sequence:

esptool.py --chip esp32s3 -p <port> flash_id      # confirms 16 MB
esptool.py --chip esp32s3 -p <port> erase_flash

The one genuinely unrecoverable mistake is burning the VDD_SPI eFuse to 1.8 V on a 3.3 V module. eFuses are one-time. Do not run espefuse.py on this board.

Physical handling, from the vendor's own warning: do not push on the screen when inserting the Type-C cable or an 18650 — support the board by its frame. Screen cracks from that are explicitly outside the warranty.

Reporting

Say what is verified and what is derived. This skill was written from the board's schematic, the vendor's own source code, and a build — not from a board on a desk.

Schematic-derived, high confidence. Every pin in board-hardware.md §2.3 is traced to ESP32-S3-RLCD-4.2-schematic.pdf (files.waveshare.com/wiki/ESP32-S3-RLCD-4.2/). That includes the facts most likely to be got wrong elsewhere: the SD slot is SDMMC with R7 (its would-be CS to GPIO17) unpopulated and card-detect grounded; LCD_TE really is GPIO6; GPIO7/42/47/48 are connected to nothing; the battery divider is 200 K/100 K 1%; the amp enable has a 10 K pull-down; and CHIP_PU has no reset button on it.

Vendor code, high confidence. github.com/waveshareteam/ESP32-S3-RLCD-4.2 is Waveshare's own repo for this board (Arduino, ESP-IDF, ESPHome and XiaoZhi examples plus prebuilt firmware). The SD pins and 1-bit width, the ADC unit/channel/attenuation and ×3 factor, the 24 MHz SPI clock, the SPI.begin(sck, -1, mosi, -1) call, and 0x38 as the end of the display init all come from code in that repo. Their peripheral tutorial supplies the HPM/LPM behaviour and the ~45 fps figure.

Corroborated by a third party. The audio pin directions, ES8311 0x18 and ES7210 0x40 are confirmed by a Home Assistant community user working with the board in hand — who reports having had DOUT/DIN backwards at first, which is rule 10.

Primary-source. The strapping tables, the GPIO35/36/37 octal-PSRAM restriction, the ADC1/ADC2 maps and the power-up glitch list in reference/esp32s3-soc.md are from the Espressif ESP32-S3 and ESP32-S3-WROOM-1 datasheets.

Known contradictions between sources — quote them as contradictions, not as facts:

• The wiki's GPIO table is wrong about TP_INT/TP_RESET/TP_SDA/TP_SCL and carries a stray QMI8658C column. The schematic PDF's own copy of the same table has those cells blank. Rule 2.
• ST7305 vs ST7306. Waveshare's product spec, their driver, their datasheet copy and U8g2 all say ST7305; only the Zephyr board port says ST7306. Use ST7305.
• U8g2 version. Waveshare says "2.36.19 or later"; the PlatformIO registry stops at 2.36.18, which already has the constructors. Pin ^2.36.18 on PlatformIO.
• Battery curve. ESPHome tutorial 2.5–4.2 V; Waveshare's own firmware 3.0–4.12 V.
• RTC backup connector. The product page says PH1.0; the schematic symbol says SH1.0. Tell the user to measure before ordering a pigtail.

Verified in this session: both template variants build clean, zero warnings, with PlatformIO Core 6.2.0 / platform-espressif32 55.3.311 / Arduino core 3.3.11 / U8g2 2.36.18, at the flash and RAM figures quoted above.

Not verified: nothing was flashed or run. No panel was lit, no card mounted, no voltage measured, no frame rate observed, and no audio played — the ES8311/ES7210 register sequences are deliberately absent from template/ for that reason. Say so rather than presenting any of it as certain.