Thanh Le / bench notes
EMBEDDED · BOARD HUB · IN PROGRESS

Raspberry Pi Zero W (BCM2835)

Bare metal with no SD card, a FreeRTOS shell on the boot cable, and the 12 QPUs driven from my own code

Last change2026-11-22: hub page with three STAR write-ups

Raspberry Pi Zero W (BCM2835)

Pi Zero W block diagram, drawn from the reduced schematic and the device tree

The Pi Zero W is the 65 x 30 mm one: a BCM2835 (the same chip as the very first Pi), one ARM1176 core, 512 MB of LPDDR2 stacked on top of the SoC, a CYW43438 for Wi-Fi and Bluetooth, one micro-USB for data and one for power. Mine has no header soldered and, for this whole project, no SD card and no UART cable either. Just the USB data cable to my PC.

3D model of my Pi Zero W My Zero W as a 3D model I made for the video series (no photo of mine yet): 65 x 30 mm, the BCM2835 with the 512 MB stacked on it, two micro-USB, no header.

The question I started with in October 2026 was simple: what does “bare metal” even mean on this chip? On a microcontroller the reset vector is yours. Here it isn’t. The VideoCore GPU boots first from its own ROM, runs closed firmware, and only then lets the ARM out of reset at 0x8000. I wanted to see how much of that I could poke at from my side of the cable.

Datasheet bus addresses vs ARM physical addresses The first thing that tripped me: the datasheet prints 0x7E... VideoCore bus addresses, my ARM code has to use 0x20.... Same GPIO, UART0 and AUX, 0x5E000000 apart.

What I got out of it: a 1381-byte bare-metal kernel booting over USB with no card (once I pinned an older firmware), then FreeRTOS with a 13-command shell on that same boot cable, after a long hunt for a USB device that kept leaving. Later the shell learned to run my own programs on the 12 QPUs, and the same rtos_cli image runs unchanged on my Pi 1 Model B (Pi 1 hub).

The stories

#StoryResultStatus
1 Booting a Pi Zero W with no SD card
My 1381-byte kernel over USB with rpiboot, a firmware regression, and a lsusb that breaks the next boot
Power-on to my code in ~10 s over USB; stable firmware fails, 1.20241126 works Done
2 A FreeRTOS shell on the boot cable, and the USB device that left after 170 ms
FreeRTOS + TinyUSB CDC on the Zero W with no UART cable; three bugs found one bit at a time
Shell on /dev/ttyACM0 ~0.8 s after hand-over; 11.3 us per task switch; tick jitter 999-1001 us Done
3 Twelve QPUs from bare metal, no Linux
Queueing my own programs on the VideoCore IV shader cores from FreeRTOS, and checking every lane against the ARM
4799 MFLOPS on 12 QPUs vs 73 on the ARM with VFP (66x); 192 lanes bit-exact Done

Still open

The camera. cam probe gets a NACK from both sensor addresses (OV5647 at 0x36, IMX219 at 0x10) because nothing is plugged in. The Unicam receiver registers read back as written, and the board’s own 640x480 test frame comes back to the PC in 1.0 s (my first try took 45.9 s), so the real frame just waits for a camera and a cable.

I still have no UART cable, so the bare-metal banner and its baud rate have never been seen on a wire. A crash only shows up as a reboot: I tried keeping a “black box” record in DRAM at 0x0F000000 across a watchdog reset and it was gone after the reset, every time. SD-card boot, Wi-Fi, the D-cache and the MMU I haven’t touched.