MCU or FPGA, Zephyr or Linux: framing the VexRiscv question

I started this project meaning to investigate “the MCU–FPGA connection” through two repos. The first surprise: neither repo is about wiring an MCU to an FPGA. They are the two ends of a different axis — Zephyr RTOS vs embedded Linux — and seeing that is what makes the project worth writing about.

The two repos, honestly

linux-on-litex-vexriscv runs full Linux on VexRiscv-SMP, a 32-bit RISC-V soft core synthesised into the FPGA. LiteX assembles the SoC and the peripherals (DRAM, Ethernet, SD); OpenSBI and Buildroot do the rest. There is no separate microcontroller here — the CPU is the fabric.

AkiraOS is the opposite world: Zephyr 4.3 on ordinary MCUs (ESP32-S3, nRF54L15, STM32), running sandboxed WebAssembly apps over a stable firmware base. No FPGA in sight.

The reframe

The two questions I’d been conflating are independent:

• MCU vs FPGA — is the CPU hard silicon, or soft logic in an FPGA?
• Zephyr vs Linux — a small real-time RTOS, or a full MMU OS?

They decouple because Zephyr ships an official litex_vexriscv board: the same VexRiscv soft core boots either Linux or Zephyr. Antmicro even ran TF Lite Micro on Zephyr/LiteX/VexRiscv.

So there are three architectures to actually compare — soft-core Linux, soft-core Zephyr, and a true hybrid where a real MCU (AkiraOS-style) drives a LiteX FPGA over SPI/Wishbone/PCIe. The last one is the literal “MCU–FPGA connection,” and it’s the one neither repo hands you.

(Draft — see README.md in this folder for the comparison table, the open questions, and where the GraphicsCourse video angle plugs in.)