GitHub
// AI-native operating system · v0.5.0

Inference is the primary workload.
Not an add-on.

SynapticOS is an inference-first runtime for microcontrollers with neural accelerators — built on Zephyr RTOS. Memory, scheduling, and hardware abstraction are all designed around one thing: the data flow of a neural network.

DUAL CORTEX-M33 @ 150 MHz eIQ NEUTRON NPU · 4.8 GOPS APACHE-2.0 · OPEN SOURCE
15 µsCross-core IPC round-trip · FRDM
10 µsLayer-preemption context save · FRDM
158/158Tests passing · 21 suites
+251%DMA ingest vs CPU copy · FRDM
0Errors in an 11,106-job soak
$15Target board · MCXN947
// Why

Most RTOS platforms treat AI inference as an afterthought.

The status quo

A neural inference library bolted onto a scheduler that was designed for control loops and sensor polling. On microcontrollers with dedicated NPUs, that mismatch wastes silicon.

The NPU sits idle while the CPU copies buffers. The memory allocator fragments the heap under tensor churn. And there is no clean way to manage model lifecycles or pipeline data from sensor to prediction.

The SynapticOS answer

  • >_
    The scheduling unit is the inference jobPriority- and deadline-aware, preemptable at NPU layer boundaries — not a generic thread.
  • >_
    Tensor-aware memory, zero fragmentationA bump allocator with persistent, ephemeral, and scratch regions — DMA-aligned by construction.
  • >_
    Hardware-agnostic accelerationOne HAL for NPU, DSP, and DMA. The same application runs on Neutron silicon or a deterministic QEMU stub.
  • >_
    Models are a managed lifecycleVersioned, registered, hot-swappable — persisted in flash with power-loss-safe A/B OTA updates and rollback, verified on the board.
// Live on hardware

The OS talks back.

Verbatim serial capture from a FRDM-MCXN947 over MCU-LINK — boot to interactive shell in about a millisecond.

picocom -b 115200 /dev/ttyACM0 — FRDM-MCXN947
<inf> syn_mpu: Cross-core guard: 0x30058000 +64 KB read-only (MPU region 7) *** Booting Zephyr OS build v3.7.0 *** <inf> dual_model: SynapticOS dual_model (CPU0, AI runtime) <inf> syn_mem: Arena: 128 KB total, 112 KB tensor, 16 KB scratch <inf> syn_ipc: IPC ready (CPU0, ring 16 entries, region 98304 bytes) <inf> syn_remote: Cross-core inference serving enabled <inf> syn_boot: Releasing CPU1 (vector table at 0x00100000) <inf> dual_model: Dual-core up: CPU1 boot 1514 us, handshake 2554 us <inf> syn_remote: Served #1: model 1 prio 2 in 3470 us (10 out bytes)   uart:~$ syn ipc status CPU1 link: UP CPU1 boot time: 1514 us (release to ready) IPC handshake: 2554 us (release to STATUS_REQ) Inferences served: 1267 (errors 0, avg 2290 us) IPC round-trip (CPU1-measured, 64 samples): last 15 us, min 15 us, max 81 us uart:~$ syn model list Registered models: 2 [1] face_detect v1.0.0 (loaded) [2] keyword_spot v1.0.0 (loaded) uart:~$

This terminal is live — click it and type help; responses mirror the v0.5.0 release (2026-08-10 board verification, dual_model sample — try syn health or syn infer stats). · Transcript above: 2026-07-14 v0.3.0 board verification · stub NPU backend (Neutron SDK integration lands in a later phase)

// Hardware

Runs on a $15 board.

The NXP FRDM-MCXN947 — two Cortex-M33 cores, a neural accelerator, a DSP coprocessor, and smart DMA. SynapticOS assigns each one a job.

BoardNXP FRDM-MCXN947
CPU2× Arm Cortex-M33 @ 150 MHz
NPUeIQ Neutron · 4.8 GOPS INT8
DSPPowerQuad coprocessor
Memory512 KB SRAM · 2 MB dual-bank flash
Price~$15 USD
CPU0 — AI COREFull SynapticOS runtime. Exclusive NPU ownership. Owns the tensor arena. Boots first, verifies CPU1's image, releases it.
⟵⟶MAILBOX IPC
15 µs measured
CPU1 — APP COREApplication logic. No FPU, DSP, or MPU — by silicon. Requests inference over lock-free shared-memory rings.
0x2000_00000x2004_00000x2005_8000
// Roadmap

Six phases to v1.0.

Open source now, engineered like a product. Phases 1 through 5 have shipped — all verified live on the FRDM-MCXN947 and tagged as releases. Next up: Phase 6, ecosystem and tooling toward v1.0.

Foundationv0.1.0 ✓Memory, HAL, registry, shell, tests
Inference Pipelinev0.2.0 ✓Pipeline engine, priority scheduler, processors
Dual-Core & IPCv0.3.0 ✓Asymmetric multiprocessing, cross-core inference, MPU guard
Model Lifecyclev0.4.0 ✓Flash-backed model store, power-loss-safe A/B OTA, hot-swap
Production Hardeningv0.5.0 ✓Layer preemption, watchdog + fault recovery, zero-copy DMA, binary OTA, 11k soak
Ecosystemv1.0.0 · nextSDK, tooling, docs, packaging
// Research

Built in the open — argued on paper.

SynapticOS: An Inference-First Runtime Architecture for Neural Processing Units on Resource-Constrained Microcontrollers

D. KAFETZIS · ARXIV PREPRINT · cs.AR / cs.SE — publishing with Phase 1

The foundation paper makes the architectural case: why the RTOS abstractions we inherited from control systems fail inference workloads, and what a runtime looks like when the tensor pipeline comes first. Every number is measured, and stub-baseline results are labeled as exactly that.

  • C·1Tensor-aware memory architecture — lifetimes as regions, not a heap
  • C·2State-machine HAL for NPU + DSP with deterministic fallbacks
  • C·3Model lifecycle registry with versioning and integrity checks
  • C·4Integrated profiling at inference-pipeline granularity
  • C·5Open implementation — ~4,100 lines of C, Apache-2.0
// Get involved

Open source now.
Built to grow.

Clone it, build it for QEMU in one command, flash it to a $15 board. Contributions open in an upcoming phase — until then, stars, issues, and hard questions are very welcome.