Platform

The Neurmorph platform stack

Three integrated layers — Compiler, Runtime, and HAL — form a complete toolchain from trained PyTorch model to silicon-resident spike-coded executable. Each layer is independently versioned and testable.

Get Started API Reference

Neuromorphic edge computing platform visualization

Architecture

Software stack layers

NMC Compiler

Graph optimizer for spiking workloads

NMC takes a torchscript or ONNX graph and runs six optimization passes: dead-spike elimination, temporal op fusion, rate normalization, threshold calibration, HAL mapping, and binary serialization. Output is a .snn binary that loads directly into the Edge Runtime.

Compiler deep dive →

nmc-cli

$ nmc compile \
  --model resnet18.pt \
  --target nt3000 \
  --output model.snn
[1/6] dead-spike elimination  ...done
[2/6] temporal op fusion      ...done
[3/6] rate normalization      ...done
[4/6] threshold calibration   ...done
[5/6] HAL mapping (nt3000)    ...done
[6/6] binary serialization    ...done
compiled model.snn  0.82 mW / 1.4 ms

Edge Runtime

Event-driven execution engine

The Edge Runtime manages spike queues, event timestep windowing, power-gated idle states, and memory pool allocation on-chip. No RTOS dependency. Written in C99 with platform-specific HAL shims — portable to any bare-metal environment where .snn executables can run.

Runtime architecture →

Event-driven dispatch

Only fires computation when spike events arrive — zero-power idle between events.

Static memory allocation

All buffers allocated at load time — no heap fragmentation, deterministic latency.

Watchdog integration

Hardware watchdog hooks with configurable spike-timeout recovery.

Hardware

Supported neuromorphic targets

Chip	Manufacturer	Neuron Count	Peak Power	HAL Status
NT3000	Synthetic reference	256K	1.2 mW	Supported
BrainPulse-2	Synthetic reference	128K	2.4 mW	Supported
SynCore-V	Synthetic reference	512K	0.8 mW	Supported
NT4000 (next-gen)	—	1M	TBD	Beta Q3 2026

Full hardware compatibility →

Ready to compile your model?

Start with the Quickstart guide to install NMC CLI, compile your first PyTorch model, and deploy to an evaluation target.

Quickstart Guide Request Pilot Access