Neuronova SDK — Early Access

This is an early-access release of the Neuronova SDK, designed to give you hands-on access to a new computing paradigm: always-on, ultra-low-power analog AI at the edge.

Some features may be incomplete, APIs may evolve, and parts of the experience will change as the technology matures. This is expected at this stage.

We’re releasing the SDK early on purpose. We believe the fastest way to build the right platform is to build it together with the people who are pushing real application to the real world, not in isolation.

Your feedback is extremely important to us. What works, what doesn’t, what feels missing, all of it directly influences our roadmap, our tooling, and the next iterations of both software and silicon.

This is not just about testing an SDK.
It’s about shaping a new always-on intelligence layer at the edge.

Thank for being with us!

OK

Releases & Downloads

Choose the SDK release you want to use, open the matching documentation site, and download the packaged bundle.

Download bundles

This bundle contains the SDK and the dockerfiles to build the compatible environments.

Version	Download ZIP
nwave-001a	nwave-001a.zip
nwave-002a	nwave-002a.zip
nwave-100a	nwave-100a.zip

Note

Version 100a contains custom HIP kernels. Due to limited compatibility of CUDA kernels, we can support only GPUs with compute capability greater than 86. If you have older NVIDIA GPUs, please use the cpu device for training.

Install nwave-100 (full package, first installation)

After downloading and extracting nwave-100a.zip, use the automated installer included in the package.

For more instructions, follow the reference page.

For users without GPU be sure that you install the cpu-only environment.

Update NWAVE from previous version

After downloading and extracting a bundle zip, move the whl file based on your GPU vendor (rocm for AMD, cuda for NVIDIA) nwavesdk-*.whl to the shared folder of your docker environment. Start a terminal inside the environment and run these two commands:

pip uninstall nwavesdk

and then install the whl package via:

pip install nwavesdk-*.whl

For users without GPU you can use any whl version.

---

Version compatibility

NWAVE SDK	SDK-Environments	Tutorials
1.0.0	0.0.1	1.0.0
0.0.2	0.0.1	0.0.2
0.0.1	0.0.1	0.0.1

Tutorials

Version	Download ZIP
Tutorials for release nwave-001a	tutorial-nwave-001a.zip
Tutorials for release nwave-002a	tutorial-nwave-002a.zip
Tutorials for release nwave-100a	tutorial-nwave-100a.zip

---

Changelog

NWAVE SDK 1.0.0 — 2026-04-29

First major release in alpha version. Note: not compatible with older releases because of the new version of H1 chip (named H1v2).

New features

• New H1 chip model now is possible to simulate the new version of H1:
  • New reset mechanism
  • No synaptic mismatch variability (easier training)
• Refactor of older layers from hw to h1v1 because of the new architecture available (functions of H1 are unchanged, for older usages is sufficient a refactor of layer's name to expect same behaviors)
• HPO module to simulate multiple networks in parallel and optimize hyper parameters initialization
• New ad-hoc weight initialization based on data features
• Surrogate schedulers to help controllability on gradients during training
• Weight scheduler to reach weight deployable to chip in an faster convergence than losses
• CPU-only environments and SDK [EXPERIMENTAL]

Bug fixes

• Test 3 of deployment check of network fixed

Documentation

• More introductory tutorials to show neurons behaviors

Known Bugs

• Depending on the data used, if taus are too close to simulation timestep dt numerical problems may occours. Ways to avoid those are either to reduce dt or increase taus.

---

NWAVE SDK 0.0.2 — 2026-03-26

New features

• GPU (HIP) path for HWLayer — the hardware neuron model can now run fully on GPU via a native HIP kernel, enabling accelerated simulation of hardware-faithful networks. LIFLayer also gained a HIP path as part of the same work.
• Multithreaded data generation for faster preprocessing pipelines
• Auto-compilation script included in the package

Bug fixes

• Fixed gradient return when detach_reset=True in HWLayer

Known Bugs

• Setting membrane learning (._mem_learn=False) to False in HWLayer and LIF seems not to be affecting training.

Documentation

• Versioned docs structure (001a / 002a)
• Added tutorial download bundles
• Added users section

---

NWAVE SDK 0.0.1 — 2026-02-06

Initial release.