Supported simulators and hardware

NIR is currently supported by 8 simulators and 5 hardware platforms, allowing users to seamlessly move between any of these platforms. The 8 simulators include jaxsnn, Lava-DL, Nengo, Norse, Rockpool, Sinabs, snnTorch, and Spyx. The 5 hardware platforms include BrainScaleS-2, Intel Loihi (via Lava-DL), Speck, SpiNNaker2, and Xylo.

The table below shows the integration progress for the respective frameworks. By “reading” a NIR graph, we mean converting it into a platform-specific representation. “Writing” a NIR graph means converting a platform-specific representation into a NIR graph.

Framework	Write to NIR	Read from NIR	Examples
jaxsnn (BrainScaleS-2)	⬚	✓	jaxsnn examples
Lava-DL	✓	⬚	Lava/Loihi examples
Nengo	✓	✓	Nengo examples
Norse	✓	✓	Norse examples
Rockpool (SynSense Xylo chip)	✓	✓	Rockpool/Xylo examples
Sinabs (SynSense Speck chip)	✓	✓	Sinabs/Speck examples
snnTorch	✓	✓	snnTorch examples
SpiNNaker2	⬚	✓	SpiNNaker2 examples
Spyx	✓	✓	Spyx examples

Why are some platforms only reading or writing but not both?

Some platforms support both reading and writing, but in other cases it does not make sense to both read and write NIR graphs. For example hardware platforms are meant as a runtime of NIR graphs, so it rarely makes sense to convert the hardware representation back into NIR.

What about other simulators and hardware platforms?

NIR is a recent invention, and we are working hard to integrate it with as many simulators and hardware platforms as possible. If you know of a simulator or hardware accelerator we should include, please get in touch with us here on GitHub.