The Nengo team at ABR is excited to announce the first public release of our new Nengo Loihi backend (currently version 0.2). This backend for the Nengo neural simulator allows standard spiking Nengo models to run on Intel’s new Loihi neuromorphic chip.
While still under heavy development, this alpha release of our latest backend allows users to build recurrent, feedforward, and many layered networks, including deep neural networks. We’ve provided several examples in the documentation accompanying the software. As an alpha release, we expect there to be growing pains, and there are limitations to how efficiently networks are mapped on to the hardware, the size of networks supported, and so on. Nevertheless we believe that this is a useful tool for those interested in quickly and easily building spiking networks on hardware (even with the Nengo GUI). We hope the feedback that early users provide can direct our development efforts and make this backend the preferred method for programming Loihi.
Thanks in advance for any feedback and suggestions for future improvements!
Additionally, to run models on real Loihi hardware, you need physical access to a board or access to Intel’s Neuromorphic Research Community (INRC). Anyone can install Nengo Loihi, however, to run Nengo models on the software-only emulator.
Setting up Nengo Loihi for the first time
is more involved than other Nengo backends
if you are using Loihi hardware.
The Nengo Loihi documentation includes detailed steps for
setting up physical hardware,
installing Nengo Loihi on the INRC cloud server.
If you are using the emulator, then you can
Once installed, Nengo Loihi simulates Nengo networks. After defining your network, instead of simulating it with
And that’s it!