Neuromorphic computing is an exciting new area of computer architecture that takes inspiration from the brain to achieve huge improvements in power efficiency. Applications are often described in the form of Spiking Neural Networks (SNNs), which are similar to conventional neural networks, but where information is encoded in time and communicated as sparse time-varying “spikes”. SNNs have been applied efficiently to a range of applications, such as machine-learning and robotic control. Hardware architectures have been proposed that execute SNNs far more efficiently than CPUs or GPUs. However, creating better architectures requires tools to model design tradeoffs.

In collaboration with Sandia National Laboratories (SNL), an initial concept for a prototype simulator called SANA-FE was created to model the energy and performance of different spiking hardware platforms and aid with architecture design-space exploration. However, SANA-FE is slow, limited to small simulations and uses text-based inputs to describe hardware architectures.

This project extends SANA-FE by optimizing the code, creating graphical user interfaces (GUIs), and providing support for a wider range of neuromorphic architectures.

Team Members:

Ignacio Gonzalez
Kunaal Jha
Lance Lui
Parth Shroff
Robin Sam