What is Neulite Kernel?

Overview

The Neulite kernel is a lightweight simulator for biophysical neuron models and network models. It specializes in simulating neuron models from the Allen Cell Types Database.

Features

Lightweight Design

Neulite features a clean codebase written in C17, providing a lightweight and highly extensible design.

High Portability

Neulite runs on a wide range of architectures from Raspberry Pi to supercomputers. This high portability enables the separation of network construction and simulation execution environments. For example, the network construction part (Bionetlite) with many dependent libraries runs locally, and only the generated files and Neulite kernel can be used on supercomputers.

Technical Details

Design Philosophy

The Neulite design is influenced by UNIX philosophy and follows these principles:

Emphasis on Simplicity: Avoid complexity and aim for easily understandable code
Modularity: Each component has independent functionality
Portability: Design that is not dependent on specific environments

Target Neuron Model

Neulite specializes in neuron models from the Allen Cell Types Database that use the Perisomatic model. For details on this model, see Specifications and Limitations.

Supported Environments

Recommended Environment

Environment: Linux, macOS, etc.
Compiler: C compiler supporting C17 (gcc 7.0 or higher recommended)

Achievements on Fugaku

Neulite has been optimized for the Fugaku supercomputer and has achieved microscopic-level simulation of a whole mouse cortex containing 9 million biophysical neurons and 26 billion synapses (see Publication).

Employed Technologies:

Scalable Vector Extensions (SVE): Full utilization of SIMD vector computation
Parallel Implementation: Efficient simulation of large-scale networks
MPI Support: Inter-node parallel processing