CORTEX: Large-Scale Brain Simulator Utilizing Indegree Sub-Graph Decomposition on Fugaku Supercomputer

TL;DR

CORTEX introduces Indegree Sub-Graph Decomposition to enable mutex-free, scalable large-scale brain simulation on the Fugaku supercomputer. By combining Area-Processes Mapping with Multisection Division and a thread-aware, delay-aware execution model, the framework achieves increased problem size while overlapping communication and computation. Verification against NEST and evaluation on a multi-scale marmoset model demonstrate improved memory efficiency and simulation throughput under realistic biological parameters. The approach advances neuromorphic computing by enabling closer-to-human brain-scale simulations and outlines domain-specific optimizations and future directions toward whole-brain, heterogeneous architectures.

Abstract

We introduce CORTEX, an algorithmic framework designed for large-scale brain simulation. Leveraging the computational capacity of the Fugaku Supercomputer, CORTEX maximizes available problem size and processing performance. Our primary innovation, Indegree Sub-Graph Decomposition, along with a suite of parallel algorithms, facilitates efficient domain decomposition by segmenting the global graph structure into smaller, identically structured sub-graphs. This segmentation allows for parallel processing of synaptic interactions without inter-process dependencies, effectively eliminating data racing at the thread level without necessitating mutexes or atomic operations. Additionally, this strategy enhances the overlap of communication and computation. Benchmark tests conducted on spiking neural networks, characterized by biological parameters, have demonstrated significant enhancements in both problem size and simulation performance, surpassing the capabilities of the current leading open-source solution, the NEST Simulator. Our work offers a powerful new tool for the field of neuromorphic computing and understanding brain function.

Figures (19)

• Figure 1: Schematic Representation of A Neuron and Its Spike: Illustrating the basic elements of (SNNs), where neurons are connected through synapses, forming a complex network. The "spike" serves as the primary unit of information processing in SNNs.
• Figure 2: Graph Abstraction of SNNs
• Figure 3: A Sub-graph of Indegree and Outdegree Format
• Figure 4: Synaptic Interactions on the Indegree Sub-graph: In Fig.\ref{['fig:in_spk_graph']}, spiking neurons with active synaptic interactions have been highlight with red border (No. 1 and No.6). At each time step, once the spiking graph has been well defined, synaptic interactions can take effect on corresponding post-synaptic neurons without dependency between different indegree sub-graphs.
• Figure 5: Synaptic Interactions on the Outdegree Sub-graph: In Fig. \ref{['fig:out_spk_graph']}, spiking neurons with active synaptic interactions have been highlight with red border (No. 1 and No.6). At this time step, there are 2 synaptic interactions taking effect on neuron 9 in 2 outdegree sub-graphs respectively. Between each synaptic interaction with its nonlinear dynamics, the states of all post-synaptic neurons with the same ID should be synchronized among all sub-graphs.