ExaFlexHH: an exascale-ready, flexible multi-FPGA library for biologically plausible brain simulations.

Area of Science:

• Computational Neuroscience
• High-Performance Computing
• Neuroscience Simulation

Background:

• In-silico simulations are crucial for understanding complex brain systems.
• Existing platforms often lack the performance, scalability, or flexibility needed for detailed models like Hodgkin-Huxley (HH) or gap junctions.

Purpose of the Study:

• To introduce ExaFlexHH, an exascale-ready, flexible library for simulating HH models on multi-FPGA platforms.
• To address the limitations of current simulation tools by providing high performance, scalability, and ease of use.

Main Methods:

• Developed ExaFlexHH utilizing FPGA-based Data-Flow Engines (DFEs) and the dataflow programming paradigm.
• Implemented a demanding extended-Hodgkin-Huxley (eHH) model of the Inferior Olive to demonstrate performance.
• Ensured parameterizability and NeuroML compliance for broad usability.

Main Results:

• Demonstrated linear scalability for unconnected networks and near-linear scalability for networks with synaptic plasticity.
• Achieved a 1.99x performance increase with two FPGAs and 7.96x with eight FPGAs.
• Showcased consistent GFLOPS per watt efficiency, indicating exascale-ready computing.

Conclusions:

• ExaFlexHH offers superior resource efficiency compared to other FPGA-based simulation implementations.
• The library pushes the boundaries of brain simulation platforms, facilitating exascale computing speeds.
• ExaFlexHH meets the criteria for high performance, scalability, flexibility, and ease of use for complex neuroscience models.