NoC simulation steered by NEST: McAERsim and a Noxim patch.

Area of Science:

• Computational neuroscience
• Computer engineering

Background:

• Understanding the brain's computational mechanisms requires simulating large-scale neural networks.
• Simulating these complex models presents challenges for existing platforms, particularly regarding inter-node communication latency.
• Faster-than-real-time simulations are crucial for studying slow neural processes like learning.

Purpose of the Study:

• To present two novel simulation frameworks that interface network-on-chip (NoC) simulators with the NEST (Neural Simulation Tool) environment.
• To enable direct definition of network traffic by neural network models for steering NoC simulations.
• To obtain statistics on network latencies and explore accelerated simulation possibilities.

Main Methods:

• Developed two simulation frameworks integrating NoC simulators with the NEST environment.
• Utilized neural network models to define and steer NoC simulations, generating network traffic.
• Applied frameworks to scaled versions of the cortical microcircuit model to analyze performance, latency, and traffic.
• Implemented a time stamp conversion mechanism to emulate acceleration factors.

Main Results:

• The frameworks successfully generated network traffic based on neural network models and provided latency statistics.
• Performance curves, latency, and traffic distributions were determined for scaled cortical microcircuit models.
• The second framework, featuring tree-based multicast support and torus topology, yielded optimal results.
• Results align with previous findings on node internals, suggesting accelerated simulations are achievable.

Conclusions:

• The developed simulation frameworks offer a viable approach for studying large-scale neural networks and their communication dynamics.
• The integration with NEST and NoC simulators facilitates efficient, accelerated simulations of complex brain models.
• Further refinement, considering NoC simulator assumptions, can enhance the accuracy and applicability of these simulation techniques.