A comprehensive workflow for general-purpose neural modeling with highly configurable neuromorphic hardware systems

Area of Science:

• Neuroscience
• Computer Science
• Hardware Engineering

Background:

• Emerging accelerated and configurable neuromorphic hardware systems present new requirements for modeling tools.
• Existing platforms may lack the flexibility and accessibility needed for neuroscientific research.

Purpose of the Study:

• To present a methodological framework for operating advanced neuromorphic hardware.
• To establish a neuromorphic system as a flexible, neuroscientifically valuable modeling tool for non-hardware experts.
• To detail a device with 45 million programmable synapses and its operational challenges.

Main Methods:

• Integration of hardware interface into the PyNN (Python Neural Network simulator) model description language.
• Automated translation between PyNN and hardware configurations.
• Executable specification of the neuromorphic system as a test bench.
• Evaluation scheme using a benchmark library to compare hardware and software simulation results.

Main Results:

• A comprehensive hardware-software workflow is established, supporting ongoing preparative studies.
• The developed ecosystem facilitates hardware design process improvements.
• The model-to-hardware mapping software demonstrates maturity and flexibility through experimental validation.

Conclusions:

• The proposed framework and workflow enable the effective utilization of advanced neuromorphic hardware for neuroscientific modeling.
• This approach bridges the gap between hardware development and practical application by non-experts.
• The system proves crucial for advancing neuromorphic computing and its application in brain research.