PAX: A mixed hardware/software simulation platform for spiking neural networks

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering
• Hardware Accelerators for AI

Background:

• Existing hardware simulators for bio-like neural networks often combine digital and analog computation.
• Spiking Neural Networks (SNNs) with dynamic adaptation rules, like Spike-Timing-Dependent Plasticity (STDP), are computationally intensive.
• Simulating neural networks in biological real time presents significant hardware challenges.

Purpose of the Study:

• To present a novel mixed hardware-software platform for simulating spiking neural networks (SNNs).
• To detail the design and implementation of analog neuromimetic integrated circuits for SNN simulation.
• To validate the platform's performance and utility in computational neuroscience research.

Main Methods:

• Development of a configurable mixed hardware-software platform for SNN simulation.
• Utilization of conductance-based neuron and synapse models with Spike-Timing-Dependent Plasticity (STDP).
• Implementation of analog neuromimetic integrated circuits for core computational tasks.
• Ensuring simulation in 'biological real time' (time difference < 50 µs).

Main Results:

• Experimental validation of the platform's system performance and accuracy.
• Demonstration of configurable neurons and networks operating in biological real time.
• Successful application of the platform in collaborative computational neuroscience projects.

Conclusions:

• The presented mixed hardware-software platform offers an efficient solution for simulating SNNs.
• The platform enables high-fidelity, real-time simulations crucial for advancing computational neuroscience.
• It serves as a valuable tool for interdisciplinary research in neurobiology and computer science.