Hardware-efficient architecture of spiking neural networks based on sign-magnitude stochastic computing

Key Points

• The architecture demonstrates significant reductions in energy consumption during computations for spiking neural networks.
• It was shown that using sign-magnitude representation in stochastic computing improved performance metrics by 35%.
• Evaluation involved assessments of various hardware platforms, indicating broad applicability across systems.
• The findings may enable more scalable and energy-efficient designs in neural network implementations, optimizing hardware usage.