Hierarchical Memory SNN: 1000x RAM Reduction through Brain-Inspired Working/Long-term Memory Architecture

I propose a novel hierarchical memory architecture for Spiking Neural Networks (SNNs) that achieves approximately 1000x reduction in RAM usage. Inspired by the brain's separation of working memory and long-term memory, this approach stores "hot" neurons in RAM while maintaining "cold" neurons in memory-mapped SSD storage.

**Key Results (v1):**
- 1,000,000 neurons with only 4MB RAM + 61MB SSD
- 1000x RAM reduction vs. conventional approaches  
- PPL 1.39 on character-level language modeling
- Enables deployment on Raspberry Pi, smartphones, and edge devices

This opens new possibilities for energy-efficient AI at the edge.