What question did this study set out to answer?

January 22, 20260 citationsOpen Access

Von Neumann vs Brain-like Architecture: A Quantitative Comparison of Information Capacity and Efficiency

Key Points

This research quantitatively evaluates the information capacity, energy efficiency, and fault tolerance of von Neumann and brain-like computing architectures.
Compared information capacity of four coding schemes: rate, phase, burst, and correlation coding.
Assessed energy efficiency by measuring bus transfer elimination.
Evaluated fault tolerance under component failure scenarios.
Utilized findings from the Blue Brain Project to analyze high-dimensional structures.
Brain-like architecture shows up to 9.3×10^13× more information capacity with burst coding.
Demonstrated 100× more energy efficiency compared to von Neumann architecture.
Achieved graceful degradation of performance with up to 30% component failure.
9D Hypercube structure achieved 90.6% accuracy in MNIST classification.

Abstract

I present a comprehensive quantitative comparison between von Neumann and brain-like computing architectures, focusing on information capacity, energy efficiency, and fault tolerance. **Key findings (v1): **- Information Capacity: Brain-like architecture achieves up to 9. 3×10¹3× more capacity using burst coding (phase + inter-spike interval) - Energy Efficiency: 100× more efficient due to elimination of bus transfers- Fault Tolerance: Graceful degradation up to 30% component failure- Real-World Tasks: 4/4 correct vs 3/4 for von Neumann **NEW in v2 - 11-Dimensional Brain Structure: **- Based on Blue Brain Project's discovery of 11D clique structures in the brain- 11D Hypercube achieves 8× faster information propagation than 2D grid- MNIST 10-digit classification: 9D Hypercube achieves 90. 6% accuracy (best) - Base-10 shows maximum advantage (+4. 2%) for hypercube topology- Parameter efficiency: 186× reduction compared to full connection This may explain why humans naturally developed the base-10 numeral system: 10 fingers + 11-dimensional brain = optimal for 10 categories. Four coding schemes compared: 1. Rate Coding: 2. 5×10¹0× capacity2. Phase Coding: 10¹0× capacity3. Burst Coding: 9. 3×10¹3× capacity (novel finding) 4. Correlation Coding: 10⁹× capacity This work provides theoretical foundation for the advantages of temporal coding and high-dimensional topology in brain-inspired computing systems, with reproducible Python simulations. GitHub: https: //github. com/hafufu-stack/brain-vs-neumann

Ask AI

Helpful

Bookmark

View Full Paper