[PERF] Optimize 50 Hz Cranial SNN Engine with Numba / WebAssembly Parallelization for 650k Neurons

[0xalydev]

Summary

The Danio biophysical simulation runtime (danio/brain/engine.py) models a 650,000 volume-constrained computational neuron population organized into 203 anatomical regions. The current vectorized Python/NumPy leaky integrate-and-fire (LIF) loop runs at 50 Hz, but per-step execution on single-thread CPU can reach 14–18 ms during high-density sensory volleys.

Goal

Accelerate the core numerical integration loop down to $< 2.5$ ms per step using:

1. Numba JIT compilation with @njit(parallel=True, fastmath=True) on membrane potential updates:
   $$V_m[t+1] = V_m[t] + \frac{dt}{\tau} \cdot (V_{rest} - V_m[t] + R \cdot I_{syn}[t])$$
2. Or a lightweight C++/Rust WebAssembly module for zero-copy in-memory tensor updates.

Key Files

• danio/brain/engine.py: Core DanioBrain runtime loop.
• danio/simulation.py: Biophysical closed-loop coordinator.
• tests/test_research_grade_simulation.py: Determinism test suite.

Acceptance Criteria

• Step time benchmarked at $< 3$ ms for 650,000 neurons on standard 4-core desktop.
• Determinism preserved (passes tests/test_research_grade_simulation.py::test_engine_determinism).
• Zero memory leakage over 10,000 continuous simulation steps.