Physics-AI Integration on Unified Memory: Zero-Copy Pipeline Between Particle Simulations and Neural Networks on Apple Silicon

Proposes a zero-copy architecture that eliminates the CPU/GPU data transfer bottleneck in Physics-AI workloads by leveraging Apple Silicons unified memory. Describes a pipeline where particle simulation data (OpenFPM/Metal) resides in shared memory that MLX neural networks can read in-place, enabling real-time physics-ML feedback loops impossible on discrete GPU architectures.