Apple Neural Engine: Architecture, Programming, and Performance

A comprehensive technical guide reverse-engineers Apple's Neural Engine, the fixed-function matrix accelerator embedded in iPhones, iPads, and Mac chips since the A11 and M1 generations. The 302-page paper documents the engine's architecture, datapath, compiler, firmware, and command protocol across A11-A18 and M1-M5chip families through direct measurement and static analysis. While the undocumented direct dispatch route enables research and on-device work, Apple's officially supported Core ML framework remains the recommended path for shipping software.