Course · Computer Science
Qualcomm Hexagon NPU
Complete reference for building custom inference engines on Hexagon NPU. Covers VLIW architecture, HVX vector programming, HTA/HMX tensor accelerators, and on-device inference optimization.
10 modules≈ 1 h reading
01Hexagon SoC Architecture Internals60 min02HVX (Hexagon Vector eXtensions) Programming Model4-6 hours03HTA / HMX (Tensor & Matrix Accelerators)4-5 hours04Memory Subsystem Mastery4-5 hours05Quantization for Hexagon4 hours06Operator Kernel Design5 hours07Inference Engine Architecture5 hours08Performance Analysis & Profiling4 hours09Toolchain & Development Environment5-6 hours10Research Frontiers & Advanced Topics4 hours