Google Coral NPU: ML accelerator core designed for energy-efficient edge AI
3 days ago
- #RISC-V
- #AI Accelerator
- #Google Research
- Coral NPU is a hardware accelerator for ML inferencing, designed by Google Research as an Open Source IP.
- It is freely available for integration into ultra-low-power SoCs for wearable devices like hearables, AR glasses, and smart watches.
- Based on the 32-bit RISC-V ISA, Coral NPU is a neural processing unit (NPU) or AI accelerator.
- It includes three processor components: matrix, vector (SIMD), and scalar.
- Features include RV32IMF_Zve32x RISC-V instruction set, 32-bit address space, four-stage processor, and in-order dispatch.
- Memory includes 8 KB ITCM and 32 KB DTCM, both single-cycle-latency SRAM.
- Supports AXI4 bus interfaces for external memory interaction and external CPU configuration.
- Compatible with Bazel 6.2.1 and Python 3.9-3.12 (with 3.13 support in progress).
- Includes commands for testing, building binaries, and running simulations.