Coral NPU is a hardware accelerator for ML inferencing. Coral NPU is an Open Source IP designed by Google Research and is freely available for integration into ultra-low-power System-on-Chips (SoCs) targeting wearable devices such as hearables, augmented reality (AR) glasses and smart watches.
Coral NPU is a neural processing unit (NPU), also known as an AI accelerator or deep-learning processor. Coral NPU is based on the 32-bit RISC-V Instruction Set Architecture (ISA).
Coral NPU includes three distinct processor components that work together: matrix, vector (SIMD), and scalar.
Coral NPU Architecture Datasheet
Coral NPU offers the following top-level feature set:
- RV32IMF_Zve32x RISC-V instruction set (specifically
rv32imf_zve32x_zicsr_zifencei_zbb) - 32-bit address space for applications and operating system kernels
- Four-stage processor, in-order dispatch, out-of-order retire
- Four-way scalar, two-way vector dispatch
- 128-bit SIMD, 256-bit (future) pipeline
- 8 KB ITCM memory (tightly-coupled memory for instructions)
- 32 KB DTCM memory (tightly-coupled memory for data)
- Both memories are single-cycle-latency SRAM, more efficient than cache memory
- AXI4 bus interfaces, functioning as both manager and subordinate, to interact with external memory and allow external CPUs to configure Coral NPU
- Bazel 6.2.1
- Python 3.9-3.12 (3.13 support is in progress)
# Ensure that test suite passes
bazel run //tests/cocotb:core_mini_axi_sim_cocotb
# Build a binary
bazel build //examples:coralnpu_v2_hello_world_add_floats
# Build the Simulator (non-RVV for shorter build time):
bazel build //tests/verilator_sim:core_mini_axi_sim
# Run the binary on the simulator:
bazel-bin/tests/verilator_sim/core_mini_axi_sim --binary bazel-out/k8-fastbuild-ST-dd8dc713f32d/bin/examples/coralnpu_v2_hello_world_add_floats.elf