HiKey970 Resources

HiKey970 development resources and documentation

Summary

This page serves as the central resource hub for the HiKey970 development board. The HiKey970 is a 96boards Super Edition platform built around the HiSilicon Kirin 970 SoC, which integrates an octa-core CPU (4× Cortex-A73 + 4× Cortex-A53), a Mali-G72 MP12 GPU, and a dedicated Neural Processing Unit (NPU) capable of accelerating AI and machine learning inference workloads. Below you will find links to hardware documentation, operating system images, AI/ML development frameworks, and community resources.

Who This Is For

This resource page is designed for hardware engineers, embedded systems developers, AI researchers, and Android platform developers who are evaluating or actively building on the HiKey970. Whether you need schematics for custom carrier board design, AOSP build instructions, or NPU SDK documentation, the links below provide a structured starting point.

Hardware Documentation

  • Board schematics: Full schematic PDFs for the HiKey970 PCB, including power regulation, high-speed expansion connector pinout, and low-speed I/O header mappings.
  • Hardware user manual: Covers board layout, DIP switch functions, jumper settings, connector specifications, and thermal management guidance.
  • Mechanical drawings: Board dimensions, mounting hole positions, and keep-out zones for enclosure design.
  • 96boards compliance: The HiKey970 follows the 96boards Super Edition specification, ensuring compatibility with standardised mezzanine boards for cameras, sensors, and additional I/O.

Operating System Images

Pre-built operating system images are available for immediate deployment:

  • Debian Linux: Desktop and server variants with GPU and NPU driver support. See the HiKey970 System Images page for download links and flashing instructions.
  • Android AOSP: Full Android build with Kirin 970 HAL support. Suitable for application development, CTS testing, and HAL bring-up work.
  • UEFI Firmware: Required before any OS installation. Available on the HiKey970 Downloads page.

AI and Machine Learning Frameworks

The Kirin 970 NPU enables on-device neural network inference with significantly lower latency and power consumption compared to CPU-only execution. The following frameworks and tools support NPU acceleration on the HiKey970:

  • HiAI Foundation: HiSilicon's native NPU SDK for model conversion, optimisation, and deployment. Supports TensorFlow, Caffe, and ONNX model formats.
  • Android NN HAL: The Android Neural Networks API hardware abstraction layer routes compatible workloads to the NPU automatically when running Android AOSP.
  • TensorFlow Lite: Community-contributed delegate implementations allow TFLite models to leverage the NPU for inference acceleration under Linux.

AOSP Build Guides

Building Android from source for the HiKey970 follows the standard AOSP build process with board-specific configuration:

  1. Initialise the AOSP repository with the HiKey970 manifest.
  2. Sync the source tree using repo sync.
  3. Select the hikey970-userdebug lunch target.
  4. Build with make -j$(nproc).
  5. Flash the resulting images via fastboot.

Detailed build instructions, including prerequisite packages and known issues, are maintained in the 96boards AOSP documentation and the HiKey970 section of the LeMaker wiki.

96boards Compatibility

As a 96boards Super Edition board, the HiKey970 supports the standardised high-speed and low-speed expansion connectors. This enables the use of off-the-shelf mezzanine boards for MIPI CSI cameras, audio codecs, sensor arrays, and prototyping I/O. Verify mezzanine compatibility with the Super Edition pinout specification before purchasing third-party expansion boards.

Troubleshooting

  • Board does not power on: Use a 12 V / 2 A power supply connected to the barrel jack. USB-C power delivery alone may be insufficient for full operation.
  • NPU not detected: Ensure the NPU kernel module is loaded. On Debian, run lsmod | grep npu. Install the NPU driver package if missing.
  • AOSP build fails: Verify you are using the correct manifest branch and lunch target. Check available disk space (at least 150 GB recommended for a full AOSP tree).
  • Mezzanine board not recognised: Confirm the mezzanine is Super Edition compatible. Check device tree overlays for the correct I/O configuration.

Related Guides

Author: LeMaker Documentation Team
Last updated: 2026-02-10