Skip to main content

Nexchip Dragon Series

Vendor: Nexchip

Category: Edge Computing / Autonomous Driving

Architecture: ARM + NPU

Introduction

Focus on high-end automotive chips. The flagship chip features a high-compute NPU designed specifically for AI inference tasks in smart cockpit and autonomous driving scenarios.

Specifications

ModelComputeMemoryInterfaceTDPProcess
Dragon One8 TOPS (INT8)8GB LPDDR5Automotive-grade SoC15W7nm
Dragon Two16 TOPS (INT8)16GB LPDDR5Automotive-grade SoC20W7nm

Official Website

Visit Official Website

OS Support

WindowsLinuxmacOSAndroid

Version History

VersionRelease DateDescription
Dragon One2023First automotive-grade AI chip, AEC-Q100 certified
Dragon Two2025NPU compute doubled, supports L4 autonomous driving

Performance Benchmarks

ModelTaskPerformance Metric
Dragon OneSmart Cockpit Multimodal Interaction8 TOPS, multi-camera support
Dragon TwoAutonomous Driving BEV Perception16 TOPS, L4-level support

Pricing

ModelReference PriceNotes
Dragon OneContact vendorSupplied to automakers (OEM)
Dragon TwoContact vendorIn mass production 2025

Quick Installation

Linux (Automotive System)

# 1. Download Nexchip SDK (requires registered developer account)
# Visit https://www.nexchip.com.cn/ to obtain SDK

# 2. Set environment variables
export NEXCHIP_SDK_PATH=/opt/nexchip-sdk

# 3. Verify NPU
npu-info

Code Examples

C/C++ (Automotive NPU Inference)

#include <nexchip_npu.h>

// Initialize NPU inference engine
npu_handle_t handle;
npu_init(&handle, NPU_MODEL_AUTO);

// Run inference
npu_tensor_t input = npu_create_tensor(input_data, dims);
npu_tensor_t output = npu_infer(handle, input);

// Release resources
npu_release(handle);

Architecture Highlights

  • ARM + NPU Heterogeneous: CPU and NPU deeply integrated, low-power real-time inference for automotive
  • Automotive-Grade Certification: AEC-Q100 Grade 2 certified, meeting stringent automotive environmental requirements
  • Unified Memory Architecture: CPU and NPU share LPDDR5, reducing data transfer overhead

Model Compatibility

Model/FrameworkSupportNotes
TensorFlow Lite✅ NativePrimary inference framework
ONNX RuntimeCommonly used in automotive
PaddlePaddle Lite⚠️Under adaptation
Proprietary Quantization ToolchainINT8/INT4 quantization support

If you're evaluating alternatives, the following products may also fit your scenario: