CH37 Series

Edge AI SoC

Featuring a proprietary architecture and highly integrated single-chip design,, this SoC consolidates CPU, GPU, NPU, GPGPU, ISP, DSP, VPU and other processing units. Targeted at the general market segments with high computing power demands such as embodied intelligence and edge computing, it delivers standout strengths including sufficient AI computing power, flexible computing precision, and multimodal perception fusion.

CPU

8 CORES

Computility

64 TOPS

Memory

LPDDR 4/4X/5

Interfaces

PCIe/XGMAC/Rapid IO

Features & Specifications

High-Performance General-Purpose AI

Equipped with a 48TOPS high-compute NPU, deeply optimized for performance and algorithms across four key application scenarios.

Integrates 1TFlops of high-performance general-purpose computing power, supporting multiple precision formats including FP64, FP32, FP16 and INT8. It is compatible with mainstream inference frameworks and the CUDA ecosystem.

Incorporates a dedicated hardware accelerator for FFT.

Multispectral

Equipped with dual ISP supporting visible light, near-infrared (NIR) and far-infrared (FIR), along with a host of proprietary imaging algorithms including dual-spectrum fusion, AINR, EIS, hardware stitching and ultra-low-light full-color imaging.

High Integration

Integrates an octa-core 64-bit heterogeneous processor, two dedicated real-time cores, NPU, GPU, GPGPU, VPU, DSP, and more.

Integrates a rich set of high-speed and low-speed interfaces.

Full-Stack Solution

Provides a comprehensive product, software and hardware solution.

Features modular product design, with software SDKs deeply optimized for specific application scenarios.

NPU	Integrates an AI acceleration engine comprising six NPU cores, delivering up to 48 TOPS of computing power; Integrates a DSP that can be used for operator extension or as a standalone processor, with 64 GFLOPS of computing power; Integrates a dedicated FFT hardware accelerator;
GPGPU	Delivers up to 1 TFLOPS@FP32; Supports precision formats including FP64, FP32, FP16, and INT8, and is compatible with mainstream inference frameworks;
Multispectral ISP	Supports 2-channel independent ISP for visible light and infrared;
CPU	8 Cores;
Real-time Control Core	Independent MCU @ 600M real-time control core;
Power Control Core	Independent MCU power control core;
Video Encoding & Decoding	4K@120fps video encoding and decoding;
GPU	3D GPU + 2D GPU; Supports OpenGL 4.6, OpenCL 3.0, OpenGL ES 3.2;
Audio	High-performance audio processor (DSP) integrated;
Video Interface	Dual-channel 4K@60fps concurrent video output, supports dual-screen different display; Supports video output interfaces: MIPI, HDMI, DP, eDP; HDMI video input interface;
Peripheral Interface	Supports interfaces including CAN, UART, PWM, SPI, IIC, SDIO;
High-speed Interface	Supports high-speed interfaces: PCIe, USB, Gigabit Ethernet, 10 Gigabit Ethernet, SRIO;
External Storage Interface	Supports SPI NOR FLASH, SPI NAND FLASH; Supports eMMC; Supports LPDDR 4/4X/5;
AI Software	Supports mainstream AI model format: ONNX; Supports general AI inference frameworks: PyTorch, TensorFlow, ONNX Runtime; Provides end-to-end full-flow NPU deployment toolchain; Supports neural networks and model deployment including CNN, DNN, RNN, Transformer; Supports CUDA general-purpose computing programming language and comprehensive general computing libraries;