Deep Learning Module

   SOM1684 is a high-performance, low-power edge computing module designed by EMA-TECH based on BM1684. It has built-in SOPHON's third-generation processor. The INT8 model can be compiled and calibrated automatically, and fully supports deep learning frameworks such as cafe, Mxnet, Tensorflow, PyTorch and PaddlePaddle. SOM1684 can decode up to 32 channels 1080P@25fps. It can process and analyze more than 16 channels of HD video at the same time.
   Equipped with rich I/O interface resources including PCIE EP / PCIE RC / Ethernet / RS-485 / RS-232 / GPIO / SDIO / PWM, it supports dual GMAC, expandable 10/100/1000Mbps network interface and RGMII mode. The size is only 80mm x 65mm, which is more convenient for engineers to design and commercialize products based on modules.


Specification and Characteristic

Product Model
Application Scenarios

UAV, intelligent NVR, intelligent gateway, HD intelligent camera

Deep Learning Processor
Deep Learning Performance
Supports FP32/INT8, capable of simultaneously processing 16 channels of high-definition video
Video Decoding Capability
1080P@960fps, support H264 & H265
Video Encoding Capability
Picture Codec Capability
1080P 480 image/sec
Network Interface
support dual GMAC, expandable 10/100/1000Mbps network interface, and support RGMII mode.
80mm x 65mm
Other Parameters

2 * PCIE, 2 * GMAC, 1 * I2C, 3 * UART, 1 * SDIO, 1 * PWM

Deep Learning Framework
Support caffe/tensorflow/mxnet/pytorch/PaddlePaddle and other deep learning frameworks

Partner Introduction

Guangzhou EMA Technology Co., Ltd. Official Website

   Guangzhou EMA Technology Co., Ltd. (hereinafter referred to as" EMA "or" the company ") was established in 2006. It is a national high-tech enterprise, a member of the edge computing industry alliance, a specialized, special and new enterprise in Guangzhou in 2021, and a gazelle enterprise in Huangpu District for three consecutive years. It is a leading intelligent hardware manufacturing and deep learning scenario solution service enterprise in the industry.

   With more than 10 years of hardware development experience, EMA has formed in-depth cooperation with SOPHGO and other first-line deep learning processor manufacturers, aiming at the computing power needed for deep learning development, focusing on the edge computing field, and providing edge computing products and solutions that meet the needs of various deep learning scenario applications for industries such as intelligent transportation, community and construction site.