엣지단에서 AI 추론 칩셋의 공급 증가가 전망되고 있다. 머신 비전과 센서 데이터 분석이 엣지 AI 관련 출하량을 견인할 것으로 보이는 가운데 엣지 AI 디바이스용 AI 반도체가 SoC 형태로도 등장했다.

▲AMD Embedded+ combining Ryzen embedded processors and Versal adaptive SoCs (Image: AMD)
Support for simplifying ODM on-device AI development
Embedded+, Complementary Architecture SoC
The supply of AI inference chipsets at the edge is expected to increase. While machine vision and sensor data analysis are expected to drive edge AI-related shipments, AI semiconductors for edge AI devices have also appeared in the form of SoCs.
AMD announced on the 6th (local time) that it is launching AMD Embedded Plus (hereinafter referred to as '+'), which combines Ryzen embedded processors and Versal adaptive SoCs into a single integrated board.
The AMD Embedded+ integrated computing platform can be used by ODMs to develop on-device AI products, and provides full-stack support for application development in areas such as medical, industrial, and automotive through the software platform. AMD emphasized that it is possible to develop low-power, small-sized, and long-life products.
Embedded+ architecture, which combines embedded processors and adaptable SoCs, leverages x86 computing, integrated graphics, and programmable hardware to support AI inference and sensor fusion applications.
Embedded+ is based on AMD Ryzen Embedded R2000 and Versal AI Edge architecture. It is emphasized that the two complementary architectures do not overlap in functions and features.
In a pre-event media briefing, AMD senior director Chetan Khona highlighted the three key pillars of Embedded+ architecture: sensor friendliness, offload processing, and time-to-market.

▲
Chetan Khona, AMD Senior Director, Industry, Vision
, Healthcare
, and Science (Image: AMD)
AMD emphasized that Embedded+ has excellent acceptance of various sensor types and interfaces. It said that image sensors, lidar, radar, ultrasonic, GPS, IMU, and IP camera sensors and products can be connected to adaptive I/O, and that up to 85 I/Os can be used.
It also supports a wide range of universal interfaces, including Ethernet, USB, COM, HDMI/DP, and AUDIO. It is equipped with high-performance Zen cores and Radeon graphics to support 4K multimedia rendering and display options, and also provides a built-in video codec for 4K H.264/H.265 encoding and decoding.
Examples of usable applications include machine vision, industrial networking, robotics, healthcare, smart cities, security, and distribution. Director Chetan Hona emphasized Embedded+'s buffer-free image preprocessing in machine vision, multi-channel camera synchronization, and low power and lightweight, and showed the possibility of using general cameras as AI cameras through the edge AI box in smart city applications.
System developers can expand their portfolio with ODM board products from the embedded+ architecture-based ecosystem, optimizing products with performance and power characteristics best suited to B2B applications.
The first Embedded+ architecture-based ODM solution is said to have been introduced to the market by Sapphire Technology's Sapphire Edge+ VPR-4616-MB.
This low-power Mini-ITX form factor motherboard product implements 30W low-power performance by utilizing the Ryzen Embedded R2314 processor and the Versal AI Edge VE2302 adaptive SoC.
The VPR-4616-MB is also available as a complete system, including memory, storage, power supply, and chassis.
“By leveraging a proven and trusted computing architecture, we have been able to focus our resources on product differentiation, reducing development time and R&D costs,” said Adrian Thompson, SVP of Global Marketing at Sapphire Technologies.