Renesas Expands RZ/V Sequence with Constructed-in Imaginative and prescient AI Accelerator

Renesas Electronics Company permits AI processing of picture knowledge from a number of cameras, providing a brand new stage of extremely correct picture recognition for imaginative and prescient AI purposes. Geared up with two 64-bit Arm Cortex-A53 cores, the brand new gadget is able to delivering excessive computing efficiency with a most working frequency of 1GHz. The RZ/V2MA incorporates a proprietary low-power DRP-AI (Dynamically Reconfigurable Processor) accelerator which might course of imaginative and prescient AI at 1 TOPS/W (tera operations per second, per watt) class efficiency.

The RZ/V2MA gadget affords high-speed interfaces corresponding to Ethernet, USB, and PCI Categorical that enable picture enter from a number of exterior cameras. Along with the DRP-AI accelerator, the RZ/V2MA consists of an OpenCV accelerator that enables rule-based picture processing concurrently. These options carry extremely correct picture recognition capabilities for machine imaginative and prescient merchandise corresponding to AI-equipped gateways, video servers, safety gates, POS terminals and robotic arms.

The brand new RZ/V2MA affords a full suite of improvement instruments to assist imaginative and prescient AI system design. Along with the present DRP-AI Translator, the brand new gadget provides DRP-AI TVM (Notice 1), which is predicated on the open-source deep studying compiler Apache TVM expertise (Notice 2). Whereas DRP AI Translator is designed to transform AI fashions into DRP-AI executables, the DRP-AI TVM compiler lets the DRP-AI accelerator work along with the CPU, permitting DRP-AI to transform and generate extra AI fashions than ever earlier than. As a primary section, Renesas helps ONNX and PyTorch AI fashions and plans to assist Tensorflow sooner or later.

“One of many challenges for embedded techniques builders who wish to implement machine studying is to maintain up with the most recent AI fashions which can be always evolving,” stated Shigeki Kato, Vice President of Renesas’ Enterprise Infrastructure Enterprise Division. “With the brand new DRP-AI TVM software, we’re providing designers the choice to develop AI frameworks and AI fashions that may be transformed to executable codecs, permitting them to carry the most recent picture recognition capabilities to embedded gadgets utilizing new AI fashions.”

“Renesas’ RZ/V collection is right for embedded gadgets because it doesn’t want followers or warmth sinks, resulting from its extraordinarily low energy consumption and low warmth functionality when working AI,” stated  Chiharu Nakabayashi, President of amnimo Inc., a supplier of IoT and AI-based companies and a completely owned subsidiary of Yokogawa Electrical Company. “With these gadgets, we’re assured that we will develop highly effective picture AI gateways that may be put in wherever.”

Key Options of the RZ/V2MA

• Two 64-bit Arm Cortex-A53 cores with a most working frequency of 1 GHz
• AI accelerator DRP-AI (1 TOPS/W class), attaining 52 fps (frames per second) when executing TinyYoloV3 applications
• OpenCV accelerator for rule-based picture processing
• Ethernet, USB, and PCI Categorical interfaces for picture enter from exterior cameras
• Video codecs (H.265 and H.264)
• DRP-AI TVM software for changing AI fashions based mostly on TVM expertise. Assist ONNX and PyTorch codecs initially.
• Excessive-speed reminiscence interfaces embody LPDDR4 (3200Mbps), USB 3.1 (as much as 5Gbps), and PCI Categorical (2 lanes)
• Accessible in 15mm sq. BGA bundle