Renesas RZ/V series Highly Power-Efficient Embedded AI MPUs for Vision AI
Vision-Optimized Built-in AI Accelerator; Real-Time AI Inference and Low-Power Consumption for Cameras Used in Industrial, Infrastructure, and Retail Applications
The Renesas RZ/V series of microprocessors (MPUs) for vision AI incorporate Renesas’ exclusive DRP-AI (Dynamically Reconfigurable Processor for AI) dedicated AI accelerator delivering excellent AI inference performance and low power consumption, targeting the vision AI market space. The RZ/V series consists of two products:
- The RZ/V2L with simple ISP, 3D graphics engine and highly versatile peripheral functions which is suitable for a wide range of applications.
- The RZ/V2M features a high performance image signal processor (ISP) with 4K/30fps support.
Key Messages for RZ/V series MPUs:
◊ Best in class heat dissipation for embedded AI MPU
- RZ/V series for Vison AI is equipped with Renesas' original AI-dedicated accelerator – DRP-AI
- RZ/V series achieves both high-speed AI inference and low-power consumption to meet the requirements of real-time AI processing
- The low power consumption of the highly efficient DRP-AI facilitates heat dissipation and eliminates the need for heat sinks and cooling fans, making it possible to shrink solution size and reduce BOM costs
- Stabilize AI inference performance without thermal throttling* and other heat-oriented problems with long-term operation * This is a function when chip becomes high temperature, it reduces the clock frequency of the CPU or AI accelerator
◊ Easy to design embedded AI MPU
- Renesas provides a free tool (DRP-AI Translator) for the RZ/V series that automatically converts AI models to executable format
- DRP-AI Translator supports the industry-standard ONNX (Open Neural Network Exchange) as its input format, enabling customers to convert their familiar environment and proven trained AI models to RZ/V series executable format. Customers can use AI development environments that can export to ONNX format or AI development environments with established conversion flows
- Using RZ/V2L with the AI development environment easily realizes highly functional embedded AI
◊ DRP library provides image processing options
- RZ/V2L makes it possible to perform a wide range of image processing using DRP-AI that supports not only AI inference processing but also flexible image processing, which is the original feature of DRP
- DRP supports simple ISP functions and allows building vision AI systems even with inexpensive camera modules without an ISP
- DRP can realize not only ISP function but also image processing suitable for the AI process users need, which can improve AI performance
◊ Compatibility between RZ/V2L and RZ/G2L
RZ/V2L is upwardly compatible with general-purpose RZ/G2L MPUs, making it easy to build a product lineup with or without AI functions. Like RZ/G2L, RZ/V2L is equipped with video input/output and functions commonly used for industrial applications.
- Equipped with ECC (error correction code) for internal memory and DDR I/F even as entry-line products
- Provides a verified Linux package (VLP) based on CIP Linux to reduce maintenance man-hours for customers. CIP Linux (Civil Infrastructure Platform Linux) is an industrial grade that has been maintained for over 10 years
- Available options to support security functions
- RZ/V2L supports faster CAN-FD (Flexible Data rate) as Controller Area Network (CAN)
- RZ/V2L is equipped with two channels of Giga-bit Ethernet and can be used for gateway applications
◊ More powerful AI performance with RZ/V2M
- Application-specific standard product is focused on Vision AI. It supports up to 4K resolution camera input and hardware ISP functions that can make good image quality for AI inference. RZ/V2M also includes AI inference using DRP-AI and CIP Linux features
- This ISP is already tuned to Renesas selected sensors by experts to allow customers to use the best image quality without additional tuning required. Customers can use the API to adjust image quality based on Renesas tuning when they want to revise some parameters
- AI inference performance is about 1.5 times that of RZ/V2L, so it has scalability in terms of AI performance
- Very low power consumption is about 30% better than competitors despite same AI performance. This feature is important for embedded systems and contributes to the end product in many ways
- No heat sink or cooling fan is needed, enabling reduced product size & BOM costs.
- RZ/V2L enables building of a vision AI system using an inexpensive camera module without an ISP.
- RZ/V2M provides “Tuned ISP” for growing list of sensors to use highest quality images without needing labor intensive ISP parameter tuning.
RZ/V2L Block Diagram
RZ/V2M Block Diagram
RZ/V Evaluation Board Kits
RZ/V2L Evaluation Board Kit
The RZ/V2L Evaluation Board Kit consists of a module board (SOM) and a carrier board that can easily start evaluation of the RZ/V2L. It also includes the MIPI Camera Module (MIPI CSI). The carrier board can be commonly used with RZ/G2L, RZ/G2LC, RZ/G2UL, and RZ/V2L modules (SOM) prepared according to the SMARC v2.1 standard. This allows you to experience seamless and flexible evaluation between these devices.
Evaluation Board Kit includes:
RZ/V2M Evaluation Board Kit
The RZ/V2M BSP, ISP, and DRP-AI can be evaluated on the RZV2M Evaluation Kit