画像
Angus Chan
Angus Chan
Senior Manager, MPU Product Department
掲載: 2020年11月6日

RZ/V2M, the first product in the RZ/V series of microprocessors (MPUs), which features DRP-AI (Dynamically Reconfigurable Processor), Renesas’ exclusive vision-optimized artificial intelligence (AI) accelerator, was selected as a winner of 2020 Aspencore’s World Electronic Achievement Awards in the Processor/DSP/FPGA category. 

The WEAA honors companies, individuals and excellent products that make outstanding contributions to innovation and development in the electronics industry worldwide. The companies, individuals and products nominated for various awards are industry leaders, which fully reflect their leading position and extraordinary in the industry. The winners are jointly selected by a judging committee composed of ASPENCORE global senior industry analysts and website users from Asia, the United States and Europe.

In the Industrial 4.0/IIoT application scenario, those manufacturing systems are designed for advanced production control and predictive maintenance. Therefore, the realization of how to make the terminal equipment become more intelligent is particularly important. Nevertheless, the huge size of an AI model and complex computational logic calculation is a very big challenge to the performance requirement of MCU, which not only takes up a lot of memory but also generates very large-scale complex mathematical operations. And huge calculations greatly inhibit the real-time operation of on-site equipment. 

In order to solve these problems, Renesas Electronics introduced RZ/V2M equipped with DRP-AI (Dynamic Reconfigurable Processor) which utilizes its hardware resources to accelerate the data calculation of an AI model embedded in the terminal equipment by peeling off the entire AI computing process from the CPU. By utilizing the powerful computing ability of hardware arithmetic units to complete the entire AI inference process without taking up CPU resources. Ultimately, efficient intelligence is achieved on terminal equipment and devices.

Solution Features

Complete AI inferences independently without CPU involvement.

画像
Picture1a

An overwhelming power factor is achieved. Compared with competitors’ products, DRP-AI has at least three times more energy efficiency. It also combines high accuracy (support for FP16) with design flexibility (compatibility with common AI models on the market).

画像
Picture2a

Deployment Flow

画像
Picture3

 

Target Application

At present, Renesas already has D-in customer to use existing RZ/A2M with DRP technology to support entry access control system development. Operation flow is using RZ/A2M in the terminal (end-point) to achieve face detection and face cropping, and then sends images to the cloud for face recognition, in order to reduce the transmission load and power consumption between the terminal (end-point) and cloud server. 

The next plan of the customer’s design is to upgrade to V2M with DRP-AI, by keeping the AI inference workload at the endpoint to further reduce the amount of data bandwidth the device occupies and speed up the device's response. According to the actual test results, the acceleration capability in the AI inference process by using DRP-AI showed dozens of times faster than the general MCUs. 

画像
Picture4a

この記事をシェアする