Renesas unveils embedded AI MPU prototype

Renesas Electronics Corp. has unveiled two new technologies that enable higher speeds and lower power consumption in microprocessors (MPUs) for advanced vision AI. These include a new AI accelerator for lightweight AI models and an embedded process technology for real-time processing. Also announced is an embedded AI MPU prototype, leveraging these two technologies.

(Source: Renesas Electronics Corp.)

The dynamically reconfigurable processor (DRP)-based AI accelerator is designed to efficiently process lightweight AI models, while the heterogeneous architecture technology enables real-time processing by cooperatively operating processor IPs, such as the CPU, Renesas said. Using both technologies, the company developed a prototype of an embedded AI MPU and confirmed its high-speed and low-power operation, achieving up to 16× faster processing at 130 TOPS and power efficiency of up to 23.9 TOPS/W at 0.8 V supply. Results were announced at ISSCC 2024.

Renesas said it developed the technologies to meet growing demand for systems—robots in factories, logistics, medical services and stores—that can autonomously run in real time by detecting surroundings using advanced vision AI. Both higher performance and lower power consumption are needed in AI chips, particularly due to demanding requirements on heat generation in embedded devices, the company said.

AI accelerator 

Renesas optimized its unique DRP-based AI accelerator (DRP-AI) for pruning, which is typically used for improving AI processing efficiency. The company analyzed how pruning pattern characteristics and a pruning method are related to recognition accuracy in typical image recognition AI models (CNN models), identifying the hardware structure of an AI accelerator that can meet both high recognition accuracy and an efficient pruning rate, then applied it to the DRP-AI design.

Renesas also developed software to reduce the weight of AI models optimized for this DRP-AI, which converts the random pruning model configuration into highly efficient parallel computing. The result was higher-speed AI processing.

The company’s pruning support technology (flexible N:M pruning technology) allows for fine control of the pruning rate according to the power consumption, operating speed and recognition accuracy required by users. It can dynamically change the number of cycles in response to changes in the local pruning rate in AI models, Renesas said, and reduces the number of AI model processing cycles to as little as one-sixteenth of pruning incompatible models and consumes less than one-eighth of the power.

Heterogeneous architecture technology

Renesas said the development of the DRP, and the CPU and AI accelerator (DRP-AI) led to the development of the new technology for advanced vision AI processing for recognition of the surrounding environment.

Because robot motion judgment and control require detailed condition programming in response to changes in the surrounding environment, CPU-based software processing is more suitable than AI-based processing, Renesas said. However, the challenge is that CPUs with current embedded processors are not fully capable of controlling robots in real time.

The heterogeneous architecture technology solves this challenge by enabling real-time processing for robot control, the company said.

On example cited is the use of simultaneously localization and mapping (SLAM) for robot position recognition in parallel with environment recognition by vision AI processing. Renesas said operating this SLAM through instantaneous program switching with the DRP and parallel operation of the AI accelerator and CPU resulted in about 17× faster operation speeds and about 12× higher operating power efficiency than the embedded CPU alone.

In addition to the high power efficiency of 23.9 TOPS per watt at a normal power voltage of 0.8 V for the AI accelerator and operating power efficiency of 10 TOPS per watt for major AI models, the AI MPU prototype also proved that AI processing is possible without a fan or heat sink.

The results help solve heat generation due to increased power consumption, one of the challenges associated with the implementation of AI chips in a variety of embedded devices such as service robots and automated guided vehicles, Renesas said.

These technologies will be applied to Renesas’ RZ/V series of MPUs for vision AI applications.