MCU for entry-level vision AI

MCU for entry-level vision AI

As part of the RZ/V Series, the new RZ/V2L has the company’s ‘DRP-AI’ artificial intelligence accelerator, with operating frequency and memory interface scaled for entry-level applications.

The DRP-AI provides real-time inference and image processing with camera support such as colour correction and noise reduction.

In this case it is described as “1Top/s/W class”) and can run Tiny YOLOv2 at 28frame’s.

“This enables customers to develop AI-based vision applications, such as point-of-sale terminals and robot vacuum cleaners, without requiring an external image processor,” according to Renesas.

It comes in 15 mm square or 21 mm square BGA packages that are pin-compatible with the existing general-purpose MPUs, allowing RZ/G2L users to upgrade an AI-capable processor.

As part of the the RZ/V2L development environment, Renesas offers DRP-AI Translator, a tool that automatically converts AI models into an executable format. The input format is industry-standard ONNX (open neural network exchange).

RZ/V2L MPUs:

  • Dual or single 1.2GHz 64bit Arm Cortex-A55
  • Cortex-M33
  • 16bit, single-channel DDR memory interface
  • Arm Mali-G31 3D graphics
  • H.264 video codec
  • MIPI-CSI and parallel CMOS camera interfaces
  • MIPI-DSI and parallel Display interfaces
  • Memory with ECC (error checking and correction)
  • ISP functions for machine vision (up to HD) in DRP library
  • Verified Linux package (VLP) based on Civil Infrastructure Platform Linux

RZ/V2L evaluation boards are available, and provide: “circuit diagram and board layout, and include a Power supply circuit and timing tree”.

A power management IC optimised for RZ/V2L in the pipeline, and examples combining them are expected in the second half of this year.

The RZ/V2L product page is here