Energy-Efficient Vision AI with the New RZBoard V2L
Avnet launches a new cost and power efficient single-board computer with a special CPU specifically for vision AI development.
It is a week of new launches for Avnet. After announcing the ZUBoard 1CG Development Kit earlier this week, the company today unveiled the RZBoard V2L. The RZBoard V2L is a Raspberry Pi form factor single-board computer with a custom CPU for vision-artificial intelligence (AI) accelerated development.
With a price tag of $149, the RZBoard is a cost-efficient platform for vision AI development. The common Raspberry Pi form factor with standard 40-pin header interface is also intended to allow users to utilize their current collection of header peripheral boards and help streamline prototyping.
The RZBoard covers just about all the interfaces needed for vision AI accelerated development with the following long list of peripherals:
- Gigabit Ethernet
- 801.11ac Wi-Fi and Bluetooth 5
- Two USB 2.0 host ports
- USB 2.0 OTG interface
- MIPI Display Serial Interface (MIPI DSI) and Camera Serial Interface (CSI)
- Controller Area Network Flexible Data-Rate interface
- Pi-Hat-compatible 40-pin expansion header
- Click shuttle expansion header
- Five 12-bit analogue-to-digital converter inputs for interfacing with analog sensors
- 5V input power via USB-C connector
- Audio codec and stereo headphone jack for voice command and audio playback
The heart of the RZBoard V2L is a general-purpose microprocessor with a 1.2GHz dual-core Arm Cortex-A55 and a 200MHz ARM Cortex-M33 along with Renesas' AI accelerator "DRP-AI" for vision. This microprocessor facilitates its video workload in part thanks to a built-in 3D graphics engine with Arm Mali-G31 and video codec (H.264) and 16-bit DDR3L/DDR4 interface.
The DRP-AI is an AI accelerator Renesas built on their Dynamically Reconfigurable Processor (DRP) special purpose hardware. It is the key to the power-efficient aspect of the RzBoard in that the DRP-AI's power efficiency does not require external heat dissipation such as heat sinks or cooling fans.
The DRP-AI also has both real-time AI inference and image processing functions with camera capabilities such as color correction and noise reduction meaning an external image signal processor (ISP) is not required.
The DRP-AI tackles neural networks layers with its two types of processors: the Dynamically Reconfigurable Processor (DRP) to handle the pooling and other layers, and the multiply-accumulate processor (AI-MAC) to handle the convolution and fully connected layers.
The software tool for the host PC to convert a trained AI model to an object code for DRP-AI, DRP-AI Translator, appears to only be available for Ubuntu 18.04 at the moment. However, the software package for writing code to run on the Arm cores though appears to be available for both Windows and Linux.
The RzBoard V2L is available for pre-order with shipping currently estimated for Q4 2022. Accessories that will also be available for the RzBoard include a 5V/3A USB Type C power supply, MIPI CSI camera, and a 7-inch display. It will also be on display at the Embedded Vision Summit May 17-19, 2022 in Santa Clara, California!