There are several RISC-V development boards on the market (or are about to be released) that use the open instruction set architecture (ISA) — including SiFive’s HiFive and HiFive Unleashed, and GreenWaves’ GAP8 development board. We can add another RISC-V-based development board to that list, with the introduction of Kendryte’s KD233 SoC, which was designed for machine vision and machine learning.
The KD233 is built around Kendryte’s K210 dual-core RISC-V processor and outfitted with a Convolutional Neural Network (CNN) KPU hardware accelerator for AI applications. The K210 chip also packs an APU audio accelerator, 6Mb of general-purpose SRAM, 2Mb of AI SRAM, and AXI ROM for loading code via SPI Flash.
Kendryte’s KD233 hosts 128Mb of SPI NOR Flash and a micro SD card slot for storage, a TFT LCD display, DVT camera, microphone header and a 3.5mm jack for audio. There are several GPIO headers (with one dedicated to Wi-Fi), Field Programmable I/O array, as well as USB to UART with CH340C and Type-C Connector for debugging. It also packs an RGB LED, RTC battery slot, power switch and reset button.
Kendryte’s KD233 can handle facial recognition at 60fps at QVGA resolution and 30fps at VGA resolution using the TensorFlow or Darknet DNN machine-learning frameworks. What’s more, it does so while consuming less than 1-watt of power and drops down to 300mW when idle. Kendryte’s KD233 RISC-V SoC is available now for $49.99, and the company offers a ton of documentation on their website to go with it, including schematics, datasheets, standalone SDKs, and more.