AI acceleration board for Raspberry Pi 5 launched in June

DX-M1M delivers 25 trillion operations per second at 3W

Raspberry Pi official newsroom spotlights AI acceleration technology

Compact VLMs and SLMs run entirely on-device

Company eyes early foothold in global developer ecosystem

The Sixfab AI HAT+, an AI acceleration board for Raspberry Pi 5 powered by Deepx's DX-M1M NPU [Source: Raspberry Pi Newsroom]
The Sixfab AI HAT+, an AI acceleration board for Raspberry Pi 5 powered by Deepx's DX-M1M NPU [Source: Raspberry Pi Newsroom]

Deepx (DEEPX) is targeting the global developer ecosystem built around the Raspberry Pi single-board computer platform, leading with a low-power NPU that draws around 3 watts. The company aims to embed the chip in physical AI products including robots, smart cameras and industrial equipment.

According to Deepx, Raspberry Pi recently featured the AI acceleration technology developed by Deepx and its official design partner Sixfab in its official newsroom.

The coverage focused on the Sixfab AI HAT+, an AI acceleration board for the Raspberry Pi 5 that carries Deepx's DX-M1M NPU, explaining how to run AI workloads on-device and outlining potential use cases.

Deepx and Sixfab jointly launched the product in June. The DX-M1M delivers up to 25 trillion operations per second (25 TOPS) while the NPU itself consumes only about 3 watts. A 13 TOPS version is planned for release later this year.

A key feature is the ability to run compact vision-language models (VLMs) and small language models (SLMs) entirely on-device at low power. The compact VLM can interpret the context of a scene, while the SLM can understand commands or summarize information in short sentences.

One example is a camera mounted above a workbench that detects a print failure on a 3D printer, analyzes the scene and explains the problem to the user. The technology can also be applied to production-line quality monitoring, environmental awareness for robots and situational analysis in smart spaces.

To help developers build such applications, Deepx provides model conversion and optimization tools, inference software, image-processing utilities and pre-optimized AI models.

That support is part of Deepx's open-ecosystem strategy, which connects developers to familiar hardware, AI models and software tools. The company says it is the only domestic NPU firm to have established partnerships across global hardware and AI software ecosystems — including Raspberry Pi, Baidu PaddlePaddle and Ultralytics.

Deepx is particularly focused on the fact that developers worldwide use Raspberry Pi to experiment with AI technology and then scale those experiments into real products. Because the tools a developer first learns can shape the industrial products they later build, the company believes expanding its presence in the global developer ecosystem will play a key role in securing a long-term customer base.

"For physical AI to spread, developers around the world need to be able to implement the AI capabilities they require and connect them to actual products," a Deepx official said. "We will support the full development environment — from semiconductors and software through to mass production — so that small experiments that start on a Raspberry Pi can become products sold in markets around the world."


jeongwan@heraldcorp.com