Analog Tech Launches 'AIR-AD-AI-001', a Compact AI PC Powered by Axelera AI's Metis® AIPU, Delivering 214 TOPS in a Small Form Factor

Analog Tech (Headquarters: Minato-ku, Tokyo; CEO: Daiji Nagayama) is launching the 'AIR-AD-AI-001', a compact AI PC equipped with the advanced 'Metis® AIPU' AI processor developed by the European AI semiconductor startup Axelera AI. This device joins the company's 'AironiA' edge AI computing series. Analog Tech has adopted Axelera AI's 'Metis® AIPU' processor to provide high-performance edge AI computing to the Japanese market.

Background: The 'Memory Wall Problem' and the GPU Barrier in Edge AI Needs are rapidly increasing for 'on-premise' AI, such as visual inspection for production lines, multi-site surveillance cameras, and digital signage. However, traditional edge AI using GPUs faces major challenges.

The 'matrix-vector multiplication' operation, essential for AI inference, requires moving massive amounts of data back and forth between DRAM and compute units. This phenomenon, known as the 'memory wall problem,' creates bottlenecks in performance and power consumption, hindering the simultaneous achievement of 'compact size, low power, and high performance' required for edge environments.

Solution: D-IMC Architecture Performing Calculations Directly Within Memory Axelera AI's 'Metis® AIPU' fundamentally solves this issue by adopting a unique D-IMC (Digital In-Memory Computing) architecture. By executing key AI inference calculations directly inside the memory cells rather than shuttling data between memory and units, data movement is minimized, achieving both significant power savings and high throughput.

As a result, the 'AIR-AD-AI-001' delivers 214 TOPS of AI inference performance while the AI inference module (Metis® AIPU) consumes only 8–15W, all packed into a compact chassis of approximately 5 liters.

Key Features of the AIR-AD-AI-001 - 214 TOPS in a 5-Liter Chassis: Industry-leading cost efficiency in a compact form factor. Measuring 200mm (W) x 250mm (D) x 95mm (H) and weighing 1.9kg, it supports VESA mounting, enabling installation on walls, behind displays, or in racks. - Real-Time AI Analysis for Up to 24 Video Channels: Using YOLOv5, a single unit can simultaneously process up to 24 video streams in real-time (varies by condition). It extends the scale of visual AI applications, such as multi-camera inspection, wide-area surveillance, and signage, ensuring privacy and low latency by eliminating cloud data transmission. - Ultra-Low Power Consumption for Field-Ready Running Costs: The AIPU accelerator consumes only 8–15W. It offers lower running costs than data center GPU servers and eliminates cloud API usage fees. - Deploy Existing AI Models Directly: With Voyager® SDK support, pre-trained models from PyTorch, TensorFlow, and ONNX can be optimized and deployed to the Metis® AIPU, allowing seamless transition to high-performance edge inference environments.

Key Applications and Use Cases - Manufacturing and Quality Inspection: Real-time visual inspection for production lines, automated visual checks. - Security and Surveillance: Multi-camera human/action detection on-premise (without sending personal data externally). - Digital Signage: Visitor attribute estimation, dynamic content switching, continuous low-power operation. - Smart Cities: Traffic counting, congestion monitoring. - Robotics: Visual AI integration for industrial robots. - PoC and Development Infrastructure: Prototyping environment for AI startups and research institutions.

About Axelera AI Founded in 2021 in the Netherlands, Axelera AI is an AI semiconductor startup incubated by the Bitfury Group and imec. With a mission to 'democratize AI,' it develops high-performance, ultra-low-power AI accelerators for the edge.