Axelera AI Raises $68M for ‘World’s Most Powerful AIPU’ for the Edge
The new funding will expand the company’s product offerings and align it with the generative AI movement.
The AI hardware marketplace is a competitive yet well-funded space. While major industry players are developing new processing solutions and growing their market share, smaller startups can still carve out a niche and establish themselves.
Axelera AI's Metis AI platform is founded on the Metis AI processing unit.
Recently, Axelera AI, a European AI startup specializing in advanced edge AI solutions, secured $68 million in Series B funding. This funding round follows Axelera's $50 million oversubscribed Series A round in March 2023. In this piece, we’ll look closer at Axelera AI’s technology and its plans for the newly secured funding.
Axelera AI Raises Funds for the Metis AI Platform
The cornerstone of Axelera AI's offerings is the Metis AI platform, a comprehensive solution combining hardware and software to optimize AI inference at the edge.
The platform includes the Metis AI processing unit (AIPU) and the Voyager software development kit (SDK). The Metis AIPU, fabricated using 12 nm CMOS technology, delivers 214 tera operations per second (TOPS) with an energy efficiency of 15 TOPS/W at INT8. It uses a novel digital in-memory computing (D-IMC) engine, enabling high-performance AI computations with low power consumption.
Axelera AI’s hardware achieves up to 214 TOPs.
What sets the Metis AI platform apart, according to Axelera AI, is its ability to execute complex neural network models with ISO-accuracy at FP32 without requiring retraining. This dramatically simplifies deployment and integration. The Voyager SDK then provides a curated model zoo and customizable application templates. For the software engineer, these resources reduce development time and time to market for AI solutions.
A Strong Folloup to the Metis Eval Kit
Earlier this year, Axelera AI announced the availability of its Metis evaluation kit. The startup designed these kits to provide developers and engineers with a comprehensive and intuitive platform for developing edge AI models on Axelera AI’s hardware. To this end, the kits are equipped with Axelera AI's AIPU, which features quad AI cores and D-IMC technology. They are paired with 1 GB LPDDR4x memory and connect via PCIe 3.0 (x4) in a full-height, half-length, and single-slot configuration.
The Aetina RK3588 + M.2 evaluation kit.
To simplify the development process, the kits come preloaded with detection and classification neural networks. The included Voyager SDK offers developers a comprehensive AI development environment compatible with major machine learning frameworks such as PyTorch, ONNX, and TensorFlow. The SDK is also compatible with various neural networks, including YOLOv5, YOLOv7, Resnet 50, MobileNetV2, and SSD-MobileNetV2.
The kits come in different SKUs to support x85 and Arm platforms. These options include an Aetina RK3588 + M.2 single board computer, a Dell 3460XE workstation + PCIe, a Lenovo ThinkStation P360 Ultra + PCIe, an Advantech MIC-770v3 or ARK-3534 + PCIe, and a FireFly RK3588 + PCIe. Regardless of SKU, the kits aim to provide a user-friendly evaluation experience that enables developers to streamline application logic and model development.
Axelera AI Moves Toward Generative AI
Axelera AI plans to use its newly secured funds to accelerate research and development, expand its global presence, and augment its product offerings to meet the growing demand for edge AI applications. The company then plans to scale operations and collaborate with industry leaders to co-develop new edge AI solutions. The company intends to build upon its existing products to expand into new markets, with a particular focus on generative AI.
All images used courtesy of Axelera AI.
