Hailo chips integrated by Axiomtek for smart city edge AI infrastructure
Taiwan-based Axiomtek announced it will be offering hardware with AI processor chips from Hailo, the latest in a series of design wins for the edge AI chip startup. Axiomtek said it is planning to target smart city applications with the new system, enabling more efficient processing of crucial tasks like facial recognition and object detection to take place closer to where data is collected.
The new RSC100 ‘Plato’ is “Well suited for use in smart city applications, including smart surveillance, smart factory, smart agriculture, and smart transportation,” states Ken Pan, the product manager at Axiomtek.
The move is the latest in a series of moves bringing Hailo processors to systems capable of supporting face biometrics at the edge.
According to market research firm IDC, the worldwide market for edge computing hardware will reach $80.7 billion in 2024. One of the key workloads these systems will run are AI/ML applications. AI software for smart city applications is set to increase to $4.9 billion in worldwide sales for 2025, up from $673.8 million in 2019, according to research firm Omdia.
The RSC100 comes with flexible and capable storage including a 16GB eMMC onboard as well as expansion slots and interfaces for SSD and NVMe storage. The system has dual GbE ports which enables IP cameras and LiDAR connectivity with greater bandwidth for video analytics, according to Axiomtek. Other expansion modules enable customers to add connectivity for their environments such as Wi-Fi/Bluetooth, 5G/4G/LTE.
Axiomtek notes that the ruggedized Plato is designed for harsh environments, and operates with a temperature range of -20°C to +70°C.
Hailo recently raised $136 million in Series C funding, bringing the total venture funding raised to $224 million. Its Hailo-8 chip is said to be based on neural networks of the brain, and its ‘hyper-efficient’ processor is better suited to edge workloads than chips more commonly found in the data center. It only uses 2.5 W of power while it can execute as many as 26 tera-operations per second.