Home > All news > Industry news > Development and Prospect of Hailo Edge AI Processor
芯达茂F广告位 芯达茂F广告位

Development and Prospect of Hailo Edge AI Processor

Founded in 2017, Hailo Technologies is an Israeli technology company deeply involved in the field of artificial intelligence. It has rapidly grown into a unicorn enterprise with 220 employees in less than seven years. Since its establishment, Hailo has been focusing on the research and development of edge AI processor and visual processing unit (VPU) with high enthusiasm.

How to understand the word "edge"? Avi Baum, Hailo's chief technology officer, explained: "In Hailo's concept, edge refers to things that are not based on the cloud, from cameras to cars." Hailo's philosophy for the development of artificial intelligence and machine learning applications has never been to improve cloud-based services, but to deploy applications originally running in the cloud on edge devices.

Artificial intelligence neural network is an important concept in the field of artificial intelligence, that is, a simple model established to imitate the way of human neural network processing information. Artificial intelligence neural network runs through its operation mode of processing information. Although traditional processors can perform this operation, there is still room for improvement in speed and ability. Therefore, Hailo decided to develop its own artificial intelligence neural network processing structure, and Hailo-8 AI accelerator was born. Hailo-8 adopts TSMC 16 nano process node technology as AI/ML accelerator of conventional CPU. Hailo-8 is an AI peripheral of a computer for the same reason that graphics processor (GPU) is used as a computer graphics peripheral. This means that the use of Hailo-8 still depends on a main processor, such as an ARM processor or X86 processor, to run traditional code in the form of a rule-based application.

 

Figure1: Hailo 8 -AI processor

 

At the same time, Avi points out that using this form of AI/ML accelerator to perform "heavy work" is only part of Hailo's blueprint. Hailo hopes that customers can easily use the complete software tool chain without extra learning about new machine structures. This is where many start-ups fail-they create cool hardware, but they can't use it effectively due to lack of software support. As for how Hailo stands out among similar companies, Avi said that the actual experience of users will prove itself.

Hailo is one of the few companies in the industry that has achieved a complete cycle from conceptualization to productization. Many start-ups have failed due to various problems in the process of concept, design, production and mass production, but Hailo-8 has now achieved mass production and has been applied to different products.

 

Figure2: Hailo 8-AI processor

 

Hailo-8, as an edge AI/ML processor, can achieve AI performance of 26 trillion operations per second (TOPS). Hailo-8 Century can support 2, 4, 6 and 8 Hailo-8 devices, providing computing power of 52 trillion times to 208 trillion times per second (TOPS). Hailo's Century PCIe card consumes 400FPS per watt on the ResNet 50 benchmark model. The 52 TOPS card starts at $249, which means that the Edge AI/ML deployment cost can be reduced by 70%. In addition, Hailo-8 Century adopts a robust design, which can adapt to different industrial temperatures and is compatible with most environments and applications.

In addition, Hailo Company successfully combined Hailo-8 technology with CPU to create Hailo-15 Visual Processing Unit (VPU). Hailo-15 vision processing unit has the same specifications as conventional CPU, and a single chip has both independent neural network load capacity and traditional functions of ARM processor. These devices and cards can cover a wide market range beyond people's imagination, including automobiles, robots, smart security cameras, monitoring systems, smart homes, smart cities, industrial applications and so on.

 

Figure3: Hailo 8-AI processor

 

Development and Prospect of Hailo Edge AI Processor-China.exportsemi.com

Related news recommendations

Login

Registration

Login
{{codeText}}
Login
{{codeText}}
Submit
Close
Subscribe
ITEM
Comparison Clear all