Chip startup Hailo Technologies Ltd. today debuted a new series of artificial intelligence processors, the Hailo-15 family, that’s optimized to run computer vision models. 

Tel Aviv-based Hailo is backed by more than $223 million in funding. It develops AI processors for vehicles, industrial equipment and other systems. The startup says that its silicon is used by hundreds of customers worldwide.

Hailo’s new Hailo-15 chip series comprises three processors: the Hailo-15L, Hailo-15M and top-end Hailo-15H. All three are designed to power smart cameras that can run AI models locally to analyze the footage they record. Running AI models locally removes the need to send data to the cloud for processing, which lowers bandwidth costs and removes the need to maintain a steady internet connection. 

Manufacturers can use AI-powered smart cameras to monitor the factory floor for production errors. A public transit provider, in turn, could leverage the technology to spot lost luggage. Hailo says its chips also lend themselves to a range of other applications including smart city automation use cases.

“Hailo-15 represents a significant step forward in making AI at the edge more scalable and affordable,” said Chief Executive Officer Orr Danon. “With this launch, we are leveraging our leadership in edge solutions, which are already deployed by hundreds of customers worldwide; the maturity of our AI technology; and our comprehensive software suite, to enable high performance AI in a camera form-factor.”

The three chips in the Hailo-15 series can manage between 7 trillion and 20 trillion computing operations per second. According to Hailo, that amounts to more than five times the performance offered by rival products. Moreover, the startup promises to provide the increased performance at a comparable price to competing offerings.

The processors in the Hailo-15 series use a system-on-chip design, meaning they comprise several different types of compute modules integrated into a single block of silicon.

Each Hailo-15 chip includes circuits optimized specifically to run AI models. There’s also an onboard central processing unit, as well as cybersecurity accelerators that help fend off hacking attempts. The accelerators provide firewall and encryption features, as well as other security capabilities.

The Hailo-15 series implements what Hailo describes as a structure-defined dataflow architecture. According to the startup, the technology allows its chips to run AI models more efficiently than would otherwise be possible.

Hailo’s structure-defined dataflow architecture splits a neural network into multiple smaller software elements. It then assigns each of those software elements to a different part of the chip that hosts the neural network. According to the company, the ability to assign each component of an AI model to the circuits that can run it most efficiently makes it possible to speed up processing. 

The Hailo-15 chips support multiple types of AI models, the company says. They can run object detection algorithms, as well as neural networks that automatically enhance low-quality camera footage. Hailo claims all three chips are capable of processing multiple video streams with 4K resolution, or a resolution of 3,840 pixels by 2,160 pixels.

The more electricity a chip uses to process data, the more heat it generates. Some processors rely on an active cooling system such as a fan to dissipate the heat they produce. 

According to the company, its new Hailo-15 chip series features a power-efficient design that doesn’t require a fan. As a result, manufacturers can more easily incorporate the chips into their smart cameras. Removing the need for a fan, which can take up a relatively significant amount of space in the device chassis, makes it easier to design compact cameras and also lowers production costs.

Image: Hailo