Hailo takes next steps with new AI chip modules for better biometrics in edge devices
Israel-based Hailo, a developer of AI chips, is now offering its Hailo-8 processor in new form factors that will make integrating the chip into edge devices easier for its customers. The end result will be a new wave of products for smart cities, smart retail, Industry 4.0, and smart homes, many of which will leverage Hailo’s accelerated AI for biometrics functions.
The Hailo-8 deep learning processor, which boasts the ability to process up to 26 tera operations per second (TOPS) is now integrated into two product packages for integration into customer products. The company’s M.2 and PCI Express Mini Card modules include the Hailo-8 in a form factor defined by standards bodies; the standards also enable the easier integration of components ranging from communications functions (Wi-Fi, Bluetooth, etc.) to solid-state drives (SSDs). These modules are designed with a specific card layout and electrical connectors that enable them to be placed on many standard motherboards.
Hailo noted that Taiwan industrial giant Foxconn has already integrated the modules into the Foxconn BOXiedge, which is a high-density and fan-less edge device. In one configuration, the system can ingest 20 camera streams through a video management system for processing and analysis on the edge. Capabilities also include pose estimation and image classification.
In short, Hailo is moving from offering a processor to an integrated system that can be plugged in to a customer’s products to add in AI acceleration, much in the way that a consumer might add a graphics card to their PC for a better gaming experience.
As for performance of the Hailo-8 itself, the company claims that a comparison of performance on multiple standard neural network benchmarks with competitors shows that Hailo’s AI modules achieve an average Frames Per Second (FPS) rate that is 26 times higher than Intel’s Myriad-X modules and 13 times higher than Google’s Edge TPU modules. Performance is important, but so is power consumption on devices such as cameras. Hailo notes that its chip can achieve 3TOPS (trillions of operations per second) per watt.
Expanding use of facial recognition will be a key market driver for Hailo, but with companies like Amazon pushing palm print biometrics for contactless payments, computer vision capabilities enabled by AI acceleration may find further traction. AI chips are being used for mapping and recognition of veins and other physiological data as part of palm recognition due to the speed of processing on the device, or in some cases, in a gateway device near the scanner. The additional processing power could be used to add in additional behavioral data like an individual’s gait in order to offer additional biometric precision for identification.