Gyrfalcon Technology, Inc. (GTI) designed the 4th generation, Lightspeeur 5801 neural accelerator chip specifically for mass consumer electronics market, IoT, Edge AI, and end-point devices offering 2.8 TOPS at 224-mW power consumption (the equivalent of 12.6 TOPS/W). Gyrfalcon uses an artificial processor-in-memory (APiMTM) architecture that is particularly power-efficient compared with others.
CIO Advisor APAC writes: “Gyrfalcon Technology Inc. (GTI) is the world’s leading developer of high-performance, low-cost Artificial Intelligence (AI) accelerators, founded in 2017 by veteran Silicon Valley entrepreneurs and Artificial Intelligence scientists. Chosen as Samsung’s global AI accelerator chip vendor, GTI started shipping first-generation chip in Q1 2018. As part of a strategic partnership agreement, GTI chips will enable LG Electronics’ AI products in 2019. The company drives adoption of AI by bringing the power of cloud Artificial Intelligence to local devices and improves Cloud performance with the utmost in AI customization for new equipment and a path to AI customers.”
This recognition underscores GTI’s commitment to optimizing AI-powered solutions delivering an unmatched ratio of high performance-to-low energy consumption with reduced production costs for the AI-Edge market.
As AI is being integrated into more applications at the edge, on-device camera processing and AI-edge vision computing are becoming more important across a wide range of vertical markets including machine vision computing, automotive, smart IoT, mobile devices, and smartphone.
GTI accelerator chip equipped with the newest AI-powered image/video enhancing techniques and advanced algorithms offer superior performance, higher energy efficiency, and reduced BOM cost by processing the raw data output of camera sensors. Having a dedicated AI processor unit on the device, provides higher reliability and precision, less latency, and improved privacy for running critical applications in real-time, always-on, anywhere operation independent of Internet connection.
Typical applications of AI-cameras are super-resolution (zoom), Bokeh, gesture and pose detection, background/foregrounds segmentation, low-light enhancement (de-noising), image recognition, object detection and tracking, facial recognition and authentication, scene optimization, and visual, search and content analysis for a broad range of vertical markets.
AI-camera models can be trained with a model development kit, and the development of applications is possible with a software development kit (SDK) that is available for Linux X86_64, Microsoft Windows, Android platforms, and also ARM v7l and ARM v8 instruction sets. It supports popular frameworks including TensorFlow, PyTorch & Caffe.