AlphaICs begins global sampling of Gluon deep learning co-processor
By GlobeNewswireElectronics Semiconductors AI Gluon Processor semiconductor
Semiconductor suited for vision AI with enhanced FPS/Watt performance
AlphaICs, a Milpitas CA-based AI fabless semiconductor company that develops edge inference and edge learning technologies, announced the availability of engineering samples of ‘Gluon’ – an 8 TOPS Edge AI inference co-processor to customers in surveillance, industrial, retail, auto, and Industrial IoT verticals which carries best-in-class FPS/Watt performance.
Gluon will be shipped with a complete (Software Development Kit) SDK that enables easy deployment of neural networks. The advanced edge inference chip delivers the capability for customers to add AI capability in the current X86 / ARM-based systems, resulting in significant cost savings.
Gluon provides the top fps/watt performance in the market for classification and detection Neural Networks – 32 Frames Per Second (FPS)/watt for Yolo-V2, a leading object detection model & 22 Frames Per Second (FPS)/Watt for VGG-19, a leading classification model.
Produced using 16nm FinFET process at TSMC
Gluon is currently being sampled to for early customers to enable the development of their vision applications. It is engineered for OEMs and solution providers targeting vision market segments, such as surveillance, industrial, retail, Industrial IoT, and edge gateway manufacturers.
Gluon, an 8 TOPS Edge AI co-processor, is produced using the 16nm FinFET process at TSMC. This chip accelerates Deep Learning Neural Network models for classification, detection, and segmentation and is focused on vision applications. Gluon incorporates PCIe and LPDDR4 interfaces to enable high-speed transfers to host processors and DRAM, respectively.
Based on many innovations, Gluon provides best-in-class benchmarks:
*153 frames per second with Yolo-V2, a leading object detection model (416x416x3 image size) in 4.73 watts.
*79 frames per second with VGG-19, a leading classification model (224x224x3) in 3.6 watts.
Print this page