AMS, a leading global provider of optical solutions, and Quadric, an IP innovator in end-to-end AI machine learning processors, have announced a strategic partnership to jointly develop integrated sensing modules that combine AMS 'cutting-edge Mira family of visible and infrared CMOS sensors with Quadric's new Chimera™ GPNPU processor. These modules achieve ultra-low power consumption, and the combination will provide innovative smart sensing solutions for the wearables, robotics, industrial and security markets.
The Quadric Chimera™ Universal Neural Network Processor (GPNPU) enables machine learning graphics processing and parallel C++ algorithms for classical data in a unified architecture, bringing advanced end-to-end AI capabilities to edge-based applications. Quadric's architecture provides efficient machine learning (ML) reasoning performance, but unlike other neural network accelerators that support a limited number of machine learning graphics operators, Quadric's solution also has universal control and signal processing capabilities that combine the desired attributes of an NPU accelerator with a digital signal processor (DSP). The Quadric GPNPU is capable of running neural network graphics and C++ code simultaneously for signal preprocessing and post-processing with significantly lower power consumption than other edge silicon solutions.
The AMS Mira series of CMOS image sensors focus on reducing sensor and system power consumption while minimizing product size to maximize resolution efficiency. The Mira sensor portfolio continues to expand and offer multiple resolutions to meet the needs of diverse applications requiring high performance global shutter image sensors. The series has released several products and is working on more in high and low resolution.
"This collaboration integrates AMS sensors with Quadric processing into a single low-power module, opening up vast new possibilities for intelligent vision sensor deployment," said Joost Seijnaeve, vice president and general manager of CMOS Image sensors at Imis Osram. Original equipment Manufacturers (Oems) no longer need 5W edge silicon wafers or 15W board level computing solutions to perform intelligent sensing functions, including human posture, face recognition, object detection, object recognition, scene analysis, and more. Today, cutting-edge intelligent vision capabilities can be embedded in a single integrated sensor and processing module and applied to a wide range of innovative products in the automotive, consumer, industrial and other sectors."
The Mira image sensor is combined with the Quadric GPNPU to integrate image acquisition and machine learning into one low-power module. Based on Quadric simulations, the product would need only a few hundred milliwatts to perform tasks that currently require a 5W or 10WGPU card. The resolution of the Mira220 and future derivatives, combined with the scalable Chimera™ processor lineup with Quadric computing ranges from 1TOP to 16TOP, will enable multiple combinations for different applications.
"Quadric is excited to join hands with AMS," said Veer Kheterpal, Quadric's chief executive. "This product enables device manufacturers to develop fully programmable smart sensing devices with ultra-low power consumption, opening up a new landscape of machine learning in edge devices."
