A multi-core SoC that supports High-quality 3d depth, SLAM accelerators, strong Computer Vision engine and Deep Learning (CNN) processor
A multi-core SoC that supports High-quality 3d depth, SLAM accelerators, strong Computer Vision engine and Deep Learning (CNN) processor
NU4000 is a superior multi core vision processor that supports 3D Imaging, Deep Learning and Computer Vision processing for Augmented Reality and Virtual Reality, Drones, Robots, and many other applications. This new generation processor enables high quality depth sensing, SLAM on-chip, Computer Vision and Deep Learning (CNN) capabilities; all in affordable form factor and minimized power consumption, leading the way for smarter user experiences.
NU4000 brings to the market unmatched imaging, vision and AI computing power, exceeding a total of 8 Terra OPS (Operations per second). It introduces an optimized Embedded Vision architecture that effectively combines a set of computing blocks, making it the most powerful vision processor available to date.
Key Features and parameters
- 3 Vector Cores that provides 500 Giga OPS
- A dedicated CNN processor that provides beyond 2 Terra OPS which enables large deep neural networks, such as VGG16 reaching the rate 40 frames (ROIs) per seconds at 10 times less power of equivalent GPU, DSP or FPGA implementations
- 3 Powerful CPU Cores that provides more than 13,000 CoreMark
- Depth processing engine that enables throughput of 120Mp/s and supports multiple simultaneous streams of both stereo and structured light
- SLAM engine that enables highly accurate key points extraction at 120fps from two cameras simultaneously
- Advanced Time-Warp HW engine that reduces the Motion-to-Photon latency toward 1msec for very extensive VR and MR use cases such as 2 displays of 2Kx2K @ 90fps
- Real time processing capable of synchronizing, time-stamping and process inputs from multiple sensors to serve as a smart sensor hub
- More than 3MB of SLAM on-chip for servicing the vision cores
- High throughput LPDDR4 interface for reducing the external memory access bottlenecks
- Connectivity to 6 cameras and 2 displays
- Advanced low power 12nm silicon process