Vision processor IP accelerates deep learning algorithms

The CNN engine operates in parallel with the vector DSPs, enabling Inuitive to implement highly accurate object detection, image classification and semantic segmentation capabilities in its SoC at a fraction of the power consumption of competing vision processor IP solutions. The MetaWare EV Development Toolkit, a comprehensive set of tools and software for the EV6x Processors, can distribute the NU4000's computations between the vision CPU and CNN resources to support new and emerging neural network algorithms as well as customer-specific CNN layers.

"To provide our customers with the most advanced, high-performance vision processing capabilities for their demanding computer vision and machine learning applications, we require vision processing IP with the best mix of performance and power consumption with the flexibility to support current and future CNN graphs," said Dor Zepeniuk, Vice President of R&D at Inuitive. "The competitive power, performance and area advantages of the DesignWare EV62 processor with its fast CNN engine were critical to achieving the real time processing capabilities of our NU4000 3D imaging and vision processor SoC."

The DesignWare EV6x Processor Family supports Inuitive's NU4000 SoC requirements for processing multiple camera inputs with high-definition resolutions up to 4K. The EV6x processors have a heterogeneous multicore architecture that includes one to four high-performance vector DSPs designed to be tightly coupled to a CNN engine that delivers up to 4.5 TeraMACs/sec in typical 16nm FinFET process technologies.

To simplify software application development, the DesignWare EV6x Processors are fully programmable and supported by the MetaWare EV Development Toolkit, which includes software development tools based on the OpenVX, OpenCV and Open CL C embedded vision standards. In addition, the Toolkit includes a CNN mapping tool that automatically converts neural networks trained using popular frameworks like Caffe and Tensorflow into an executable for the CNN engine.

"Artificial intelligence is enabling intelligent, real-time identification and recognition at high precision for applications such as ADAS, video surveillance and virtual/augmented reality," said John Koeter, Vice President of marketing for IP at Synopsys. "By using Synopsys' EV6x Vision Processor IP, companies like Inuitive can create SoC designs that execute deep learning algorithms at TeraMAC-per-second speeds with the lowest power consumption and highest accuracy."