Alongside NeuPro-S, CEVA also introduced the CDNN-Invite API, an industry-first deep neural network building technology that supports heterogeneous coprocessing of NeuPro-S cores with custom neural network engines in a unified firmware runtime neural network optimization. NeuPro-S, together with the CDNN-Invite API, is ideal for any vision-based device requiring edge AI processing, including autonomous vehicles, smartphones, surveillance cameras, consumer cameras, and emerging use cases in AR/VR headsets, robots, and industrial applications.
Designed to optimally process neural networks for object segmentation, detection, and classification within videos and images on edge devices, NeuPro-S includes system-aware enhancements that deliver significant performance improvements. These include support for multi-level memory systems to reduce costly external SDRAM transfers, multiple weight compression options, and heterogeneous scalability that allows various combinations of CEVA-XM6 vision DSPs, NeuPro-S cores, and custom AI engines on a single unified architecture. This enables NeuPro-S to achieve, on average, 50% higher throughput, 40% lower memory bandwidth, and 30% lower power consumption than CEVA's first-generation AI processor.
The NeuPro-S family includes the NPS1000, NPS2000, and NPS4000, pre-configured processors with 1000, 2000, and 4000 8-bit MACs per cycle, respectively. The NPS4000 offers the highest CNN performance per core with up to 12.5 Tera operations per second (TOPS) at 1.5 GHz and is fully scalable to reach up to 100 TOPS.
By addressing the growing diversity of application-specific neural networks and processors, the CDNN-Invite API enables the seamless integration of customer-designed neural network engines into CEVA's award-winning Deep Neural Network (CDNN) framework. CDNN then comprehensively optimizes and enhances the networks and layers to leverage the performance excellence of each CEVA-XM6 vision DSP, NeuPro-S, and custom neural network processor. The CDNN-Invite API is already being adopted by customers working closely with CEVA engineers to implement it in commercial products.
CEVA's vast experience and holistic strategy for consolidating processing, acceleration, software, and tools deliver an open, efficient, and programmable platform for deep learning. The fully programmable CEVA-XM6 vision DSP embedded in the NeuPro-S architecture facilitates concurrent processing of image, computer vision, and general DSP workloads, in addition to AI runtime processing. This also allows customers and algorithm developers to take advantage of CEVA's extensive vision and imaging software and libraries, including the CEVA-SLAM software development kit for 3D mapping, the CEVA-CV and CEVA-VX software libraries for computer vision development, and the recently acquired wide-angle imaging software suite that includes dewarp, video stitching, and image data sensor fusion technology.
Analog Devices' A²B 2.0: Automotive audio with Ethernet, increased bandwidth, and low latency
Analog Devices, Inc. has announced that its A²B 2.0 (Automotive Audio Bus) is now production-ready. As the next evolution of ADI's A²B technology, A²B 2.0 is designed to help original equipment manufacturers (OEMs) and Tier 1 suppliers in the automotive industry...
