CEVA introduces the second generation SensPro DSP family for AI and DSP hub processing workloads

CEVA, the world's leading licensed provider of wireless connectivity and smart sensing technologies, announced the launch of the second-generation SensPro DSP family for AI and DSP hub processing workloads, comprising a wide range of sensors including cameras, radar, LiDAR, time-of-flight, microphones and inertial measurement units (IMUs). The SensPro2 family builds on CEVA's industry-leading sensor hub DSP leadership, offering a six-fold DSP processing performance improvement for computer vision, an eight-fold DSP performance improvement for radar processing, and a two-fold increase in AI inference performance at the same process nodes, with a 20% increase in power efficiency compared to its predecessor.

● Compared to the first-generation SensPro with the same process node, SensPro2™ delivers six times better computer vision performance, two times better AI reasoning, and 20 percent lower power consumption

● The new low-power entry-level SensPro2 DSP delivers a tenfold increase in performance compared to the CEVA-BX2 DSP for AI networks for voice assistants, natural language processing and spatial audio

● The SensPro2 DSP with high precision floating point is available for automotive applications for powertrain battery management and radar systems

The SensPro2 family has been expanded to include seven vector DSP cores that are scalable in terms of power and performance. The new entry-level core can meet requirements up to 1 TOPS AI performance, while the high-end core can achieve 3.2 TOPS performance. Each SensPro2 family member can be configured with an application-specific instruction set architecture (ISA) for radar, audio, computer vision, and SLAM, as well as parallel vector computing units for floating-point and integer data types, resulting in the most efficient sensor hub DSP for specific use cases.

Ran Snir, vice president of Research and Development at CEVA, said: "Our new SensPro2 family of high-power sensor hub DSPS offers scalable performance, multiple accuracies and high utilization for the increasingly complex and diverse AI/ sensor workloads of contextion-aware devices. SensPro2 architecture is a unique and innovative architecture with a universal ISA that enables seamless software reuse across all SensPro2 DSPS. As customers increasingly use SensPro2 cores in their product designs, they value this feature and the value of an application-specific ISA."

The SensPro2 architecture features a number of technology upgrades that improve performance and increase efficiency for multi-task sensing and AI use cases, such as the new low-power vector DSP architecture. For automotive powertrain applications, the upgraded floating point DSP has high precision performance and meets the demands of the electrification trend with a powerful processor. The SensPro2 architecture and kernel have been certified for ASIL Level B hardware Random failure and ASIL Level D system failure for use in cars. In terms of performance, SensPro2 is capable of running at 1.6GHz to deliver up to 3.2 TOPS performance for 8x8 network inference operations, with twice the memory bandwidth of the first generation to handle the data-intensive full connection layer more efficiently.

Members of the second generation SensPro DSP family include:

● SP100 and SP50 DSPS with 128 and 64 INT8 Macs, respectively. These DSPS feature minimal chip size and boost the performance of DeepSpeech2 speech recognition neural networks by a factor of ten compared to CEVA-BX2 scalar DSPS for audio AI workloads such as conversation assistants, voice analysis, and natural language processing (NLP).

● SP1000, SP500, and SP250 DSPS with 1024, 512, and 256 INT8 Macs, respectively, offer the highest performance and accuracy in the SensPro2 family and the best configurability for computer vision, SLAM, radar, and AI workloads.

● The SPF4 and SPF2 floating point DSPS feature 64 and 32 single precision floating point MACS, respectively. These DSPS are optimized for electric vehicle powertrain control and battery management systems and are equipped with a full set of Eigen Linear Algebra, MATLAB vector libraries, and Glow graphics compiler support.

SensPro2 has extensive software infrastructure support to accelerate system design, including the LLVM C/C++ compiler, the Eclipse-based Integrated Development Environment (IDE), the OpenVX API, the OpenCL software library, and the CEVA Deep Neural Network (CDNN) graphics compiler. Includes CDNN-Invite API for adding custom AI engines, CEVA-CV image functionality, CEVA-SLAM software development kit and vision software library, and Radar SDK, ClearVox noise reduction, WhisPro speech recognition, MotionEngine sensor fusion, Tensor Flow Lite Micro support, and SenslinQ software framework.