Following the launch of its SensPro sensor hub DSP architecture last year, CEVA this week announced its second generation SensPro2 family, which now features seven vector DSP cores for boosting performance and a range of application-specific instruction set architectures (ISAs) for greater efficiency.
The SensPro2 is hub for artificial intelligence (AI) and DSP processing workloads associated with a wide range of sensors including camera, radar, lidar, time-of-flight, microphones and inertial measurement units (IMUs). Compared to the first generation, SensPro2 delivers 6X more DSP processing for computer vision, 8X more DSP performance for radar processing, a 2X improvement in AI inferencing, and is 20% more power efficient versus its predecessor, at the same process node.
The family now includes seven vector DSP cores, scaling in power and performance. New low power entry-level cores address DSP and AI workloads requiring up to 1 TOPS AI performance and the high-end cores reach 3.2 TOPS. Each of the SensPro2 family members can be configured with application-specific instruction set architectures (ISAs) for radar, audio, computer vision and SLAM, along with parallel vector compute options for floating point and integer data types, to enable an optimum efficiency sensor hub DSP for a specific use-case.
The SensPro2 architecture adopts a range of enhancements that have increased the performance and boosted efficiency for multitasking sensing and AI use cases, such as a new low power vector DSP architecture. For automotive powertrain applications, the upgraded floating-point DSPs offer high-precision performance, addressing the electrification trend with a powerful processor. The architecture and cores are automotive ready with ASIL B hardware random faults and ASIL D systematic fault certification. In terms of performance, SensPro2 is capable of delivering up to 3.2 TOPS for 8×8 networks inferencing running at 1.6GHz, and doubles the memory bandwidth from the first generation, to more efficiently address data-intensive fully-connected layers.
In a briefing with embedded.com, Moshe Sheier, VP of marketing at CEVA, said, “This is a self-contained DSP with the capability to do AI workloads. The amount of configurability and the scalability is unique.” He said the first generation SensPro is already being licensed into silicon in automotive applications which may get to market later this year. In terms of opportunities for SensPro2, he said these will be targeting radar in automotive and internet of things (IoT), as well as audio, where there is considerable requirement for AI capability. However, he said, “Computer vision remains a prime target for this new family of products.”
The second generation SensPro2 DSP family consists of:
- The SP100 and SP50 DSPs, with 128 and 64 INT8 MACS, respectively. These DSPs offer the smallest die size in the family and deliver a 10X performance improvement for DeepSpeech2 speech recognition neural network, compared to the CEVA-BX2 scalar DSP, and are ideal for audio AI workloads, such as conversational assistants, sound analytics, and natural language processing (NLP).
- The SP1000, SP500 and SP250 DSPs with 1024, 512, and 256 INT8 MACs, respectively. These DSPs offer the highest performance and precision in the SensPro2 family, with optimal configurability for computer vision, SLAM, radar, and AI workloads.
- The SPF4 and SPF2 floating point DSPs, with 64 and 32 single precision floating point MACs, respectively. These DSPs are optimized for electric vehicle powertrain control and battery management systems, complemented by a full suite of Eigen Linear Algebra, MATLAB vector libraries and support for Glow graph compiler.
SensPro2 is supported by a broad portfolio of software infrastructure to expedite system designs including an LLVM C/C++ compiler, Eclipse based integrated development environment (IDE), OpenVX API, software libraries for OpenCL, CEVA deep neural network (CDNN) graph compiler including the CDNN-Invite API for inclusion of custom AI engines, CEVA-CV imaging functions, CEVA-SLAM software development kit and vision libraries, radar SDK, ClearVox noise reduction, WhisPro speech recognition, MotionEngine sensor fusion, Tensor Flow Lite Micro support, and the SenslinQ software framework.
- CEVA introduces new configurable sensor hub DSP architecture
- Decisions, decisions: Hardware accelerator or DSP?
- 4D imaging radar chipsets enhance object identification
- Imaging radar development platform offers 2K resolution
- Low-power radar chip uses spiking neural networks
- Smart cities: the case for lidar in intelligent transport systems