SensPro has been developed to fulfil the need for specialized processors that can efficiently handle the proliferation of different types of sensors. The sensors provide different data types and bit-rates from imaging, sound, RF and motion, which together are used to create a 3D contextually-aware device.
The SensPro architecture combines the high performance single and half precision floating-point math needed for high dynamic range signal processing, point cloud creation and deep neural network (DNN) training, besides the large amount of 8- and 16-bit parallel processing capacity required for voice, imaging, DNN inference processing and Simultaneous Localization and Mapping (SLAM).
SensPro has a highly-configurable 8-way VLIW architecture to allow it to be tuned for different applications. It uses a micro-architecture that combines scalar and vector processing units and incorporates an advanced, deep pipeline enabling operating speeds of 1.6GHz in a 7nm process. SensPro uses a CEVA-BX2 scalar processor for control code execution with a 4.3 CoreMark/MHz score. It features a wide SIMD scalable processor architecture for parallel processing and is configurable for up to 1024 8×8 MACs, 256 16×16 MACs, dedicated 8×2 Binary Neural Networks support, as well as 64 single precision and 128 half precision floating point MACs. This allows it to deliver 3 TOPS for 8×8 networks inferencing, 20 TOPS for Binary Neural Networks inferencing, and 400 GFLOPS for floating point arithmetic. SensPro also has a memory architecture providing a bandwidth of 400GB/s, 4-way instruction cache, 2-way vector data cache, DMA, and queue and buffer managers for offloading the DSP from data transactions.
SensPro is accompanied by software and development tools to quicked system designs including an LLVM C/C++ compiler, Eclipse-based 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, ClearVox noise reduction, WhisPro speech recognition, MotionEngine sensor fusion, and the SenslinQ software framework.
SensPro DSPs will initially be available in three configurations, each including a CEVA-BX2 scalar processor and various vector units configured for optimal use-case handling:
• SP250 – single vector unit with 256 8×8 MACs targeting imaging, vision, and sound centric applications
• SP500F – single vector unit with 512 8×8 MACs and 64 single precision floating point MACs targeting SLAM centric applications
• SP1000 – dual vector units with 1024 8×8 MACs and binary networks support targeting AI centric applications