HISAOR is a rad-tolerant high performance (12 TOPS, 64 GFLOPS) AI SOC designed for high-reliability embedded AI applications. Its quadcore ARM Cortex-A9 cores power the supervisory real-time operating system which manages the collection of data for AI processing from its high-speed communications interfaces. The AI number crunching is carried out by octacore GPU and neural network acceleration coprocessors based on Verisilicon IP. Vision processing applications take advantage of the media processors for encode and decode of H264/5 and JPEG2000 conversions. A memory controller supporting DDR4, DDR3, and LPDDR3 is available with non-volatile controllers for NAND and SPI NOR flash. A high-speed AXI interconnect matrix provides for high-throughput transfers beween peripherals and memory.
AI models trained and developed with cloud applications (e.g. Amazon SageMaker) can easily be ported to run on the HISAOR AI coprocessors using the Verisilicon Vivante Acuity tools and IDE. Support for OpenCL and OpenVX is provided as the standard interface for vision processing applications. Driver sources are provided for applications requiring bespoke AI optimisation. Host operating systems supported include Linux & OCEOS, the space qualified RTOS from OCE Technology.
AI models trained and developed with cloud applications (e.g. Amazon SageMaker) can easily be ported to run on the HISAOR AI coprocessors using the Verisilicon Vivante Acuity tools and IDE. Support for OpenCL and OpenVX is provided as the standard interface for vision processing applications. Driver sources are provided for applications requiring bespoke AI optimisation. Host operating systems supported include Linux & OCEOS, the space qualified RTOS from OCE Technology.