Research And Implementation Of AI Engine For Internet Of Things Devices

With the development of Internet of Things technology,people put forward higher requirements on the security,intelligence,network and energy saving of hardware equipment.Thanks to the development of AI engine technology that provides computing services,Io T devices have become much more intelligent.However,due to the development of anti-attack technology and the complex and diverse environment that Io T devices face,the intelligent process of Io T devices is faced with huge security problems.Therefore,how to provide safe computing environment services for artificial intelligence applications under the condition of balanced performance and power consumption of AI engines on the Internet of Things devices has gradually become a research and development direction for the application of artificial intelligence in the Internet of Things.Different from the AI engine that only pursues high-performance computing,the innovation and focus of this paper is to build an AI engine with high security and a variety of computing environment services around the requirements of Internet of Things devices.The main work of this paper is as follows:Based on Trust Zone technology,a high-security AI engine for Internet of Things devices is designed,so that the analysis of network models and data input can be carried out in a secure environment and the data will not be damaged.In order to make the engine face different application scenarios and provide a variety of computing environment services,a task distribution mechanism is designed.The engine can provide a variety of computing environment services under the condition of balancing computing speed,model precision and device power consumption.The limitation of computing resources,the information interaction mechanism and the specification of Open CL interface of ARM computing library under the protection of Trust Zone technology are studied.The design and implementation of the engine can be based on the parallel computing mechanism of multi-core CPU resources,so that the engine can provide secure high-performance computing environment services for models with heavy computing capacity.The RPMSG communication mechanism of heterogeneous platform and CMSIS quantitative computing interface specification are studied.The design and implementation of the engine can call the mechanism of low-power computing resources based on quantitative computing,so that the engine can provide safe low-power computing environment services for models with small computing amount.Optimize the required computing resources in the low-power computing environment and adopt the memory reuse strategy.Reduce fragmentation and latency problems caused by frequent memory requests and releases.Based on the CMSIS quantization interface specification,the quantization mechanism of the model data is realized,which enables the model to speed up the operation on the Cortex-M processor platform.Study the mechanism of the trust check,and give the scheme of the trust check according to the calculation results of the network model.Finally,a credible verification scheme under the quantitative calculation mechanism is given to enhance the security of AI engine.