基于MCIMX6Q6AVT10AC的可配置神经网络硬件设计  

Hardware design of configurable neural network based on MCIMX6Q6AVT10AC

作　　者：朱春燕[1] Zhu Chunyan(Suzhou Top Institute of Information Technology,Kunshan 215300,China)

机构地区：[1]苏州托普信息职业技术学院,江苏昆山215300

出　　处：《无线互联科技》2023年第3期152-155,共4页Wireless Internet Technology

基　　金：江苏省高职院校教师专业带头人高端研修项目,项目名称:智慧医疗技术研究及应用,项目编号:2021GRFX059。

摘　　要：当前硬件在不同环境下适应性较差,为此需要提高硬件的运行性能,文章提出基于MCIMX6Q6AVT10AC的可配置神经网络硬件设计研究。将STMicroelectronics的ARM微控制器STM32F429BI作为硬件的MCIMX6Q6AVT10AC微处理器,在硬件电源管理方面集成线性调节器,以实际引脚输入/输出电源状态为基础在硬件内部生成不同区域的电压水平。为满足MCIMX6Q6AVT10AC为核心硬件的可配置需求,本研究设置了具有交织模式的16位、32位和64位3种模态内存接口以及辅助驱动装置。测试结果表明,设计硬件运行期间的正峰值衰落和负峰值衰落程度分别为8.94%和8.89%,每个周期内相位峰值失真小于15 mV,能够满足相关规范要求。Current hardware has poor adaptability in different environments,so it is necessary to improve the performance of hardware.This paper proposes the design and research of configurable neural network hardware based on MCIMX6Q6AVT10AC.The ARM microcontroller STM32F429BI of STMicroelectronics is used as the MCIMX6Q6 AVT10AC microprocessor of the hardware.The linear regulator is integrated in the hardware power management,and the voltage level of different regions is generated in the hardware based on the actual pin input/output power state.In order to meet the configurable requirements of MCIMX6Q6AVT10AC as the core hardware,the 16-bit,32-bit and 64-bit memory interfaces with interleaved mode and auxiliary drivers are set up.The test results show that the positive and negative peak fading degrees of the designed hardware during operation are 8.94%and 8.89%,respectively,and the phase peak distortion is less than 15 mV in each cycle,which can meet the requirements of relevant specifications.

关 键 词：MCIMX6Q6AVT10AC 可配置神经网络 线性调节器 交织模式 内存接口 辅助驱动装置 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]