随着计算机软硬件技术的持续进步,机载计算平台集成的任务功能日益增多,导致平台内部的计算需求在规模和复杂性上不断攀升。面对智能化应用的迅猛增长,传统的单一处理器架构已不足以应对多样化的复杂任务。为此,基于OpenVPX标准,定义并设计了一种符合硬件开放式架构的3U异构融合处理模块,以适应多种复杂任务的需求。文中还提出了一种异构计算资源池化技术,旨在实现多类型任务应用的快速部署和高效运行,同时降低通信延迟,显著提升计算平台的处理能力和适用性。最后进行了实验验证,结果表明与多CPU架构相比,文中所设计的异构融合处理模块在执行特定神经网络算法时,处理时间缩短了约4.8倍,证明了其在性能上的显著提升。该研究成果不仅展示了异构融合处理模块在机载智能计算应用中的显著性能优势,而且为航空计算平台的未来发展提供了创新的解决方案和技术支持。

With the continuous progress of computer hardware and software technology, the task functions integrated in the airborne computing platform are increasing day by day, resulting in the scale and complexity of the internal computing needs of the platform. In the face of the rapid growth of intelligent applications, the traditional single processor architecture is no longer enough to cope with a variety of complex tasks. Therefore, based on the OpenVPX standard, this study defines and designs a 3U heterogeneous fusion processing module that conforms to the hardware open architecture to meet the needs of a variety of complex tasks. This research also proposes a heterogeneous computing resource pooling technology, which aims to achieve rapid deployment and efficient operation of multi-type task applications, while reducing communication latency, and significantly improving the processing power and applicability of computing platforms. Experimental verification shows that, compared with multi-CPU architecture, the processing time of the heterogeneous fusion processing module designed in this paper is about 4.8 times shorter when executing specific neural network algorithms, which proves that its performance is significantly improved. The results of this study not only demonstrate the significant performance advantages of heterogeneous fusion processing modules in airborne intelligent computing applications, but also provide innovative solutions and technical support for the future development of aviation computing platforms.