Green Computing Innovations for Biomedical Imaging Applications

我们的研究计划旨在推动绿色嵌入式计算技术的发展，目标是移动计算密集型生物医学成像应用。鉴于这类应用程序运行在电池供电的便携式设备上，嵌入式系统设计者和程序员面临着多个相互冲突的要求，如高计算吞吐量、低功耗、高数据质量和低实施成本等。这一技术挑战可以通过为运行时系统适配提供特定于应用的体系结构支持来解决，由此动态地调整一组功率管理且灵活的硬件/软件组件(例如，嵌入式CPU、GPU、FPGA、专用处理单元及其编程库)，以在目标应用执行(例如，超声成像)期间提高系统能量效率。我们的短期目标是研究和开发：1)创新的低复杂性算法，用于专门针对超声成像应用的多维数据采集、处理和分析；以及2)创新的可配置硬件/软件架构，支持这些算法的高效和灵活执行。我们的长期目标是研究和开发：1)预测系统级模型，生成电池感知的能量预算估计，以指导能源约束下的应用程序执行；以及2)基于图形的算法，用于动态硬件/软件配置管理，以最大化建筑能效，同时保持所需的应用感知服务质量。我们的最终目标是促进嵌入式系统资源、流程和数据的模型驱动设计和优化，旨在实现目标生物医学应用程序执行期间交付的性能和电池寿命之间的有效平衡。该建议支持强大的HQP培训环境，在该环境中，受监督人员将获得涉及多个技术科目的适用知识和技能，如超声成像、信号处理、硬件/软件协同设计、可重构架构、能效计算、系统级建模和优化算法。我们预计，我们的研究成果和训练有素的HQP将为加拿大高科技公司和公共部门组织在移动嵌入式计算系统和应用领域开发创新解决方案带来实质性好处。

Our research program aims to advance green embedded computing technologies targeting mobile computation-intensive biomedical imaging applications. Given that such apps run on battery-powered portable devices, embedded system designers and programmers are faced with multiple conflicting requirements, such as high computational throughput, low power consumption, high data quality, and low implementation cost, among others. This technical challenge can be tackled by providing application-specific architectural support for runtime system adaptation, whereby a set of power-managed and flexible hardware/software components (e.g., embedded CPUs, GPUs, FPGAs, specialized processing units, and their programming libraries) are dynamically tuned to enhance system energy efficiency during targeted application execution (e.g., ultrasound image formation). Our short-term objectives are to research and develop: 1) innovative low-complexity algorithms for multidimensional data acquisition, processing, and analysis that specifically target ultrasound imaging applications, and 2) innovative configurable hardware/software architectures that support efficient and flexible execution of these algorithms. Our long-term objectives are to research and develop: 1) predictive system-level models producing battery-aware energy budget estimates to guide application execution under energy constraints, and 2) graph-based algorithms for dynamic hardware/software configuration management to maximize architectural energy efficiency while maintaining a desired application-aware quality of service. Our ultimate goal is to facilitate model-driven design and optimization of embedded system resources, processes, and data, aiming to achieve an effective balance between delivered performance and battery life during execution of targeted biomedical applications. This proposal supports a strong HQP training environment, in which the supervised personnel will gain applicable knowledge and skills cutting across multiple technical subjects, such as ultrasound imaging, signal processing, hardware/software codesign, reconfigurable architectures, energy-efficient computing, system-level modeling, and optimization algorithms. We anticipate that our research findings and trained HQP will be of substantial benefit to Canadian high-tech companies and public-sector organizations developing innovative solutions in the area of mobile embedded computing systems and applications.