SHF: Small: Disruptive Hardware for Energy-Starved Autonomous Nano-Systems

由于数字处理器在动态环境中执行导航所使用的底层软件算法时，数字处理器的功率消耗过大，因此主要依赖能源回收的移动自主机器人昆虫和虚拟虫子经常萎缩。NSF的这项研究将研究潜在的自适应控制理论软件算法，以能够像大脑一样强化学习的节能硬件来取代它们。拟议中的安装在机器人昆虫上的生物启发硬件将有利于国防相关应用、灾区远程监控的第一反应人员以及敌对和危险环境的测绘。这样的算法硬件也将有利于在飞行控制、生化过程、发电机和电信方面的应用。教育改进和研究整合是这项综合性和跨学科研究的主要目标，从而提供机会来升级当前的计算机科学和工程课程，并培养未来的工程劳动力，他们将应用多学科的知识来设计超低功率自主纳米系统。自适应动态规划(ADP)算法广泛应用于自适应最优控制系统和信号处理等大规模工程应用中。然而，ADP在微处理器上的成功限制了其在移动计算和自主机器人昆虫中的应用范围，在这些领域，电池节能至关重要。合理的ADP算法不能在便携式低功耗机器上运行，因为ADP需要很大的内存块，并且需要在合理的运行时间内进行并行处理。这项研究的目的是开发将更高级的算法从微处理器上的软件实现移植到混合信号芯片设计，然后再通过结合CMOS和忆阻器技术进行超低功耗芯片设计的方法。

Mobile autonomous robotic insects such as dragon flies and virtual bugs that chiefly rely on energy scavenging frequently languish due to the excessive power dissipation by digital processors while executing the underlying software algorithms used in navigation through dynamic environments. This NSF research will study the underlying adaptive control-theoretic software algorithms to replace them by energy-efficient hardware capable of brain-like reinforcement learning. The proposed biology-inspired hardware mounted on robotic insects will be beneficial to the defense related applications, first responders to disaster areas for remote monitoring, and mapping of hostile and hazardous environments. Such algorithmic hardware will also benefit applications in flight control, biochemical processes, power generators, and telecommunications. Educational improvement and research integration is a major goal of this integrative and cross-disciplinary research, thereby providing opportunity to upgrade the current Computer Science and Engineering curriculum and train engineering workforce of the future who will apply knowledge from multiple disciplines to design ultra-low-power autonomous nanosystems. Adaptive dynamic programming (ADP) algorithms are used in many large-scale engineering applications involving adaptive optimal control systems and signal processing. However, the success of ADP on the microprocessor has limited its scope in mobile computing and autonomous robotic insects where the battery energy preservation is paramount. Reasonable ADP algorithms cannot run on portable low power machines because ADP needs a large memory bank and requires parallel processing for a reasonable runtime. This research aims to develop methods for porting higher-level algorithms from software implementation on the microprocessor to a mixed signal CMOS chip design at first and then an ultra-low-energy chip design by combining CMOS and memristor technologies.