Collaborative Research: FuSe: A Reconfigurable Ferrolectronics Platform for Collective Computing (FALCON)

数字计算一直是现代信息革命的基石。然而，能源效率的提高和数字硬件计算成本的降低已经减速。在解决具有计算挑战性的问题时，这种减速的影响最为明显，例如组合优化(CO)中的问题，其中计算资源(能量、时间、内存)需要随问题大小呈指数级增长。此外，这些问题在现实世界中得到了广泛的应用，从人工智能到自动驾驶，再到航空公司调度，再到配电，这创造了一种实际需要，即开发新的替代方法来有效地解决此类问题。像耦合振荡器这样的模拟动力系统为解决这类难题提供了一种很有前途的基于物理的方法，因为它们表现出数字系统中没有的集体性质。然而，当前的耦合振荡器平台解决的CO问题非常有限，几乎没有可重构性，并且缺乏使数字计算如此成功的硬件算法生态系统。因此，这项研究的目标是开发一种新的模拟耦合振荡器平台，Falcon，通过跨领域的努力克服这些挑战，从开发基于振荡器的新计算模型到设计新的铁电材料、器件和实现它们的电路。这项研究将使模拟计算取得根本性进展，随后将转化为实际应用的性能改进。此外，为了扩大这项工作的影响，该团队将创建一个开放源码库，其中包含将在此项目过程中开发的计算模型、耦合体系结构和设计方案。该团队还将通过各种活动专注于劳动力发展，如组织行业日、开发新课程，以及为STEM领域中代表性不足的学生创造研究机会。在此项目中开发的基于振荡器的耦合猎鹰平台将提供定制的具有不同相位同步动力学的耦合铁芯，这些铁芯经过专门设计，可以直接绘制并在硬件中解决特定类别的CO问题。所提出的范例标志着从根本上背离了到目前为止使用的“一刀切”的方法，其中振荡器同步动力学只能被映射到单个计算模型(例如，伊辛模型)，该模型对于要求解的CO问题可能并不总是在计算上有效。这种方法可能会导致大量的预处理开销，并需要远远超出原始问题大小的额外硬件要求(振荡器节点)。此外，额外的预处理和硬件需求可能会降低(如果不是消除)模拟方法的任何性能优势，并限制其可扩展性。相比之下，Falcon耦合核心将提供多种类型的同步动态，每个核心都有助于以最小的开销将大量CO问题直接映射到硬件上。猎鹰平台将通过在铁电材料、器件和混合信号电路方面的跨堆栈创新来开发，同时紧密结合基于耦合振荡器的计算的理论基础的进步。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Digital computing has been the bedrock of the modern information revolution. However, improvements in energy efficiency and reductions in compute costs for digital hardware have decelerated. The impact of this slowdown is felt most acutely when solving computationally challenging problems, such as those in combinatorial optimization (CO), where the computational resources (energy, time, memory) required scale exponentially with problem size. Moreover, such problems find extensive real-world application in fields ranging from artificial intelligence, to autonomous driving, to airline scheduling, to power distribution, creating a practical need to develop new alternative approaches to solving such problems efficiently. Analog dynamical systems such as coupled oscillators offer a promising physics-based approach for solving such hard problems since they exhibit collective properties that are unavailable in digital systems. However, current coupled oscillator platforms address a very limited set of CO problems, exhibit little reconfigurability, and lack the hardware-algorithm ecosystem that made digital computing so successful. Therefore, the goal of this research is to develop a new analog coupled oscillator platform, FALCON, that overcomes these challenges using a cross-cutting effort that spans the development of new oscillator-based computational models to the design of new ferroelectric materials, devices, and circuits for implementing them. The research will enable fundamental advances in analog computing that will subsequently translate to performance improvements for practical applications. Furthermore, to broaden the impact of this work, the team will create an open-source repository of computational models, coupling architectures, and design schemes that will be developed through the course of this project. The team will also focus on workforce development through various activities, such as organizing an industry day, developing new courses, and creating research opportunities for students underrepresented in STEM fields.The coupled oscillator-based FALCON platform developed in this project will offer tailored coupling cores with differentiated phase synchronization dynamics that are specifically engineered such that specific classes of CO problems can be directly mapped and solved in hardware. The proposed paradigm marks a radical departure from the ‘one-size-fits-all’ approach used until now, wherein the oscillator synchronization dynamics could only be mapped to a single computational model (e.g., Ising model) that may not always be computationally efficient for the CO problem to be solved. This approach can result in significant pre-processing overhead and entail additional hardware requirements (oscillator nodes) that far exceed the size of the original problem. Moreover, the additional pre-processing and hardware needs can reduce, if not eliminate, any performance advantage of the analog approach, as well as limit its scalability. In contrast, the FALCON coupling cores will offer multiple types of synchronization dynamics, with each core facilitating the mapping of a large number of CO problems directly onto the hardware with minimal overhead. The FALCON platform will be developed through across-the-stack innovation in ferroelectric materials, devices, and mixed-signal circuits, in close conjunction with the advancement of the theoretical foundations of coupled oscillator-based computing.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.