FET: Medium: ROCS: Recurrent Oscillatory Computing Systems for Rapid Solution of NP-Complete and Deep Learning Problems

在当今世界，从航班调度到医疗保健，计算难题比比皆是。这些难题中的大部分可以简化为一种称为伊辛问题的形式，它与铁磁材料的基础物理密切相关。该项目基于最近设计的一种方法，使用连接的互补金属氧化物半导体（CMOS）振荡器网络，即节拍器和祖父时钟的电子等效物，快速有效地解决电子硬件中的Ising问题。使用这种方法能够比目前可能的方法更快地解决现实世界中的大型问题，将对社会产生广泛而有益的破坏性影响。该项目的活动包括课程开发、向高中生推广、传播和互动的年度讲习班以及科学和设计工具基础设施开发。与之前的大型、昂贵且不适合低成本批量生产的伊辛机器方法不同，该方法是一种纯粹的经典方案，不依赖于量子现象或新型纳米器件。它可以使用传统的CMOS电子器件来实现，它具有许多优点：可扩展性/小型化（即，在物理上小的系统中有非常多的旋转），完善的设计流程和工具，基本上保证首次工作的硬件，非常低的功耗操作，与控制和I/O逻辑无缝集成，易于通过USB等标准接口编程，以及低成本的批量生产。另一个关键优势与可变性有关，这是纳米级CMOS的一个重要问题。与其他方案不同，由于可变性而导致性能下降，这种方法可以通过简单的基于vco的校准来消除可变性，从而使所有振荡器达到相同的频率。然而，另一个关键的潜在优势源于所提出方案的连续/模拟性质（与纯数字算法相反）。计算实验表明，该方案找到伊辛问题的好解所需的时间随着自旋数的增加而缓慢增长。这是相对于数字算法的一个重要的潜在优势，因为硬件尺寸可以扩展到大量的旋转。此外，几乎任何类型的非线性振荡器（不仅仅是CMOS）都可以用于实现该方案，包括光学，微电子机械系统，生化，基于自旋扭矩器件等振荡器。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Hard computational problems abound in today's world, from airline scheduling to healthcare. Large classes of these hard problems can be reduced to a form known as the Ising problem, which is closely related to the underlying physics of ferromagnetic materials. This project is based on a recently devised way to solve the Ising problem quickly and effectively in electronic hardware using networks of connected Complementary Metal Oxide Semiconductor (CMOS) oscillators, i.e., electronic equivalents of metronomes and grandfather clocks. Being able to solve large real-world problems much more quickly using this approach than what is currently possible will have broad and beneficially disruptive effects on society. The project's activities include course development, outreach to high-school students, yearly workshops for dissemination and interaction, and scientific and design tool infrastructure development.Unlike previous Ising machine approaches, which are large, expensive and ill-suited to low-cost mass production, the proposed approach is a purely classical scheme that does not rely on quantum phenomena or novel nano-devices. It can be implemented using conventional CMOS electronics, which has many advantages: scalability / miniaturisability (i.e., very large numbers of spins in a physically small system), well-established design processes and tools that essentially guarantee first-time working hardware, very low power operation, seamless integration with control and I/O logic, easy programmability via standard interfaces like USB, and low cost mass production. Another key advantage relates to variability, a significant problem in nanoscale CMOS. Unlike other schemes, where performance deteriorates due to variability, this approach can essentially eliminate variability by means of simple VCO-based calibration to bring all the oscillators to the same frequency. Yet another key potential advantage stems from the continuous/analog nature of the proposed scheme (as opposed to purely digital algorithms). Computational experiments indicate that the time the scheme takes to find good solutions of the Ising problem grows only very slowly with respect to the number of spins.This is a significant potential advantage over digital algorithms as hardware sizes scale up to large numbers of spins. In addition, virtually any type of nonlinear oscillator (not just CMOS) can be used to implement this scheme, including optical, micro-electronic mechanical systems, biochemical, spin torque device based, etc., oscillators.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.