Collaborative Research: FET: Medium: Probabilistic Computing Through Integrated Nano-devices - A Device to Systems Approach

世界对计算能力的需求一直在稳步增长，而且看不到尽头。允许用户找到几乎所有问题的答案的搜索引擎预计将在不到一秒的时间内提供结果，同时只需输入几个字就可以很高概率地理解客户。这不是一种精确并给出“正确”答案的计算类型；相反，它提供了更具关联性的解决方案，模仿了人类解决问题的方式。事实上，现实世界中的许多问题本质上都是概率问题，传统的计算方案并不能针对这些任务进行优化。1982年，诺贝尔奖得主理查德·费曼在承认这一事实时说：“模拟概率性质的方法可能仍然是用计算机来模拟，而计算机本身就是概率的……因此，它变成了我所说的概率计算机，其中的输出不是输入的唯一函数。为了促进科学进步，这项提议旨在探索、建模和建造最终实现这种概率计算机的硬件组件和电路，这将极大地造福于整个社会。此外，正在为本科生和研究生创建教育工具、新课程和培训机会，使他们接触到概率计算的设备到系统研究计划，以便为新的电子学时代做好准备。该项目采用设备到系统的方法，涵盖从单个设备和小电路到数千个设备的模拟。它从各种不同的角度解决了如何在硬件中实现概率功能的问题。这项建议的主要目标是通过实验演示集成概率位(p位)和p电路，并根据实验输入通过系统级应用中的关键优点系数来量化规模化概率计算机的优势，从而在发展概率计算方面迈出下一步。这是通过在磁隧道结(MTJ)配置中使用不稳定的磁体作为随机数生成器来实现的，该随机数生成器通过使用场效应晶体管在适当的电路布局中变得可调。初步预测估计，如果与传统晶体管相结合的集成MTJ能够扩大规模，那么与传统半导体技术相比，在关键的优点系数方面有望实现数量级的改进，例如概率采样器可以通过的每秒统计独立样本数(也称为每秒翻转数)。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。

The world has seen a steady increase in demand for computational power, with no end in sight. Search engines that allow users to find answers to pretty much all of their questions are expected to deliver results in less than a second, while understanding the customer with a high probability based on just a few words of input. This is not the type of computation that is precise and gives the “correct” answer; it instead provides solutions that are more associative, mimicking the human way of addressing problems. In fact, many problems in the real world are probabilistic in nature, and conventional computing schemes are not optimized for these tasks. In 1982 Nobel Prize winner Richard Feynman stated in recognition of this fact: “The way to simulate a probabilistic nature might still be by a computer which itself is probabilistic... So, it becomes what I’ll call a probabilistic computer, in which the output is not a unique function of the input.” To promote the progress of science, this proposal aims to explore, model and build hardware components and circuits that ultimately enable such a probabilistic computer, which will greatly benefit the society as a whole. Moreover, educational tools, new courses and training opportunities both for undergraduate and graduate students are being created that expose them to a device-to-systems research program on probabilistic computing in order to prepare them for the new era of electronics.The project adopts a Device-to-Systems approach that covers experiments from single devices and small circuits all the way up to simulations with thousands of devices. It addresses the question of how to implement probabilistic functionality in hardware from a variety of different angles. The key objective of this proposal is to take the next step in the development of probabilistic computing by experimental demonstrations of integrated probabilistic bits (p-bits) and p-circuits and quantifying advantages of scaled probabilistic computers through key figures-of-merits in system-level applications based on experimental input. This is being achieved by employing unstable magnets in a magnetic tunneling junction (MTJ) configuration as random number generators that become tunable by the use of field-effect transistors in a suitable circuit layout. Initial projections estimate that if integrated MTJs coupled with conventional transistors can be scaled up, one can expect to achieve orders of magnitude improvements compared to what is achievable in conventional semiconductor technology in key figures-of-merits, such as the number of statistically independent samples per second (also referred to as flips per second) that a probabilistic sampler can go through.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.