Collaborative Research: FET: Small: Reservoir Computing with Ion-Channel-Based Memristors

通用数字计算范式面临着严重的限制，这可能会通过采用更多受大脑启发的方法来解决。然而，尽管人工智能算法最近取得了重大进展，但在开发能够模拟大脑功能和能效的硬件方面仍有大量工作要做。该项目旨在开发利用基于离子通道的忆阻器(ICM)的高度非线性动力学的新型物理油藏计算体系结构。这项工作将开发新的制造方法、学习算法和网络设计，以利用所提议的材料的独特性质。这将导致一种具有生物兼容性和高能效硬件的大脑启发计算的新范式。长期目标是开发低成本、高能效、高度可调、模块化、容错和自我修复的生物分子神经网络和组织。预期的应用包括信号处理、传感器内和近传感器计算、神经工程、人工智能和后硅技术。教育影响利用了这个项目与工程、生物、物理和化学主题相结合的事实。参与其中的学生将接受专门的科学培训，这将帮助他们为在多个领域做出贡献做好准备。传统的储备库计算机使用一个储存层，该储存层包含具有随机分配的突触权重的神经元的循环连接，然后是一个读出层，其非易失性突触权重被训练。假设储存层和读出层都可以使用表现出易失性和非易失性存储器配置的ICM组合成单一层。这些装置中的基本元件是模仿生物膜的组成和功能的绝缘类脂膜。在电压激活离子通道的存在下，这些合成类脂膜可以表现出由膜和离子通道动力学引起的电压依赖的记忆。这项拟议的研究特别旨在：1)了解如何利用ICM的特定属性来进行储层计算；2)设计针对我们的设备在储集层和读出层的非线性短期记忆动力学的体系结构，并可能将它们合并到单一层中；3)使用ICM的交叉开关阵列进行实验验证；以及4)推广新的储液器架构，使其可以与其他非易失性和易失性忆阻器一起使用，可能是固态的。该项目由计算机和信息科学与工程局(CESE)计算和通信基础部门(CCF)的软件和硬件基础集群和既定的激励竞争研究计划(EPSCoR)共同资助。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 general-purpose digital computing paradigm faces severe limitations, which could potentially be addressed by adopting more brain-inspired approaches. However, despite significant recent advancements in artificial intelligence algorithms, substantial work remains in developing hardware capable of emulating the functionality and energy efficiency of the brain. This project aims to develop novel physical reservoir computing architectures that harness the highly nonlinear dynamics of ion-channel-based memristors (ICMs). This work will develop new fabrication methods, learning algorithms, and network designs to take advantage of the unique properties of the proposed materials. This will lead towards a new paradigm for brain-inspired computing with biocompatible and highly energy-efficient hardware. The long-term goal is to develop low-cost, energy-efficient, highly tunable, modular, fault-tolerant, and self-healing biomolecular neural networks and tissues. The intended applications include signal processing, in-sensor and near-sensor computing, neuro-engineering, artificial intelligence, and post-silicon technologies. The educational impact leverages the fact that this project interfaces with topics in engineering, biology, physics, and chemistry. Students who are involved will receive exclusive scientific training, which will help prepare them for making contributions in multiple fields.Traditional reservoir computers use a reservoir layer comprising a recurrent connection of neurons with randomly assigned synaptic weights, followed by a readout layer whose nonvolatile synaptic weights are trained. The hypothesis is that both reservoir and readout layers can be combined into a single layer using ICMs that exhibit collocated volatile and non-volatile memories. The basic element in these devices is an insulating lipid membrane that mimics the composition and function of biological membranes. In the presence of voltage-activated ion channels, these synthetic lipid membranes can exhibit voltage-dependent memristance caused by membrane and ion channel dynamics. This proposed research specifically aims to: 1) understand how the specific properties of the ICMs can be harnessed for reservoir computing; 2) design architectures tailored to the nonlinear short-term memory dynamics of our devices in both the reservoir and the readout layers, and possibly combine them into a single layer; 3) experimentally validate using crossbar arrays of the ICMs; and 4) generalize the new reservoir architecture so it can be used with other nonvolatile and volatile memristors, possibly solid-state in nature.This project is jointly funded by the Software and Hardware Foundations Cluster of the Division of Computing and Communication Foundations (CCF) in the Directorate for Computer and Information Science and Engineering (CISE) and the Established Program to Stimulate Competitive Research (EPSCoR).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.