NCS-FO: Biomimetic membrane networks as adaptable neuromorphic computation circuits

NCS-FO：仿生膜网络作为适应性神经形态回路garrett S. Rose （PI）和C. Patrick Collier （co-PI）田纳西大学，诺克斯维尔提案目标：使用生物学启发的方法，更好地模仿生物神经网络的结构和信号传导，并为多功能神经形态行为的新形式提供未开发的潜力。摘要：神经形态计算作为一个领域，包括模拟哺乳动物大脑行为的计算硬件。这样的系统非常适合执行对生物大脑来说通常很简单但对传统人工计算机来说很难的应用程序，例如，从任何角度识别人脸。虽然多年来研究了许多神经形态架构，但纯粹基于传统硅基电路的实现无法模仿生物神经网络的基本特性，并且可能因此需要更多的功率来实现类似的计算能力。为了解决这一限制，并揭示诸如可调离子传输和突触-神经元组织等特性在大脑复杂计算中的作用的新见解，该项目考虑了一类新的适应性大脑启发电路，该电路由具有可重构传输特性的仿生膜组成，模拟了在真实突触和通过这些突触进行通信的固态人工神经元中发现的可变权重。技术摘要：该项目与先前的电子神经形态硬件和数字软件系统有很大的不同。具体来说，在这项工作中，PI研究了：1)如何使用物理刺激来控制通过仿生膜的分子运输，以可重复地改变人工突触模拟物的突触重量；2)学习如何将基于膜的突触模拟与固态（即基于晶体管的）神经电路集成，以开发具有可控和可复制可塑性的混合设备；3)利用网络建模技术预测多神经元结构中的可变突触权重如何影响集体感知和学习功能。这个高风险、高回报的项目以研究和教育活动的综合整合计划为特色，这样以前的工作结果就可以为许多人提供信息，包括研究大脑复杂性的神经科学家，开发神经形态计算设备和大脑硬件接口的工程师，以及理解连续变化的信号通路如何帮助学习、个性和群体行为的社会科学家。拟议工作的更广泛影响在本质上是变革性的，并对科学、工程和医学的多个领域感兴趣。通过利用环保、易于配置的软材料和模仿自然的设计，PI寻求超越Exascale的计算策略，从而为下一代低功耗自主计算、分布式传感和信息存储奠定基础。我们的方法也有助于推进大脑启发的计算系统，节能电路和能够与活组织通信的合成设备，包括多功能医疗植入物，药物输送设备，疾病监测和治疗系统，以及可以帮助大脑学习和认知的技术。建模和仿真方面也将产生新的开源软件，使学生和研究人员能够探索网络排列和突触可塑性对网络功能的特征。该项目将工程、生物、物理和化学等学科结合起来，这一事实对教育产生了影响；参与这项工作的博士生将接受专门的科学培训，为他们在多个领域做出贡献做好准备。这些活动还通过激发对STEM的兴趣、增加代表性不足群体的参与和扩展工程课程来支持更广泛的影响。

NCS-FO: Biomimetic membrane networks as adaptable neuromorphic circuitsGarrett S. Rose (PI) and C. Patrick Collier (co-PI)University of Tennessee, KnoxvilleProposal Objectives:Using biologically-inspired approach that better mimics the architecture and signaling currency of biological neural networks and to offer untapped potential for new forms of multifunctional neuromorphic behaviors.Nontechnical Abstract:Neuromorphic computing as a field comprises computational hardware engineered to mimic the behavior of the mammalian brain. Such systems are well suited to perform applications that are often simple for a biological brain but prove to be difficult for classical artificial computers for example, recognizing human faces at most any angle. While many neuromorphic architectures have been studied over the years, implementations based purely on traditional silicon-based circuitry fail to mimic the basic properties of biological neural networks, and, possibly as a result, they require far more power to achieve similar computational capability. To address this limitation and to uncover new insights into the role of properties such as tunable ion transport and synapse-neuron organization on complex computation in the brain, this project considers a new class of adaptable brain-inspired circuits comprised of biomimetic membranes with reconfigurable transport properties that mimic the variable weighting found in real synapses and solid-state artificial neurons that communicate via these synapses.Technical Abstract:This project delineates significantly from prior electronic neuromorphic hardware and digital software systems. Specifically, in this work the PI investigates: 1) how molecular transport through biomimetic membranes can be gated using physical stimuli to reproducibly vary synaptic weights in artificial synaptic mimics; 2) learn how to integrate membrane-based synaptic mimics with solid-state (i.e. transistor based) neural circuitry to develop hybrid devices that exhibit controllable and reproducible plasticity; and 3) utilize network modeling techniques to predict how variable synapse weighting within multi-neuron architectures affect collective sensing and learning functionality. This high-risk, high-reward project features a comprehensive integration plan across research and educational activities such that the results from previous work inform many, including neuroscientists studying complexity in the brain, engineers developing neuromorphic computation devices and brain-hardware interfaces, and social scientists in understanding how continuously-variable signaling pathways aid in learning, individuality, and group behavior. The broader impacts of the proposed work are transformative in nature and of interest to multiple fields of science, engineering, and medicine. By tapping into environmentally friendly, easily configurable soft materials and emulating nature's design, the PI seeks to advance computational strategies past Exascale, thus setting the stage for the next generation of low-power autonomic computation, distributed sensing, and information storage. Our approach also contributes to advancing brain-inspired computing systems, energy-efficient circuitry, and synthetic devices capable of communicating with live tissues, including multifunctional medical implants, drug-delivery devices, systems for disease monitoring and treatment, and technologies that can assist the brain in learning and cognition. The modeling and simulation aspects will also generate new open-source software that will allow students and researchers to explore the characteristics of network arrangement and synapse plasticity on network functionality. The educational impact leverages the fact that this project interfaces topics in engineering, biology, physics, and chemistry; PhD students involved in this work will receive exclusive scientific training to prepare them for making contributions in multiple fields. These activities also support broader impacts by generating interest in STEM, increasing participation of underrepresented groups, and expanding engineering curricula.