CRII: FET: 3D Printed Application-Specific Neuromorphic Circuits: Design, Fabrication, and Implementation

该奖项全部或部分由2021年美国救援计划法案（Public Law 117-2）资助。现代硅基CMOS系统在处理速度和能耗方面面临限制，无法再通过摩尔定律驱动的缩小规模努力来解决。传统冯·诺依曼计算机体系结构中的主要固有挑战是由计算机处理和存储器单元之间的数据传输引起的延迟，也称为冯·诺依曼瓶颈。神经形态计算，模仿人类大脑的节能和高度并行的操作，是一个有前途的平台，以解决这一挑战。建立在忆阻器设备上的神经形态电路使得能够以电阻状态的形式进行有效的信息存储、生物突触功能的仿真和存储器内计算（即，经由矩阵向量乘法）。神经形态计算技术的大规模采用目前受到材料，设备和算法层面的多方面挑战的阻碍，这些挑战因传统微制造技术的长原型周期而加剧。NSF的这项提案旨在开发一个基于纳米级3D打印的数据驱动、高吞吐量和可定制的神经形态电路制造平台，这将在材料和器件架构层面实现设计自由。该项目的更广泛影响包括新一代专用计算硬件，这对于推动未来人工智能的实施是不可或缺的，对提高美国在IC行业的竞争力至关重要。该项目还将支持本科和K12级别的多个跨学科研究和教学活动，特别关注代表性不足的社区的参与。该项目旨在实施纳米级3D打印技术，以开发特定应用的神经形态电路。这一目标将通过最先进的3D打印机实现，该打印机通过双光子聚合技术实现亚微米分辨率，并支持一系列功能性和可调节的打印树脂。将执行一种新的贝叶斯优化方法，以揭示最佳的材料和工艺参数，将产生具有所需特性的神经形态设备。将利用集成的电气表征框架来探索3D打印忆阻器交叉杆阵列的突触功能和器件性能指标。最终确定的特定应用的神经形态电路将承担标准手指数据集和真实肿瘤细胞图像数据的图像分类的硬件级基准测试。使用纳米级3D打印平台制造复杂的神经形态电路将为定制和可扩展制造特定应用的计算硬件系统提供新的机会。该奖项反映了NSF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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Modern silicon-based CMOS systems face limitations in terms of processing speeds and energy consumption that can no longer be addressed via Moore’s-law-driven scaling-down efforts. The main inherent challenge in conventional von Neumann computer architecture is the latency caused by the data transfer between computer processing and memory units, also known as von Neumann bottleneck. Neuromorphic computing, emulating the energy-efficient and highly-parallel operations of the human brain, stands as a promising platform to address this challenge. Neuromorphic circuitry built on memristor devices enables efficient information storage in the form of resistance states, emulation of biological synaptic functionalities, and in-memory computing (i.e. via matrix-vector multiplication). Large-scale adoption of neuromorphic computing technology is currently hindered by multifaceted challenges in the materials, device and algorithm levels that are exacerbated by long-prototyping cycles of conventional microfabrication techniques. This NSF proposal aims to develop a data-driven, high-throughput and customizable manufacturing platform for neuromorphic circuits, based on nanoscale 3D printing, that will enable design freedom in both the materials and device architecture levels. The project’s broader impacts include a new generation of application-specific computational hardware that is indispensable for fueling the future implementations of AI and critical to increase U.S. competitiveness in the IC industry. The project will also support a multitude of interdisciplinary research and teaching activities at undergraduate and K12 levels with a specific focus on the involvement of underrepresented communities.The project aims to implement a nanoscale 3D printing technique to develop application-specific neuromorphic circuits. This goal will be enabled through a state-of-the-art 3D printer that achieves sub-micron resolutions via the two-photon polymerization technique, and supports a range of functional and adjustable printing resins. A novel Bayesian optimization approach will be executed to uncover the optimum set of material and process parameters that will yield neuromorphic devices with desired characteristics. An integrated electrical characterization framework will be utilized to explore the synaptic functionalities and device performance metrics of 3D printed memristor crossbar arrays. Finalized application-specific neuromorphic circuitry will undertake the hardware-level benchmarking on image classification of standard hand-digit datasets and real-life tumor-cell image data. Use of a nanoscale 3D printing platform for complex neuromorphic circuitry fabrication will enable new opportunities for customization and scalable manufacturing of application-specific computational hardware systems.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.