FuSe-TG: Advanced Device and System Opportunities for future Neuromorphic Integrated Circuits (NICs) and Their Applications

点击翻译按钮获取中文摘要

The semiconductor industry has improved beyond imagination over the past 50 years. In 1971, the first Intel microprocessor contained 2,300 transistors. Now, in 2023, Tesla's self-driving chip contains roughly 6,000,000,000 transistors. This improvement of six orders of magnitude of added capability is unheard of in any other industry worldwide. Although the semiconductor industry has made significant progress over the years, it is now reaching the physical limits of silicon, where a transistor contains only a few silicon atoms. Hence the question is, "What is the future of semiconductors over the next 50 years?" The objective of this project is to address this question by building a research team to develop a co-design approach that includes materials, devices, and systems to create a revolutionary breakthrough for shoring up US capabilities in the next generations of the semiconductor industry. The research team in this award represents an interdisciplinary cross-section of several important disciplines within engineering. In addition, the project revolves around participation from underrepresented groups by involving them in science and technology. The neuromorphic integrated circuit (NIC) developed in this project uses the monolithically integrated 3D stacks of a novel floating-gate carbon nanotube field-effect transistor (FG-CNFET) to create a revolutionary breakthrough in semiconductor technology by extending Moore's law beyond transistor scaling in a single layer. It is an enabling technology that provides additional non-deterministic computing capability to conventional deterministic computing hardware to perform complex computation in a highly energy-efficient analog circuit. Moreover, due to its massive parallelism and reconfigurability, the proposed NIC hardware is inherently fault-tolerant, perfectly suitable for devices compatible with monolithic 3D integration, but typically suffers from high defect rates, as for instance exhibited by FG-CNFETs. The proposed NIC has three distinctive features that no existing technology can offer: 1) reconfigurable and scalable neuromorphic architecture, 2) built-in analog signal processing units capable of highly energy-efficient non-deterministic computing, and 3) utilization of advanced devices that are compatible with monolithic 3D integration, rectifying their imperfections.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.