ASCENT: Heterogeneously Integrated and AI-Empowered Millimeter-Wave Wide-Bandgap Transmitter Array towards Energy- and Spectrum-Efficient Next-G Communications

由于数据密集型应用的出现和通信网络的快速增长，无线生态系统的能耗一直在不断增加。目前，全世界所有通信网络的总用电量估计为每年几千太瓦时，预计未来几十年这一数字还会成倍增加。然而，在无线通信中为增加带宽而新采用的毫米波(MmW)无线电系统，与工作在较低频段的传统无线电系统相比，能量效率低得多。该项目旨在通过异质集成(HI)和先进封装，利用高效宽带隙(WBG)半导体技术，从根本上提高毫米波系统的能效，并将通过人工智能(AI)进一步增强，以实现实时效率优化。通过降低能耗，基于WBG的高效毫米波系统的无处不在的部署将缓解无线网络的大量碳排放，并最终有助于无线产业向净零排放演进。此外，该项目还将促进毫米波频段的资本化，缓解6 GHz以下频谱的拥堵，以满足我国迫切的频谱可持续性需求。该计划的影响将通过综合教育努力和外展活动进一步扩大。该项目旨在通过先进的高功率WBG功率放大器(PA)电路的异质集成以及人工智能辅助从单个元件到阵列系统级别的动态优化，极大地提高毫米波阵列系统的能效。首先，将开发多芯片组装和封装集成，将WBG芯片与封装基板无缝嵌入，并与其他功能块和天线互联。开发的HI工艺还将采用基于“淋浴式”微型喷嘴结构的兼容冷却解决方案，以便在系统运行期间有效地管理WBG层的热状况。此外，高分辨率的热测试结构将被共同集成，为人工智能培训提供实时温度传感数据。其次，提出了一种新型的WBG毫米波功率放大器结构--混合非对称负载调制平衡放大器(H-ALMBA)。这种新颖的架构提供了无与伦比的效率和高线性度。H-ALMBA还配备了强大的内部无变容管可重构性，使PA能够针对封装高度紧凑的毫米波系统中的严重环境波动进行自我修复，包括天线扫描阻抗、温度、机械变形和工艺变化。第三，将开发一个人工智能辅助控制器来动态调整PA阵列的操作参数，其中控制回路通过空中观测链路闭合，以获取主波束数据。在该控制方案中，将研究算子学习和多模式数据融合，通过考虑各种观测参数的不同影响来学习控制策略。为了实现AI引擎的低功耗和低延迟实施，将开发一个硬件友好的培训框架，以高精度自动压缩和量化AI辅助控制器。这一奖项反映了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。

Due to the advent of data-intensive applications and rapid growth of communication networks, the energy consumption of wireless ecosystem has been ever-increasing. Currently, the total electricity consumption of all communication networks in the world is estimated to be a few thousand terawatt-hours per year, and this number is projected to be multiplied in the coming decades. Nevertheless, the millimeter-wave (mmW) radio systems newly adopted in wireless communications to increase the bandwidth are much less energy-efficient than the traditional radio systems operating at lower frequency bands. This project aims to fundamentally improve the energy efficiency of mmW systems by exploiting the highly efficient wide-bandgap (WBG) semiconductor technologies through heterogeneous integration (HI) and advanced packaging, which will be further enhanced by artificial intelligence (AI) for real-time efficiency optimization. By reducing the energy consumption, the ubiquitous deployment of high-efficiency WBG-based mmW systems will mitigate the massive carbon emission of wireless networks and eventually contribute to the evolution of wireless industry towards 'Net-Zero' emission. Moreover, this project will promote the capitalization of mmW frequency bands and ease the congestion of the sub-6-GHz spectrum to address the emergent need of spectrum sustainability in our nation. The impact of this program will be further expanded through integrated educational efforts and outreach activities.This project aims to vastly enhance the energy efficiency of mmW array system through advanced heterogeneous integration of high-power WBG power amplifier (PA) circuits in conjunction with AI-assisted dynamic optimization from individual elements to the array system level. First, multi-chip assembly and packaging integration will be developed to seamlessly embed WBG chips with the package substrate and interconnect them with other functional blocks and antennas. The developed HI process will also involve compatible cooling solutions based on a "shower-type" microfabricated nozzle structure to effectively manage the thermal condition of the WBG layer during system operation. Moreover, high-resolution thermal test structures will be co-integrated to provide real-time temperature sensing data for AI training. Second, a novel WBG mmW PA architecture, called hybrid asymmetrical load-modulated balanced amplifier (H-ALMBA), is proposed. This novel architecture offers unparalleled efficiency as well as high linearity. The H-ALMBA is also equipped with a powerful intrinsic varactor-less reconfigurability, enabling self-healing of the PA against severe environmental fluctuations in the highly compact mmW system in package, including antenna scan impedance, temperature, mechanical deformation, and process variation. Third, an AI-assisted controller will be developed to dynamically tune the operation parameters of the PA array, in which the control loop is closed via an over-the-air observation link for acquisition of the main-beam data. In this control scheme, operator-learning and multi-modal data fusion will be investigated to learn the control policy by considering the different impacts of various observation parameters. To achieve low-power and low-latency implementation of the AI engine, a hardware-friendly training framework will be developed to automatically compress and quantize the AI-assisted controller with high accuracy.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.