Multiplexing Techniques for Scalable Wireless Interconnects at sub-THz Frequencies: Circuits-EM-Communication Codesign Approach

ECCS-1408490 PI：Kushik Sengupta，普林斯顿大学该提案旨在研究和开发亚太赫兹频率下无线互连的空间复用架构，作为可扩展的节能解决方案，以达到每秒1 TB（1 Tb/s）。随着我们进入地球级计算时代，这些处理器处理的大量数据将需要非常大的带宽，而目前电气或光学互连解决方案都无法提供。将电互连缩放到更高数据速率的当前方法或者受到可用带宽密度（Gb/s/mm 2）、能量成本、在驱动高速数据通过长且有损的物理迹线时的电路复杂性的限制，或者受到在受约束的形状因子中可能容纳的并行物理迹线的最大数量的限制。THz频率附近的无线互连很有前途，但10 Gb/s的无线数据速率和高能量/比特要求远远不能满足未来片外互连的带宽要求。在这个建议中，我们的目标是调查技术，其中的信道容量可以增加许多倍，使用通信理论的空间域复用技术。在相同的总功率约束下，这样的架构具有多个数量级的信道容量，从而提供了一个可扩展的解决方案，以无线Tb/s互连。该建议中的一个关键组成部分是将联合收割机、高频电路和系统以及天线与通信理论技术无缝地结合起来，以按数量级增加容量和数据速率，这在单向分区方法中是不可能的。印刷电路板（PCB）上的金属基互连走线是最常见的芯片-芯片互连方法。然而，在专用服务器系统、高性能计算或甚至便携式设备中，越来越需要计算能力来处理越来越多的数据，这要求从处理器到外围设备的通信数据速率成比例地缩放。在大多数情况下，输入输出引脚的数量受到形状因子的限制，这对所有处理器之间的通信能力造成了瓶颈。在这项提案中，我们研究了使用位于太赫兹频谱部分（微波和红外线之间）的极高频电磁波在芯片组之间建立无缝无线通信链路的技术。移动到如此高的频率使我们能够利用维持如此高的数据速率所需的更高带宽。此外，我们调查的技术，以增加通信链路容量的另一个顺序，通过空间复用技术在短距离通信设置。该项目的成功将为从高性能计算到互联网数据中心的各种应用带来新形式的智能互连解决方案。这项研究工作的结果预计也将对推进太赫兹电子领域产生重大影响，使成像和传感等多种应用受益。从更广泛的角度来看，这将对通信和计算领域的全新技术产生重大影响，这不仅使我们成为一个更加互联的社会，而且还将推动其他应用科学领域的研究。这项研究还有望培养多学科领域的研究生和本科生，这对解决未来具有挑战性的研究问题至关重要。

Multiplexing Techniques for Scalable Wireless Interconnects at THz FrequenciesECCS-1408490PI: Kushik Sengupta, Princeton University This proposal aims to investigate and develop spatially multiplexed architectures for wireless interconnects at sub-THz frequencies as scalable, energy-efficient solution towards one terabit per second (1 Tb/s). As we enter the era of terra-scale computing, massive amounts of data crunching by these processors will require inordinately large amount of bandwidth, not currently served by either electrical or optical interconnect solutions. Current methods of scaling of electrical interconnects to higher data rates are either limited by the available bandwidth density (Gb/s/mm2), energy cost, the circuit complexities in driving high-speed data through the long and lossy physical traces, or by the maximum number of parallel physical traces possible to accommodate in a constrained form factor. Wireless interconnects near THz frequencies are promising , but wireless data rates of 10Gb/s and the high energy/bit requirement, falls way short of meeting the bandwidth requirements for future off-chip interconnects. In this proposal, we aim to investigate techniques where the capacity of the channel can be increased many-fold using communication theoretic spatial-domain multiplexing techniques. Under the same total power constraint, such architectures have orders of magnitude more channel capacity, thereby providing a scalable solution towards wireless Tb/s interconnects. A key component in this proposal is to combine seamlessly, high-frequency circuits and systems and antennas with communication-theoretic techniques to increase capacity and data-rates by orders of magnitude, not otherwise possible in a single directional partitioned approach. Metal-based interconnect traces on printed circuit boards(PCB) serve as the most common method of chip-chip interconnects. However, increasing need of computational power to crunch more and more data in specialized server systems, high-performance computing or even portable devices, requires that communication data-rate from the processor to the peripherals be scaled proportionately. In most cases, the number of input-output pins is limited by the form factor, which puts a bottleneck on communication capacity among all the processors. In this proposal, we investigate techniques to use very high-frequency electromagnetic waves located in the Terahertz portion of the spectrum (between microwaves and infra-red) to establish seamless wireless communication links among the chipsets. Moving to such high frequencies enables us to exploit orders of magnitude higher bandwidth needed for sustaining such high data rates. Additionally, we investigate techniques to increase the communication links capacity by another order through spatial multiplexing techniques in a short-range communication setting. The success of this project is envisioned to bring new forms of smart interconnect solutions for a host of various applications from high-performance computing to internet data centers. The results of this research effort are also expected to have major impact in advancing the field of THz electronics benefitting diverse applications such as imaging and sensing. In a broader vision, this will have major impacts in radically new technologies in communication and computation, which not only makes us a more connected society, but also fuel research in other areas of applied science. This research is also expected to train both graduate and undergraduate students in multi-disciplinary fields, which are vitally important for solving challenging research problems for the future.