A High Throughput Parallel Processing System Using Three Dimensional Optoelectronic Interconnect

9422452威尔这项活动将研究一种高度并行的计算机，其特点是具有三维光学互连网络。这种极其紧凑的系统旨在处理图像处理等应用中的高吞吐量数据流。这种高密度和大I/O带宽源于硅基处理器之间的光通信通道。通过使用工作在硅对其透明的波长的集成光电子器件，在堆叠的硅电路之间形成贯穿晶片的垂直光通信通道。薄膜光电子器件直接与SI集成电路结合，实现这种自包含的垂直光互连。为了构建该系统，包含数字处理器、网络接口和路由器的硅芯片以及用于薄膜器件的模拟接口电路通过标准IC铸造厂制造。芯片随后被后处理，以使用标准、低成本、高成品率的微制造技术来集成薄膜光电子器件。然后将这些芯片的二维平面堆叠起来，形成三维系统，该系统通过垂直光学通道连接。从长远来看，集成穿透晶圆光互连的使用将导致极高密度的系统(4英寸立方体中的4096个处理器具有高吞吐量(820 Gb/s外部I/O带宽，3.2 Tb/s总带宽)。拟议的系统将加速在并行计算架构、模拟电路、光电子器件和集成以及晶体生长领域建立一个由教职员工和学生组成的跨学科合作小组。正是这种类型的合作能够实现一个革命性的、主要是电子的系统，具有简单的点对点光电互连，并有助于为每个研究人员确定必须从系统角度考虑的关键点和权衡。

9422452 Wills This activity will study a highly parallel computer that features a three dimensional, optical interconnection network. The extremely compact system is design to process high throughput data streams in applications such as image processing. The high density and large I/O bandwidth result from optical communication channels between Si-based processors. By using integrated optoelectronic devices operating at wavelengths to which silicon is transparent, through-wafer vertical optical communications channels are formed between stacked silicon circuitry. Thin film optoelectronic devices are bonded directly to SI integrated circuits to realize this self contained vertical optical interconnect. To construct the system, silicon chips containing a digital processor, network interface, and router, plus analog interface circuitry for the thin film devices, are fabricated through a standard IC foundry. The chips are then post processed to integrate the thin film optoelectronic devices using standard, low cost, high yield microfabrication techniques. Two dimensional planes of these chips are then stacked to form the three dimensional system which is connected by the vertical optical channels. The use of integrated through-wafer optical interconnect will lead, long term, to extremely dense systems (4096 processors in a four inch cube with high throughput (820 Gbits/sec external I/O bandwidth, 3.2 Tbits/sec aggregate bandwidth). The proposed system will accelerate an established, interdisciplinary collaborative group of faculty and students in the areas of parallel computational architectures, analog circuits, optoelectronic devices and integration, and crystal growth. It is this type of collaboration that enables the realization of a revolutionary, predominantly electronic system with simple point-to-point optoelectronic interconnects, and helps to identify for each of the researchers the critical points and trade-offs which must be considered from a system perspective.