EEPW论坛

Luxtera: Photonic Revolutionary
铜互联的道路是否已经终结？Is the Fate of Copper Interconnect Sealed?
By Kevin Krewell
光通信在某些市场大行其道，在这些市场上，使用光纤的成本和复杂性是无法避免的。对于大批量应用而言，铜线缆作为通信介质仍然占据统治地位，但未来将属于光纤和无线通信。尽管很多公司，包括Intel，都正在研究这些未来的光解决方案，但一家起步公司Luxtera却计划在2006年向市场推出一种成本经济性好的解决方案，来加速这一技术转移进程。Optical communications are prevalent in some markets where the expense and complexity of using optical fibers are necessary. For volume applications, copper wire still dominates as the communications medium, but the future belongs to optical and wireless communications. Although many companies, including Intel, are researching these future optical solutions, one startup, Luxtera, plans to accelerate the transition by providing the market with a cost-effective solution in 2006.
Luxtera所宣布的技术方案是一种可以让传统的硅圆片于光学元件相匹配并在硅集成电路中进行调制的方法，这种调制避免了使用更奇异（不兼容）的材料。今年早些时候，Intel展示了如下结果：在硅上可以制作激光器件，但Luxtera的解决方案可以让激光器在另一块单独的芯片上制造。Luxtera的设计消除了使用昂贵的III－V化合物半导体材料（如砷化镓和磷化铟）的必要。虽然硅对于可见光而言是不透明的，但这种材料对红外光却具有惊人的透光性，而且硅结构可以用来对红外激光进行引导和修正。The breakthrough Luxtera claims to have is a way to mate traditional silicon-wafer technology to optical components and to modulate the light wave in the silicon without using more-exotic materials. Earlier in the year, Intel demonstrated that a laser could be fabricated in silicon as well, but Luxtera's solution allows the laser to be fabricated on a separate die. Luxtera's design eliminates the need to use expensive group III–V compounds, such as gallium arsenide (GaAs) and indium phosphate (InP), in the semiconductors. Although silicon looks opaque to visible light, the material is amazingly transparent to infrared light and silicon structures can be constructed that can guide and modify infrared laser light.
光通信面临的挑战是如下方面的成本问题：光学激光器、对激光束进行调制/解调所需的元件；让芯片元件与光纤实现配合的极其精密的机械设计。Luxtera宣称找到了一种成本经济性非常好的方法，以将激光与调制芯片实现很好的配合，并将硅片上的调制器与光纤实现配合。The challenges of optical communications links have been the cost of the optical lasers; the components required to modulate and demodulate the laser light; and the extreme-precision mechanical designs required to mate the chip components to the fibers. Luxtera claims to have found a very cost-effective way to mate the laser to a modulation chip and to mate the silicon-chip modulator to the fiber. 不幸的是，Luxtera的革新性技术的细节目前尚被封锁，因此目前该公司的许多声明尚无法得到验证。某些基本的研究最初来自于加州理工学院（Caltech），该校继续成为R&D伙伴，但大部分知识产权（IP）是由Luxtera开发的。该公司在三月后半月发动了第一波媒体攻势。Unfortunately, the details of Luxtera's innovations are currently under lock and key, so many of the company's claims cannot be verified at this time. Some of the basic research originally came from the California Institute of Technology (Caltech), which continues to serve as an R&D partner, but much of the intellectual property (IP) was created by Luxtera. The company launched its first press wave in late March.
使得Luxtera特别有吸引力的一点，就是该公司并没有仅仅依赖一个好的注意——它建立在多个好的主意上，这些思想综合起来就可以制造出完整的解决方案。这些概念都建立在长达7年的研究和开发工作的基础上，代表了芯的光纤连接技术和硅调制器设计。在建立上述的解决方案时，该公司形成了一个丰富的IP库，其中有约100项即将获得专利的成果。What makes Luxtera particularly interesting is that the company isn't based on just one good idea—it's based on a number of good ideas that are combined to allow complete solutions to be manufactured. The concepts are based on seven years of research and development and represent new fiber-coupling techniques and silicon-modulator designs. In building the solution, the company has access to a broad IP portfolio of around 100 patents pending.

图1。 Luxtera的测试半空Figure 1. Luxtera test card. Source: Luxtera
Intel也在探索如下的技术：对发射光的调制、硅结构中的光导能力、光纤到硅片的机械配合以及最近研发的第一种硅激光器件。Intel拥有一家建在Santa Clara的光电子实验室，MPR最近有幸参观了一番。虽然硅的特性并不适合于传统的激光产生，但Intel能够利用Raman效应在硅中制造出激光器。不过，到目前为止，Intel的努力还局限于实验室中的试制工作，该公司还没有公开声明它何时交付已经投入生产的解决方案。Intel is also exploring techniques to modulate optical transmissions, guiding light in silicon structures, mechanical mating of fiber to silicon, and, most recently, the first silicon laser. Intel has a photonics lab in Santa Clara that MPR recently had the chance to tour. While silicon's properties are not suitable for traditional laser generation, Intel was able to use the Raman effect to create a laser in silicon. So far, though, Intel's efforts are still lab experiments, and the company has not gone on record to say when it would ship production-level solutions.
铜互联走到了尽头？Is Copper Down to the Wire?
多年以来，设计界依赖于带有铜线层的印刷电路板来完成芯片间的互联，而且依靠底板和铜线缆完成机箱和网络间的互联。但是，整个电子生态系统日益增长的、对带宽的需求正不断推动铜互联架构的极限的扩展。不断变得更为复杂的信号调理的和精密的位和时钟编码方法已经使得铜线传输可能的速度扩展到5Gb及5Gb以上，但在较长的距离上10Gb的速度则达到了铜线传输的极限。For many years, the design community has relied on printed-circuit boards (PCB) with copper layers to provide chip-to-chip interconnect, and on backplanes and copper cabling to connect boxes and networks together. But increasing requirements for bandwidth all through the electronics ecosystem are pushing the limits of that copper infrastructure. Increasingly complex signal conditioning and sophisticated bit and clock encoding have made it possible to push data-over-copper speeds to 5 gigabits (Gb) and above, but getting to 10Gb over any significant distance means pushing copper to the limits.
不过，永远不要低估市场延长铜电缆、Ethernet和TCP/IP协议的寿命的能力。在1990年代早期，人们预测网络的未来是针对光纤分布式数据接口（FDDI）的（ANSI）X3T9.5标准。其目标是以一种更健壮、可管理性更好、性能更高的标准来取代铜线Ethernet。However, never underestimate the market's ability to extend the life of copper, Ethernet, and the TCP/IP protocol. In the early 1990s, the future of networking was supposed to be the (ANSI) X3T9.5 standard for fiber distributed data interface (FDDI). The goal was to replace copper Ethernet with a more robust, more manageable, higher-performing standard. 然而，FDDI却仅仅成为一种面向校园局域网的细分（niche）解决方案，而在双绞线上传输的10/100M Ethernet则投入了大规模应用。Ethernet后来被扩展到1Gb，最新的尝试则是找出一种成本经济性好的10G解决方案（IEEE 10Gbase－T），它可以实现至少55m的传输。现在尚不清楚这一最新的、再一次扩展铜线Ethernet容量的努力是否会成功。10Gbase－T的实现将需要一种优于常见的Cat5电缆的电缆。Instead, FDDI became only a niche solution for campus-area networks, and 10/100M Ethernet over twisted-pair wiring became the volume solution. Ethernet was later extended to 1Gb, and the latest attempts are to find a cost-effective 10G solution (IEEE 10Gbase-T) that can handle distances of at least 55 meters. It is not clear if this latest effort to extend Ethernet on copper one more time will be successful. The 10Gbase-T implementation may require a better cable than common Cat5 cable.
10G数据率级的芯片间解决方案很有可能来自于不同的IP提供者，但这些方案的大多数都需要大量的信号处理，走线长度方面存在极限，而且承受着功耗和电磁辐射的增加所带来的负担。Chip-to-chip solutions for 10G data rates will likely come from various IP providers, but most of these solutions will require significant signal processing plus limits to trace lengths and will carry burdens of increased power consumption and electromagnetic radiation.
铜线的主要优势在于，它已经存在（现有的电缆连接和FR4 PCB基础），而且它的运用十分方便。光纤线缆的分布渗透则比不上铜线的，除了远距应用和高EMI环境。光纤电缆的问题是使用时的成本和复杂性。光纤的成本不是一个长期性舞女体，因为光纤都采用了更为便宜的材料（玻璃vs.铜），而且由于可用于远程传输，单模光纤的批量现在相当大。光纤的昂贵之处就在于其连接器和激光驱动器。The main advantages of copper are that it's already there (existing cabling and existing FR4 PCB infrastructure), and it's easy to work with. Optical cables are far less ubiquitous than copper is, except for long-haul applications and high-EMI environments. The problems with optical cabling have been cost and complexity to work with. The cost problem of the fiber is not a long-term problem, as optical fibers are based on cheaper material (glass vs. copper), and the volume of single-mode fiber is now fairly high because of its use in very long haul applications. The expensive aspects of fiber are the connectors and laser drivers.
虽然LED 可以与多模光纤一起使用，而且价格并不昂贵，但大多数的应用则针对激光和单模光纤的使用，因为它们的传输距离更远。采用激光后，通信范围得到了扩展，通过使用多种波长来承载信息以提高带宽的技术也成为可能，该技术已经用于昂贵的远距离骨干网，被称为密集波分复用（DWDM）。采用DWDM后，带宽可以递增相加，一直累计到太位（1012bit）级带宽。While LEDs can be used with multimode fibers and are not very expensive, most of the research has gone into using lasers and single-mode fiber, because of its longer-distance transmission capabilities. With a laser, the range is extended, and it is also possible to increase bandwidth by using multiple wavelengths to carry information, a technique used in expensive long-haul carrier backbones—called dense wavelength division multiplexing (DWDM). With DWDM, bandwidth can be added in increments, up to terabits of aggregated bandwidth.
电子无法长时间阻碍光子的脚步
Electrons Can't Hold Back Photons for Long
2种基本的信息传输介质间爆发了战争：电子和光子。最终的结果很容易就可以预测出——光子最终降会赢得这场战争。光子的传输举例比电子远（在适当的介质中），所消耗的功率更少，而且发热更小。物理学的基本规律决定了，随着时间的推移，一旦某些障碍得以克服，光子学将取代电子学。当然，光电子的元件不会像电子元件那么便宜，或者不会那么容易制造。The battle is between two fundamental information transports: electrons and photons. The end result is easy to predict—photons will eventually win that battle. Photons can travel further than electrons (in the proper medium), use less power, and generate less heat. The fundamentals of physics dictate that photonics will replace electronics over time, once certain obstacles are overcome. Certainly, photonic components are not as inexpensive, or as readily manufacturable, as electronic components are.
这种介质的转换从远距离的电信电缆开始，然后发展到中等距离传输，形式是光纤到路边（fiber-to-the-curb）的电缆设施建设。在数据中心内部，光纤用于必须具备如下特性的场合：高带宽、抗干扰、远距离传输。一旦价格和制造问题可以让光电子的成本进一步削减的话，我们就会看到数据中心将添加更多的电缆——首先用于机架间的连接，然后用于电路板间的连接，最终会用于芯片间的连接。即使10G调制和光纤到硅芯片连接的成本得以降低，人们还需要制定光电子底板方面的标志并开发出印刷电路板上的光路由的大批量制造技术。The migration started with long-haul carrier cables, then moved to medium distances with fiber-to-the-curb cable installations. In data centers, fiber is used where high bandwidth, interference resistance, and long distance are required characteristics. Once the price and manufacturing issues reduce optical costs further, we should see additional fiber in the data center—first in rack-to-rack connections, then in board-to-board, and eventually in chip-to-chip. Even with reductions in the costs of 10G modulation and fiber connections to silicon die, standards will need to be developed for optical backplanes, as will high-volume manufacturing techniques for optical routing on PC boards.
来源于Luxtera技术的首批测试部件目前正由Freescale在其主流的130nm SOI硅工艺线上制造。第一个生产型的部件将于06年二季度开始提供样品，而且将采用标准的XFP 10Gb模块形式。该XFP模块是用于OC192/STM—64信道、10G Fibre 信道和10G Ethernet的标准的光电部件。（如要了解更多关于XFP 10G的信息，可以访问see http://www.xfpmsa.org/。）XFP模块还需要一个10Gb差分电气连接来实现串行的数据连接。The first test parts from Luxtera's technology are presently being manufactured by Freescale in its mainstream 130nm SOI silicon. The first production parts will sample in 2Q06 and will be available in a standard XFP 10Gb module. The XFP module is a standard optical component for OC192/STM-64 channels, 10G Fibre Channel, and 10G Ethernet. (For more information on the XFP 10G module, see http://www.xfpmsa.org/.) The XFP module still requires a 10Gb differential electrical connection for the serial data.
Luxtera的第一种产品可以让该公司顺势落入现有的基础架构，我们可以认定，它的报价将定在一个出奇的价位上，以证明Luxtera实现方案在成本上的优势。该公司在这些模块之外还有什么计划，尚不得而知，但Luxtera可以继续将该技术留给自己的硅基网络物理层（PHY）元件。如果该公司的最终目标是弥漫式的、渗透到系统芯片级的光学网络，就需要给予主流集成电路制造商相应的许可，使之能集成到高性能的微处理器和系统元件中。Luxtera's first product allows the company to drop into an existing infrastructure, and we would assume it will come at an aggressive price point to prove the cost advantages of the Luxtera implementation. The company's plans beyond these modules are unknown, but Luxtera could continue to reserve the technology for its own silicon for networking physical-layer (PHY) components. If the company's ultimate goal is ubiquitous optical networking down to the system-chip level, Luxtera will need to license the technology to mainstream silicon manufacturers for incorporation into high-performance microprocessors and system components.
对Luxtera而言，而且一般来说对整个光电子技术而言，未来确实十分光明。该公司已经向少数几家关键性的伙伴，包括Sun Microsystems，展示了其技术。我们将一直关注该公司的进展，而且我们希望将来能报道更多的细节。我们还预计多家公司将继续把铜技术扩展到至少10G的水平，但其传输距离会受限。For Luxtera, and for photonics in general, the future really does look bright. The company has shown its technology to a few key partners, including Sun Microsystems. We will be watching the progress of this company, and we hope we can report more details in the future. We also expect that companies will continue to push copper technology up to at least 10G, but distances will be limited.