Future fiber systems in computer communications applications must meet growing bandwidth requirements, while
maintaining feasible power and cost targets in addition to maintaining manageable volumes of fiber cabling. Therefore,
bandwidth-per-fiber represents a critical design metric for next-generation systems. Here, a multicore fiber technology
based on multimode graded-index cores is reviewed. A full link demonstration using six cores transmitting up to 20 Gb/s
each is achieved between custom transmitter and receiver assemblies, which interface directly to the multicore fiber. The
demonstrated technology may provide the added bandwidth per link required in next generation HPC systems.
Scaling computing systems to Exaflops (1018 floating point operations per second) will require tremendous increases in
communications bandwidth but with greatly reduced power consumption per communicated bit as compared to today's
petaflop machines. Reaching the required performance in both density and power consumption will be extremely
challenging. Electrical and optical interconnect technologies that may be part of the solution are summarized, including
advanced electrical printed circuit boards, VCSEL-array based optical interconnects over multimode fibers or
waveguides, and singlemode silicon photonics. The use of optical interconnects will play an ever-larger role in
intrasystem communications. Although optics is used today primarily between racks, it will gradually migrate into
backplanes, circuit cards, and eventually even on-chip.
Keywords: optical interconnects, supercomputers, exascale,
Aggregate chip bandwidths in server and high performance computing have exceeding Tb/s, and if present trends are to
continue would lead to doubling the number of signal pins in each generation. For high bandwidth switch and server
applications, bandwidth requirements could exceed the package pin limit as early as 2012.
We defined metrics to compare the performance of electrical and optical interconnects, which includes bandwidth
density (Gb/s/mm2/port), media bandwidth*distance product (GHz*m), power consumption (mW/Gb/s/Port), and
technology comparison metric (Gb/s/mm2/port * GHz*m/mW/Port). We will show that optical interconnects offer a
performance metric improvement factor of greater than 25 over electrical interconnects.
The IBM Terabus program has developed parallel optical interconnects for terabit/sec-class chip-to-chip
communications through printed circuit boards with integrated optical waveguides. 16 TX + 16 RX channel transceiver
"Optomodules" were assembled and fully characterized, with fiber-coupled full links operating up to 15 Gb/s, for an
aggregate bi-directional data transfer rate of 240 Gb/s. Furthermore, we have demonstrated a complete link between two
Optomodules through polymer waveguides on a printed circuit board, with all 32 uni-directional links operating error-free
at 10Gb/s, for a 160 Gb/s bidirectional aggregate data rate. This is the fastest, widest, and most integrated
multimode optical bus ever demonstrated.
Traditionally, large core (greater than 100 micron) step index multimode optical fiber has occupied a reactively small niche of applications in data communications. While the large diameter of this type of fiber makes it easy to align to optoelectronic devices, its bandwidth*distance (BW*D) product is low due to modal dispersion between the large number of modes supported by a fully filled fiber. Recently, interest has been renewed in using an underfilled launch to excite 62.5 micron core graded index multimode fiber as a way to improve its bandwidth performance. With the proper launch conditions, this same effect has been measured in large core fiber. A vertical cavity surface emitting laser (VCSEL) is used to provide a low numerical aperture launch into a large core fiber, which has a relatively large numerical aperture. The laser thus underfills the modes of the fiber, and a bandwidth enhancement for the fiber is obtained. Results of experiments performed on step and graded index large core multimode fibers using a direct VCSEL launch are presented. In addition, these relaxed alignment tolerance fibers allow the utilization of very low cost cabling and connectorization procedures for parallel optical fiber cables. Data, including skew, bandwidth, and insertion loss, are presented on these cables.
Detailed analysis has been performed of modal noise reduction by superposition of high frequency modulation. Data obtained revealed that, for both direct and external modulation, the modal noise reduction factor depends on the modulation frequency and fiber length. The fiber mode coupling, chromatic dispersion, and the mode partition noise have been neglected in the analysis. Experimental results confirmed that, for multimode fiber, the modal noise reduction depends almost only on the modulation depth.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.