Browse Prior Art Database

Free-space optical backplane

IP.com Disclosure Number: IPCOM000016017D
Original Publication Date: 2002-Aug-28
Included in the Prior Art Database: 2003-Jun-21
Document File: 8 page(s) / 137K

Publishing Venue

IBM

Abstract

Signal speeds within data processing systems (such as switch/routers, servers, high-end computers) are approaching 10 Gb/s and more. At the same time, system size (in particular for Internet switches) is expected to increase, because the speed of the processing electronics (network processors) isn't increasing as fast as fiber bandwidth (driven by the introduction of wavelength division multiplexing, or WDM). This means that more electronics working in parallel is needed to process the incoming datastreams. As a consequence, future switches are expected to be multi-rack systems instead of single-shelf systems. Obviously, this increase in system size is also reflected in the length of the intra-system interconnects, which will be increased to tens of meters. At these speeds and lengths, conventional copper interconnects are reaching fundamental bandwith limits. It is generally believed that the use of optical interconnects is a valuable approach to overcome these limitations. While the longest interconnects (rack-to-rack) will be the first that have to "go optical", it can be expected, that optics will also be required within a single shelf or box. Apart from the "bandwidth-length-product", connector density (e.g. at the switch card edge) is an important bottleneck that might call for a trasition to optical interconnect technology. At which time the crossover-point will be reached, will be determined by many factors, such as interconnect length, bitrate, cost, power, connector density, customer migration paths, etc. The question here is not so much "if", but rather "when, where and how" optical interconnects within systems will be needed. There are already various approaches to implement optical interconnects from one board to another board within the same box/shelf/enclosure, such as individual fibers, fiber laminates, imaging fiber bundles, embedded waveguides or free-space optical interconnects. Common problems of all approaches are the interfacing and the integration of optics with electronics. Ideally, the interfacing (alignment, connectorization, etc) should be cheap, cause little losses and consume little space. The same is true for the whole optical subassembly, which in addition should integrate well into existing systems (e.g. 19" racks with plug-in cards that have rather sloppy mutual alignment) and into established electronic board fabrication and packaging processes. Looking at the above mentioned approaches, it can be seen that they all have some shortcomings: Individual fibers for example have excellent transmission, but they don' thave the desired degree of integration with the electronics, and their connectorization is expensive and bulky. Similar statements can be made for fiber laminates and imaging fiber bundles. Currently known embedded waveguide prototypes mostly suffer from too high attenuation (a typical value of 0.1 dB/cm causes 10 dB attenuation over the typically required 1 m interconnect length) or a too low melting point (which is a problem for board packaging technologies such as solder reflow). Moreover, there is usually a compromise between overall size and mass producibility (very larger structures have to be written serially). Many of the known approaches for free-space optical interconnects base on line-of-sight concepts, and they typically suffer from high cost, bulk system size, stringent alignment tolerances and poor integration with the rest of the system. 1

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 20% of the total text.

Page 1 of 8

Free-space optical backplane

   Signal speeds within data processing systems (such as switch/routers, servers, high-end computers) are approaching 10 Gb/s and more. At the same time, system size (in particular for Internet switches) is expected to increase, because the speed of the processing electronics (network processors) isn't increasing as fast as fiber bandwidth (driven by the introduction of wavelength division multiplexing, or WDM). This means that more electronics working in parallel is needed to process the incoming datastreams. As a consequence, future switches are expected to be multi-rack systems instead of single-shelf systems. Obviously, this increase in system size is also reflected in the length of the intra-system interconnects, which will be increased to tens of meters.

At these speeds and lengths, conventional copper interconnects are reaching fundamental bandwith limits. It is generally believed that the use of optical interconnects is a valuable approach to overcome these limitations. While the longest interconnects (rack-to-rack) will be the first that have to "go optical", it can be expected, that optics will also be required within a single shelf or box. Apart from the "bandwidth-length-product", connector density (e.g. at the switch card edge) is an important bottleneck that might call for a trasition to optical interconnect technology. At which time the crossover-point will be reached, will be determined by many factors, such as interconnect length, bitrate, cost, power, connector density, customer migration paths, etc. The question here is not so much "if", but rather "when, where and how" optical interconnects within systems will be needed.

There are already various approaches to implement optical interconnects from one board to another board within the same box/shelf/enclosure, such as individual fibers, fiber laminates, imaging fiber bundles, embedded waveguides or free-space optical interconnects. Common problems of all approaches are the interfacing and the integration of optics with electronics. Ideally, the interfacing (alignment, connectorization, etc) should be cheap, cause little losses and consume little space. The same is true for the whole optical subassembly, which in addition should integrate well into existing systems (e.g. 19" racks with plug-in cards that have rather sloppy mutual alignment) and into established electronic board fabrication and packaging processes. Looking at the above mentioned approaches, it can be seen that they all have some shortcomings: Individual fibers for example have excellent transmission, but they don' thave the desired degree of integration with the electronics, and their connectorization is expensive and bulky. Similar statements can be made for fiber laminates and imaging fiber bundles. Currently known embedded waveguide prototypes mostly suffer from too high attenuation (a typical value of 0.1 dB/cm causes 10 dB attenuation over the typically required 1 m...