Page 364 - System on Package_ Miniaturization of the Entire System

336 Cha pte r S i x

consumer market as digital home appliance interfaces, home networks, and automotive
networks where reliability requirements are less demanding.
The most comprehensive data comes from Corning on fluorinated acrylates [39].
Exposure to air at 100°C for 270 days shows no measurable change in the index of
refraction. Thermal aging at 170°C for 12.6 days decreases propagation losses, and optical
2
aging under 130 mW in a waveguide with a 7 μm × 7 μm cross section (260 kW/cm ) at
1319 nm for 117 days shows no changes.

6.5 Evolution of Optoelectronic SOP Technology
Because of increasing bandwidth requirements and the limitations of copper
interconnects, optoelectronic data communication has migrated from fiber-based long-
distance communications in the 1980s to local area optical networks in the late 1990s
where it has abruptly stopped. The technical reasons for the abrupt stoppage are fairly
easy to understand. The replacement of copper cables outside the box requires relatively
few optical channels, and until now one could borrow from long-haul and LAN optical
transceiver technology, and with minor modifications adapt that technology to “box-to-
box” data sharing in a high-performance computing environment. Standard optical
transceiver packs, though bulky and having a bandwidth of only 2.5 Gb/s per channel
over 12 channels, can be used over several hundred meters of parallel plastic or glass
optical fiber ribbons [40]. An optical transmitter module obtains power and signals
from a transmitter board, converts the signals to optical coding and transmits them
optically. The signals are then received by a receiver module where they are converted
back to digital signals. Optical transceivers have become miniaturized for this application
[26] but still offer very low interconnect density. This is where further progress toward
“optics to the processor” has slowed; the pool of ready-made solutions has been
exhausted. At this point, high-density optical interconnect technology to the processor
has become the subject of research and development in many laboratories around the
world [15]. The next evolutionary step is to extend optical signaling between cards as in
blade center servers, then between processors on a card, and eventually integrated
optics on the processor chip itself, as depicted in Figure 6.1.

6.5.1 Board-to-Board Optical Wiring
The ideal optical backplane or midplane consists of optodigital cards that plug in to optical
and electrical ports as is done with electrical cards in Nuclear Instrumentation Module
(NIM) bins used in photon counting and nuclear counting, a concept that has migrated to
rack-mounted computers such as blade servers. The concept of plug-in optical cards has
been previously attempted by General Electric [41–43] and a consortium of universities
and industry companies and organizations who leveraged MCM technology to package
VCSEL arrays and use MT connectors for optical coupling. Reliable optical plug-in cards
are difficult to achieve because of the difficulty of achieving and maintaining reliable
optical alignment in a mechanically and thermally harsh environment. In addition, the
optical channel density is roughly commensurate with electrical pin density. Consequently,
the plug-in optical backplane remains an unsolved technical challenge in reliability and
edge density. An intermediate solution to high-speed interboard optical interconnects that
seems to be gaining momentum for enhancing the performance of high-end machines is
the miniaturized optical transceiver module that uses either optical fiber array ribbons or
optical polymer WG array ribbons, as shown in Figures 6.5 and 6.6.