Optical Interconnects


Parallel optical interfaces can be conceived that consist of arrays of optoelectronic devices of the order of one thousand optical channels each 'running at speeds around I Gbit/s and hence offering and overall capacity of 1 Gbit/s to a single integrated circuit. Although there are still unresolved difficulties in the areas of architectural design, manufacturing processes, simulation and packaging (as explained later), the technology has now developed to the point that it is possible to contemplate its use in commercial systems within a time-frame of 5-10 years. Fig 1st shows the concept of chip-to-chip communication using optics.

The idea of using optical techniques to address the chip-to-chip interconnection problems has been around for a long time. However, it is only in the last few years that technology with a realistic promise of eventual commercial applications has emerged. Progress can be attributed to a shift away from trying to develop custom VSLI techniques with in-built optoelectronic capability, towards developing techniques to allow parallel arrays of separately fabricated optoelectronic devices to be tightly integrated with standard foundry VLSI electronics, e.g. CMOS

2. Motivation

The optics can reduce the energy for irreversible communication at logic level signals inside digital processing machines. This is because quantum detectors, quantum sources can perform and effective impedance transformation that matches the high impedance of small devices to the low impedances encountered in the electromagnetic propagation. This energy argument suggests that all except the shortest intrachip communication should be optical.

We see that there is a limit to the total number of bits per second, of information that can flow in simple digital electrical interconnection set by the aspect ratio at the interconnection. This limit is largely independent of the details of the design at the electrical lines. As the limit is scale-invariant, neither growing nor shrinking the system substantially changes the limit. Exceeding the limits will require additional multi-level modulation. Such a limit will become a problem for a high band-width machine. Optical interconnect can solve this problem since they avoid the resistive losses that gives the limit.

3. Capability and Limitations of Electrical Interconnects

Now the physical origins of the limitation of conventional electrical interconnects are listed

3.1 Frequency Dependent loss: - The main physical limitation on the use of electrical signaling over long distances in frequency dependant loss due to the skin effect and dielectric absorption. Attenuation due to the skin effect increases in proportion to ?f above a certain critical frequency. This given rise to a so called 'aspect-ratio' limit on the

The constant of proportionality Bo is related to resistivity of copper interconnects and is only dependant on the particular fabrication technology. It ranges from a 1015 bit per second to 1016 bit per second

The aspect-ratio limit is scale invariant and applies equally to band to band interconnect as well as to connection on a Multichip-Module. Also, for a fixed cross-section, the limit is independent of whether the interconnect is made up of many slow wires or a few fast wires. The aspect ratio limit is part of the reason why fiber-optics has replaced co-axial cables in telecommunication networks.

Attention due to dielectric absorption increases in proportion to frequency leadint and upper limit on operating speed, which is inversely proportional to distance.

It is independent of conductor Cross-section and is not scale-invariant. For a 1 Gbit/s interconnect, it would limit the distance to 1 m in a standard fiber-glass interconnect and may be to 10m in a good low-loss material like poly tetra fluoroethene (PTFE).
However, attenuation due to dielectric absorption does not limit the overall band width of an interconnect over a certain distance in the same way as the skin effect, because a higher overall bandwidth could be obtained by using more conductors within the same cross-section.