Electrical versus optical - history repeats itself

Changes in data communication on board level

In one of my earlier blogs, I wrote about how the future of computer power migration can be predicted based on history. The computing power of a supercomputer today is available on your laptop 12 years later, and waiting another three years allows you to have this computing power on your smart phone. Imagine what has to happen on a product level to make this possible! Let’s take a look into the future.

In September, I went to ECOC (European Conference on Optical Communication) in Cannes, France. Several new technologies were shown, like the Wide Band Multi Mode Fibre (WBMMF), where four multimode (MM) signals can transmit through a MM fibre at 25Gb/s using multiplexing technology. Despite this great development and its importance for 400Gb/s communication, it is still a small incremental step compared to what we need to get to 1 Exaflop (1 ExaFlop = 1018 FLoating-point Operations Per Second) computing power in a smart phone. Some are talking about new optical communication channel materials instead of glass, but no major announcements have been made.

If you grew up in an electrical world like I did, and are old enough, you should know about wire wrapping technology. Copper insulated wires were stripped on both ends and wrapped around a square pin with a special tool. These pins were mounted on a carrier and connected to the discrete components. I don’t know about you, but from my experience, this meant a large spaghetti of copper wires on a board! Replacing a wire or changing its path was a nightmare to get done. Eventually, the Printed Circuit Board (PCB) was developed, on which copper tracks were created using UV-lighting and copper etching. This was a major improvement and is still used extensively today.

I believe the same will happen in the optical domain. If you look at some of the PCB’s containing both electrical and optical components, you can see a large spaghetti of optical fibres on top. The optical fibres cannot connect as easily as the copper wires did in the past around a square pin – yet! A serious over-length of fibre is required because of fusion splicing or connector termination. However, all of this excess fibre has to be stored on top of the PCB. Some are using fibre circuits where fibres are mounted on a carrier, but an excess length of fibre is still required because of its connectors. So history is repeating itself…

Polymer waveguides embedded in a carrier, like a PCB, seem to be the solution. It’s like the transition from wire wrapping to the electrical PCB. However, quite a lot of development has to be done to get there. The intrinsic optical attenuation due to the materials is going down and reaching a level of 0.03dB/cm, but this is still very large compared to a fibre; the attenuation is about a 1000 times higher than of a multimode glass fibre. It doesn’t have to be as low as a fibre because of the limited length requirements (on a PCB), but still, it means that we should be driving the electronics harder. This does, however, result in power dissipation, which is exactly what we don’t like in a data centre. Nevertheless, it’s an interesting technology which is finding its usage already in the sensor market, and will find its way into our data centres, most probably in combination with integrated Silicon / III-V photonics.

If you would like to learn more about fibre innovation, please visit: http://www.te.com/tefiberoptics/en/home.html

Next time I would like to write about structured cabling versus direct attach type of cabling. Hope to ‘meet’ you again!