Intel's Silicon Optics Announcement: Big News on Two Levels
Intel has announced that its R&D group has developed an all-silicon optical modulator that can run at gigabit speeds. Though this is the first time the story has hit the big media, the team that's made this breakthrough disclosed the general direction to Intel's developer partners nearly two years ago (PDF file).
(In what follows, I'm making an assumption: This is not FUD. To be specific, that the Intel design will be manufacturable, in volume and with good yield, in a conventional chip fab production line, within a couple of years. I won't know if that's true for sure until I have someone smarter than me look at the Nature article, and do a little gossiping. You have been warned.)
The first level where this is big news is in device interconnection. Here's the key quote from Markoff's NYT article:
"The lines running in and out of chips are becoming the bottleneck,'' Dr. Miller said. "As we start scaling up the information capacity of broadband networks, we run into a basic problem with electrical wiring. In the future, we will be able to use optics at all levels of electronic systems.''
Find a silicon flip chip and look at the back. Note all the connection pads that must be precision soldered onto the carrier printed circuit board. Every one of them carries a direct manufacturing cost, an indirect cost in reduced yields, and potentially spews interference into the rest of the system. If Intel can further increase the speeds of their new silicon optics, they can eventually eliminate many of those connections, and instead transfer the data optically, using just a few data paths that do not generate electrical interference. That can be big news for the architecture of pretty much any device built with high speed silicon.
The news at the second level might be even bigger: Intel has served notice that the silicon juggernaut is still very much alive, and will now start extracting value from the optics markets directly. A lot of us have been wondering where the margins to fund succeeding generation of Moore's Law would come from, as it gets harder and harder to extract premium prices for every faster MPUs. Here (along with mobility) is a portion of Intel's answer.
Which has a knock-on effect into another area you might have heard of: nanotechnology. Everyone in my business has, over the last few years, seen multiple plans to use things like quantum dots and other fine-structured materials to modulate and switch light, and create fully integrated optical circuits. Some of these have been funded. Many have languished because their ultimate consumers, network operators, haven't been buying. This partial hiatus, born of the telecom bubble implosion, may created a window of opportunity for silicon.
Silicon may not be the optimal material to implement optical switching and integration, but it might just be good enough to meet some market demands as they evolve. This would engage the silicon learning curve in direct competition with other approaches, and potentially forestall the creation of competing industrial bases in the optics sector. Intel's announcement indicates they will try it. Maybe that's part of what the NNI committee knew when they decided on funding priorities?