"Within five years, we hope to connect microprocessors and memory chips right to the optochip, producing the optical analogue to the electrical multichip modules in today's big-iron machines."
At only 10 Gb/s, there are already buses available that do better than that. The good part is the small size I think. At another point they say 160 2 Gb/s links in a small package. That's the goodness.
The value comes from eliminating electromagnetic crosstalk and reducing the amount of PCB real estate dedicated to those buses. A single line of fiber can outperform several lines of copper, is made of cheaper material and sidesteps a whole swathe of engineering challenges (resonance, power dissipation, synchronization of signals across multiple lines etc.)
"sidesteps a whole swathe of engineering challenges"
Isn't it funny that as time marches on and certain technologies improve, that other areas become much simpler and cheaper to deal with. I can imagine that components such as this making it easier for more engineers and hobbyists getting into electronics, embedded systems, etc.. because of this.
As a kid my parents had one of those fiber optic swans you put water in and set atop a "light bulb"; this was in the early 80s or late 70s and I thought at the time (being a geek and familiar with RF) that this fiber optic technology is going to have long reaching ramifications in a good way...
> I can imagine that components such as this making it easier for more engineers and hobbyists getting into electronics, embedded systems, etc.. because of this.
Easier ??? If anything it will make it unbelievably much harder, just like VLSI chips are harder to interface to for hobby purposes than TTL and soldering SMD is an art, but 1/10th" pitch can be done by anybody that isn't legally blind.
As tech has steadily advanced over the years the entry level skills required have gone up dramatically.
I found it fascinating and shows the development from first concept through research phases to practical/commercial uses. It's a few years old now, but should be good for a lot of information anyway.
As you said, we already have buses that can do as much as even 32GB/s (2 lane PCIe 3). I think the bottleneck wrt main memory is the scalability of DRAM. Having lots of DRAM chips helps you scale _out_ just like graphics cards. To put in comparison, Geforce 8800 Ultra has an ~100 GB/s memory bandwidth available to 128 cores (shared).
Another big plus will be the huge reduction in latency, it can't get lower than that. I'm wondering at how much they will be able to sell it, from what i've seen working in this field common opto-electrical transceivers are not usually cheap (but they will need a less complex/very low power design, so this could not be a problem).
This is true. Though, I assure you, as someone who worked on the generation of optical interconnects before this one: The invention of the VCSEL really has changed the game, as has the invention of cheap planar arrays of microlenses. These are not the lasers of 1990.
http://www.anandtech.com/print/3834 http://techresearch.intel.com/articles/None/1813.htm