INTEL IS TALKING about Silicon Photonics again, the real advances are masked behind a breathless press release about lasers replacing electronics. The advances announced today are nothing nearly as spectacular nor as breathless, it is simply a speed bump and a connector.
The last time Intel talked lasers, they had the technology working with parts transferring data at 40Gbps. Today, there are two big jumps, an aggregate 50Gbps and a connector. While neither sounds spectacular, together they will allow some fast chip to chip transmission in the mid-term future.
Intel’s four wavelengths
Instead of one wavelength doing 50Gbps, Intel splits the laster into four wavelengths, and modulates each one separately. Each frequency carries 12.5Gbps for an aggregate 50Gbps. These four wavelengths are then combined, sent over a single fiber cable, and then split back into four channels on the receiver.
In the future, you could move to dozens of wavelengths and simply scale the bandwidth with the number of optical grates you use. Intel is keen to point out that 40 wavelengths would get you to 100Gbps, or enough bandwidth to bring the RIAA down on Intel on general principle.
All of this splitting is nothing new, it has been done for optical communication for years now. What is new is doing it on silicon, and doing it in a cost effective way. That is the novel thing that Intel brought to the table this time.
Plug and play mark XII or so
The other interesting part is the connector. Intel is showing off a pluggable optical connector, the last missing piece before this can be product-ized. If the connector is idiot proof, humanity will evolve a better class of idiots. Until then, the connector should enable a workable solution for a few years.
Once humanity evolves, the next step, one that Intel is undoubtedly at work on now, is to stack a laser on to a CPU to multiply I/O bandwidth by orders of magnitude. This will change how motherboards are laid out, add noise resistance, and drop pin counts. Together, these changes will make for a faster and more reliable system and related interconnects. In the end, that is the point.S|A
Latest posts by Charlie Demerjian (see all)
- HyperX ships it’s 60 millionth enthusiast memory module - Oct 15, 2018
- Bittware/Nallatech water cools 300W of Xilinx FPGA - Oct 12, 2018
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018