Intel's New Silicon Photonics Module For Data Centers Beams Info at 100Gbps Across 2km (zdnet.com) 31
An anonymous reader writes:Intel has announced it's launching silicon photonics, a product 16 years in the making, to enhance the use of optics for data center traffic management. It has a tremendous advantage over other silicon solutions, Intel executive vice president Diane Bryant said at the Intel Developer Forum in San Francisco. Intel is "the first to light up silicon," she said, integrating the laser light-emitting material onto silicon. It uses silicon lithography to align the laser with precision, she said, creating a cost advantage because it's automatically aligned. It has a performance advantage because of the precision of lithography. The module Intel is introducing delivers 100 gigabits per second. With network traffic in the data center doubling every 12 months, electrons running over copper cable just won't cut it, Bryant said. However, the cost of fiber optics is growing, positioning silicon photonics as the next solution.The company adds that the silicon photonics module can deliver the data across two kilometers.
At last, at last! (Score:2)
Finally, I can watch my streaming 16K 3-D reruns of the re-re-re-mastered original star trek series on my smart watch.
Maybe then, 'Spock's Brain' will finally be watchable...
The cost of fiber optics is growing? (Score:2)
And? Won't this still require fiber optic connections between equipment?
This looks to be a new type of transponder not a new type of cable.
Re:The cost of fiber optics is growing? (Score:5, Informative)
I had no idea either until I looked it up.
Silicon photonics is the study and application of photonic systems which use silicon as an optical medium. The silicon is usually patterned with sub-micrometre precision, into microphotonic components.[4] These operate in the infrared, most commonly at the 1.55 micrometre wavelength used by most fiber optic telecommunication systems. The silicon typically lies on top of a layer of silica in what (by analogy with a similar construction in microelectronics) is known as silicon on insulator (SOI).
Re: (Score:3, Funny)
A person with this much subtlety has no place on Slashdot.
Re: (Score:3)
"I had no idea either until I looked it up."
If you now tell us that you RTFA as well, we're sorry, but we have to let you go.
Re: (Score:2)
Won't this still require fiber optic connections between equipment?
Sure it will. So even if this thing was twice as amazing as the hype wants it to be, we'd still be at the mercy of the telecommunications companies to actually incorporate it into their networks, which they have zero incentive to do, which means that we probably won't see this thing in the wild for decades.
On the other hand, I suppose there's no reason they couldn't put this thing in some sort of freespace optical system. But all that would accomplish is to put us all at the mercy of Big Mountaintop inste
Re: (Score:2)
MCI made major inroads into AT&T's business, building line of sight microwave towers all over the place - lasted for about 20 years before the tech was obsoleted, by fiber I think.
Problem with data rates like this LOS, is the loss from raindrops, birds, etc. If the signal spreads and multipaths too much, the data rate will have to drop.
Re: (Score:2)
MCI = Microwave Communications Inc. The first radio system was between Downers Grove, IL and Olivette, MO. Used to carry traffic for over the road trucking. I decomed the system myself in 2000. It carried 7 DS3s.
Re: (Score:3)
This isn't a telco play (2km doesn't get you anywhere... Plus who needs 100G to the home today - I would settle for 1G) - it is a data center play. 2km is a data center distance, large clusters are easily hosted in rooms of this size - and even inter rack distances can get into the 10's of meters. Frankly others have had 100G silicon photonics out for a while (read about a year).
Re:The cost of fiber optics is growing? (Score:4, Informative)
And? Won't this still require fiber optic connections between equipment?
This looks to be a new type of transponder not a new type of cable.
Fibre optic cables usually hook up to a fibre optic transceiver that has fancy lasers on it made on a different process to digital silicon chips, that then has to hook up to a comms chip, thus adding cost.
The silicon photonics just announced involves building a fibre optic transceiver on a silicon chip process so it can be integrated with the associated digital logic, eliminating unnecessary hardware and more closely coupling the data handling hardware with the fibre link.
Re: (Score:2)
So it's essentially a replacement for the HBA, the SFP, or both?
Re: (Score:2)
So it's essentially a replacement for the HBA, the SFP, or both?
I don't know what the specific products are. I haven't looked. But the technology certainly could replace just the SFP or both, depending on how you partitioned the hardware network stack. FWIW, the picture on the low-information web page linked in TFS looks like an SFP to me.
Re: (Score:2)
Nobody has used FDDI for a long time now. Well at least not in a new deployment.
Super computers also don't necessarily need fibre optics either. Mellanox will happily sell you a 3m EDR passive copper cable and a 5m one for 100GbE. We used them where ever possible because they are cheaper and use less power that the active fibre optic ones.
The main advantage in theory of silicon photonics is that it should be in theory cheaper because it uses bog standard silicon wafers and not the specialist and hence more
Re: (Score:2)
And? Won't this still require fiber optic connections between equipment? This looks to be a new type of transponder not a new type of cable.
Fibre optic cables usually hook up to a fibre optic transceiver that has fancy lasers on it made on a different process to digital silicon chips, that then has to hook up to a comms chip, thus adding cost.
The silicon photonics just announced involves building a fibre optic transceiver on a silicon chip process so it can be integrated with the associated digital logic, eliminating unnecessary hardware and more closely coupling the data handling hardware with the fibre link.
Yep. NOW you will see the true advantage of LightPeak/Thunderbolt.
Re: (Score:2)
The problem with the vision is that there are really good reasons to use separate transceivers. Is that a SX, LX or EX transceiver you have there? Or perhaps it is a BX one because you are sort of fibres and this allows you to effectively double the number without laying extra fibre. Perhaps I am really short of fibre and decide to replace all my SFP's with CWDM matched pairs and employ a passive multiplexer.
So unless silicon photonics is going to offer at least LX transceivers for SX prices it is not going
Re: (Score:2)
The problem with the vision is that there are really good reasons to use separate transceivers. Is that a SX, LX or EX transceiver you have there? Or perhaps it is a BX one because you are sort of fibres and this allows you to effectively double the number without laying extra fibre. Perhaps I am really short of fibre and decide to replace all my SFP's with CWDM matched pairs and employ a passive multiplexer.
So unless silicon photonics is going to offer at least LX transceivers for SX prices it is not going to have the impact that Intel are hoping for.
What makes you think it's all about data centers?
Is the cost of fiber optics really growing? (Score:4, Interesting)
Inflation doesn't mean growing cost. 20% inflation and an object going from $100 to $115 means the cost is shrinking.
There's all kinds of ways to make weird economic statements. You can ignore inflation and just point to raw dollars (i.e. $5 in 2016 is expensive; the thing cost $1 in 1938!). You can claim the cost of X is growing because the utilization is growing--we're running 50,000 meters of fiber optics instead of 500, so the cost of fiber optics now represents a bigger chunk of the data center's budget (100x more stuff, even if it costs 1/10 as much, is still 10x the cost).
Technology shifts how we do things. New technology means making a thing is cheaper; scarcity means your technology doesn't scale that high (more labor per unit to make that much that fast); and the deployment of a new technology to replace an old technology increases its market share, raising that technology's total marginal cost. I find it surprising that fiber optics would increase in cost per linear foot, considering fiber optics are relatively easy to produce in mass quantity and shouldn't have any scarcity issues.
Re: (Score:2)
To lay backbone fiber in the ground cost $500 per meter. Including materials, labour and right of way in the US.
Re: (Score:2)
Yes, but I don't think that you are going to lay fibre in the ground within a data centre which this technology is aimed for. Or else you are doing your data centres wrong.
Re: (Score:2)
How does that compare to laying that same fiber in 1998?
It is about integration (Score:3)
The advantage of silicon photonics is to integrate the optical elements (lasers & PIN diodes) into the silicion drivers and amplifiers, theoretically reducing cost.
There are already 100 Gbps CWDM4 QSFP28 (4 wavelengths of 25 Gbps on 2 fibers) and 100 Gbps PSM4 QSFP28 (single 25 Gbps wavelength on 8 parallel fibers) transceivers out there [cisco.com], but they need discrete lasers & PIN diodes in InP or GaAs, not silicon.
So we will see how Intel's silicon photonics 100 Gbps CWDM4 QSFP28 [intel.com] and 100 Gbps PSM4 QSFP28 [intel.com] transceivers end up being priced.
My impression is that 50 Gbps wavelengths are coming soon (using 4-level PAM), so two of those will be 100 Gbps. But the holy grail is the one wavelength 100 Gbps [finisar.com], likely some kind of high-order modulation (HOM). AppliedMicro has demonstrated a 100 Gbps single-wavelength PAM4 [apm.com], but no word on distance.