Intel's Light Peak Will Replace Copper Wires

Enlarge

Light Peak module with 4 fibers. Each fiber capable of carrying data at 10 gigabits per second. Credit: Intel

(PhysOrg.com) -- At the Intel Developer Forum in San Francisco Wednesday, the company announced a new optical cable that will be able to transfer data, between electrical devices, starting at speeds of 10 gigabits per second.

Vice president, Dadi Perlmutter, of Intel's Mobility Group, hopes to ship an optical cable, called Light Peak, by 2010. Light Peak will first be introduced into the market as being able to transfer data at 10 gigabits per second. Future versions will be able to transfer data at 40 and 100 gigabits per second as the manufacturing process becomes cheaper.

A single Light Peak cable will be capable of transporting multiple types of data simultaneously such as transferring data to a hard drive, connecting to the internet and transferring video.

Each end of the Light Peak cable will be connected to chips that contain light producing devices, encode data, and transmit data. The chips will also amplify data and convert the light to electrical signals.

Researchers are hopeful that silicon photonics will eventually replace copper wires on motherboards and microprocessors by making high-bandwidth connectors cheaper.

The first generation of Light Peak cables will use the same type of optical chips used in telecommunication devices today. Intel will be able to drive down the cost of these chips because the manufacturing standards are less stringent.

The lasers and detectors inside the chips are not required to be high performing. The chips don't need to transmit data over great distances as required in the telecommunication industry.

Intel is currently working with other companies to form partnerships. Sony is supportive of Intel's Light Peak technology, with more announcements coming.

Additional information can be found at Intel's Tech Research

Via: Technology Review

© 2009 PhysOrg.com

   

• h20dr - Sep 24, 2009
  • Rank: not rated yet
  This is not new technology; just an application on a smaller scale but it will have huge implications for personal communication devices as miniaturization of electronics continues to scale down towards the nano level.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• El_Nose - Sep 24, 2009
  • Rank: 4 / 5 (1)
  May not be new - per se but this is something intel has been researching for like 7 years. This FINALLY coming to a mobo near you means removing a rather insignificant bottle neck on PC's but a larger one on the huge corp servers. If these can be used in conjunction with gigabitwifi and sharing of memory across boxes and cpu's may be a great boon to parallel computing
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• E_L_Earnhardt - Sep 24, 2009
  • Rank: 5 / 5 (1)
  Current power transfer is by electron. Insert one into the orbit of an atom and one drops out of the orbit of another miles away. Photons may be faster,
  but "fade" is a long-distance limiting problem. We lose a lot of light between here and the sun!
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Bob_Kob - Sep 25, 2009
  • Rank: 5 / 5 (1)
  Does a harddrive even have the capacity to read or write at 10 gigabits per second?
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Roj - Sep 25, 2009
  • Rank: 5 / 5 (1)

  capable of transporting multiple types of data.. such as transferring data to a hard drive..

  
  
  If these devices are still booting from hard drives, rather than solid state memory, its a Marketing Gimmick.
  
  Unless of course there's a window carved out of the tower so we can watch the blinking lights.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• guiding_light - Sep 25, 2009
  • Rank: 5 / 5 (1)
  I have a couple doubts about this otherwise hopeful technology.
  
  First is the power consumption of the light sources (they can't be that efficient). Second 3D interconnects make more sense (less distance to travel, less power) and save space.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• plasma_guy - Sep 25, 2009
  • Rank: 5 / 5 (1)
  It looks like a step in the wrong direction, being more wiring, albeit different type. I don't see this working well among motherboards for personal devices, maybe server arrays.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• lengould100 - Sep 25, 2009
  • Rank: 3 / 5 (2)
  This is cute, but not going to change anything for personal users. short-distance links are alredy faster than anyone uses, its those plodding unreliable "last mile" links which need fixing.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Anyone - Sep 25, 2009
  • Rank: 1 / 5 (2)
  Re: "fade", E_L_Earnhardt -- not at all the same kind of thing; not at all the same distance; not the same wavelength; the only common factor between the two is photons. try Google, polymer optic fiber.
  
  Re: hard drive I/O, Bob Kob -- not to my knowledge, for magneto-mechanical types. as was mentioned, probably use SSDs.
  
  Re: power consumption, 'guiding light' -- yes, they can, especially at short distance. try Google -- VCSELs, also LEDs. Intel likely will use VCSELs.
  
  Re: 'wrong direction', 'plasma guy' -- no technical reason to doubt it, if Intel can reduce manufacturing and assembly cost. One big issue is alignment of emitter with fiber. One way to mitigate that is with large core polymer fiber.
  
  Re: 'last mile link', lengould100 -- agreed, there must be sufficient demand, sufficiently large end-use need to justify this kind of upgrade. it's not clear to me what that demand or end-use will be, aside from switch rooms and back-office for now.
  
  p.s. -- I don't work for Intel
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• guiding_light - Sep 25, 2009
  • Rank: 5 / 5 (1)
  Anyone, if you're trying to take the signal over a long distance (like meters) optical is indeed compelling, but I cannot agree for chip-scale you want to add an electrical-to-optical conversion unit. Not if I can use an equally wide (multi-micron scale) copper via.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• dhughes - Sep 26, 2009
  • Rank: 5 / 5 (1)

  Does a harddrive even have the capacity to read or write at 10 gigabits per second?

  
  
  10 Gigabits (bits) per second is 1.25 Gigabytes (bytes) per second, not exactly beyond the capability of a modern (but not exactly commonplace!) RAID array of Intel X25-E SSDs.
  
  There have been hardware websites who have shown a 16 drive RAID of Intel X25-E SSDs at 3GB (bytes) / second and a PCI card SSD RAID at 6 GB / second.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Husky - Sep 26, 2009
  • Rank: not rated yet
  How about linking 32GPU's spread over several pc's with your optical cable in a virtual crossfire setup, finally getting a decent framerate in farcry2 or realtime raytracing in ultra HD ?
  
  Or,
  
  A huge Raid on wich you store a 1000 virtual machines and a pagefile to swap them in and out of your memory for a few milliseconds, so it looks as if you got them running local all at once on just a server with 1 board and 4 gig mem, wich shouldn't normally be possible. This could shift the cost from buying more rackservers AND userspace/ storage to just 1 or 2 fast servers and userspace
  
  
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Bob_Kob - Sep 26, 2009
  • Rank: not rated yet
  I thought that this would have been common practice on servers and the like, and that this tech was specifically for user end applications now.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Arikin - Sep 28, 2009
  • Rank: 5 / 5 (2)
  El_Nose, fiber-optic connections between servers is being done now. Right now they use it between the servers and say a disk array. Sun is doing it and some others.
  
  This is a nice idea. But they moved the bottle neck from the motherboard to the devices again. Until the encoding and decoding is removed and the light is used by the devices directly then your faster speed isn't going to be realized.
  
  There are already boards out there that can match or beat the devices input/output speeds. So no real advantage with this setup.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Anyone - Sep 28, 2009
  • Rank: 5 / 5 (1)
  Re: distance, 'guiding light' -- VIAs, and the like, are part of the problem with copper interconnect at chip-scale. people are talking about 40 - 100Gbps now (802.3ba); no clear indication when there will be a market. people need to decide when the cost of fine-tuning copper interconnect and use of exotic dielectric material will meet or exceed the cost of optics, then choose.
  
  Re: bottleneck, Arikin -- correct. I bet it's legacy issues with the (now) older copper interconnect (e.g. - baseline wander and jitter). perhaps this would be an avoidable problem if there weren't a need for backwards-compatability.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• otto1923 - Sep 28, 2009
  • Rank: 5 / 5 (1)

  Does a harddrive even have the capacity to read or write at 10 gigabits per second?

  Mechanical rotating or switching parts and pieces are dead end tech. The goal would be completely solid state, multi-layered, nano-engineered and assembled, 3D interconnected, flexible, disposable card pda phones, folios, screens, and bricks. No movable, fixable, or replacable parts.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• freelanc3r - Sep 30, 2009
  • Rank: not rated yet
  Forgive my ignorance on some of the more technical aspects involved in this, but if one were to replace the copperwiring found on most consumer motherboards with polymer optic fiber wouldn't that also cut down on a lot of heat generated by the computer? If not a lot, at least quite a bit as well as prep the standard PC for the next round of upgrades for performance and speed by allowing faster interfaces with each of the components?
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• guiding_light - Oct 01, 2009
  • Rank: not rated yet
  Well no one is going to argue the worsening problems of staying with copper. The trouble with optics is you can't make it smaller than micron scale. To collect data from the hundreds of transistors within the area occupied by the optical element, you still need individual copper wires, which overwhelm the benefit offered by the optical part. For scales of mm, cm upward, you have to consider the signal integrity and power of the light source over such scales. So, in short, there is still going to be an incompatibility of scales between chips and cables.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5