Robert J. Bradbury wrote:
>
> So unless you have read my papers, you are probably unaware that
> during 1998, I became convinced VSCEL lasers were going to play
> an important role in connecting the CPUs (nano or not) together in
> the not so distant future. Well that insight appears to have
> been only ahead of the reality by perhaps 5-6 years (probably
> more Signs of the Singlularity).
>
As you are probably aware, VCSELs are the basis of inter-chassis
connections in modern core internet routers. Today's VCSELs can be used
at 2.5Ghz in current designs. This is for real products, not just
lab stuff. Faster lasers are vastly more expensive and are used
for long-haul fiber. the current "standard" device has 12 vcsels
and a fiber connector that connects to 12 fibers for a total of
30Gbps.
Several companies are working on a neat new VCSEL device: four
VCSELs of slightly different wavelengths mounted in a package that
directs them into the same fiber. multiple wavelengths per fiber is
is called "wavelength division multiplexing" or WDM. This low-cost
version is called coarse WDM, or CWDM. This puts 10Gbps into one fiber
fairly cheaply. This is the proposed standard for 10Gig-e, which
is the latest generation of ethernet. You can design products using
this technology now. It's not just in the lab. The 4-VCSEL/1 fiber
arrangement is market-driven by 10Gig-e, not by any particular
technical constraint. I anticipate that within a year we will see
denser versions with 16 or 32 VCSELS/fiber. Note that dense and
ultra-dense DWDM and UWDM systems are already used for long-haul,
but these are hand-built from expensive parts bulky parts, not
mass-produced VCSEL devices.
For the last several years, the rate of improvement for data
transmission (dollars/bps) has substantially outpaced the rate
of improvement for computing, which itself is improving. This in
turn reduces the penalties associated with massively parallel
computing.
This archive was generated by hypermail 2b30 : Sat May 11 2002 - 17:44:14 MDT