ok, I'll put my propeller hat on.

the highest data rates achieved (error-free BER of less than 1e-12 without FEC) with NRZ directly-modulated 850nm VCSELs to-date are 44Gb/s for 850nm, 49Gb/s for 980, 40Gb/s for 1100nm, and 35Gb/s for 1550nm.

that's not POET, that's the world. in the best labs money can buy.

in their labs, finisar has tried to push this to 55Gb/s but the bathtub curve eye diagrams look like something Jackson Pollock painted. they're just awful.

to put this in perspective, commonly used high-end rates in datacentres today are 25.78Gb/s (ethernet, infiniband), and 28.05Gb/s (fibre channel)

980nm actually has GREAT advantages, but the bulk of the industry still uses 850.

(explains the shift: you sell what they use, until they use what you sell...)

why 850? because that's what cheap consumer digital cameras use.

850 is a bit of a sweet spot between sensitivity and interference from visible light. lower wavelengths are often polluted by things like the image source (projector or LCD) or background lighting (e.g., the lightbulbs in your house).

the monstrous manufacturing volume in the camera space makes it really cheap for the stingy industry to adopt for datacomms application.

with 980, advantages compared to 850

• threshold current is reduced by a factor of 2
• voltage is reduced by .2V
• power output is increased,
• external quantum efficiency is not affected.
• with detectors, responsivity is .6dB better
• and it's compatible with 1300nm links

with a 300m 12.5 gigabaud serial link, using 980 vs 850 you should get around this:

• power range -1.5 to -6.5 dBm
• wavelength floor 970nm
• rise/fall time 26 picoseconds
• extinction @7dB
• OM3 modal bandwidth (EMBc) of 1900MHz/km (at 2200MHz/km, link margin of +0.8dB); OM4 would get you around double that
• eye-centre sensitivity -14.5 dBm (class1 laser safety, better by 2.6dB)
• power budget +8dB (link spec margin improvement of 3.2dB)

although I'm sure Geoff would never release hard data until there was a datasheet, I'd expect these sorts of improvements in POET vs others

• current better than 600 uA
• slope efficiency better than 0.3W/A
• series resistance better than 100 ohms
• 3dB bandwidth better than 22GHz at -3V
• system consumption well under 25pJ/bit