Optical antennas working

You guys have heard me talk about my antenna-coupled tunnel junction optical modulators and detectors before. They're basically stuff from the Radio Amateur's Handbook, scaled down by a factor of 100,000 and used in the infrared--basically 200 THz crystal radios. They're supposed to talk to each other via ridiculously small silicon waveguides (0.45 micron x 0.22 micron cross section), and eventually may be used to replace wiring in computers, though there are other competing approaches, of course. I've been coding simulators, optimizing designs, laying out masks, babysitting fab runs, and building characterization setups for the past few years, with nothing much to go on but faith and remarkably patient management. Between salary, overhead, fab costs, and lots of equipment, IBM has probably sunk 2 million bucks into these gizmos, based on nothing much more than suggestions in the literature and my say-so. A big bet, career-wise, and one that has been getting a lot less comfortable lately.

The new news: they work.

It isn't the absolute biggest signal in the world, but I can actually detect 1.55 micron light in a waveguide with what looks like reasonable efficiency, all with little bits of mildly oxidized metal. I'll tell you more about them as the results come in, but it won't take so much nerve from now on.

Cheers,

Phil Hobbs

PS: I've tried sending this three times in two days, and it hasn't shown up. Apologies if you're seeing it more than once. On the other hand, I've been seeing it in my dreams for awhile now....;)

Congratulations! Please accept this virtual cigar.

--
Bill Sloman, Nijmegen

Congrats!

Sounds like quite a struggle with a final positive output.

Robert

Hi, Phil,

Gonna publish?

If they're basically crystal-radio diode detectors, they'll be speed-limited by the "baseband" stuff, diode impedance and all the various capacitances. I guess the next step, to get the speed up, would be some sort of integrated active detector.

I'm guessing that other apps will emerge before these are used to replace metallic conductors in computers. Maybe something even more interesting.

John

My congrats, Phil. At liquid nitrogen/helium or at room temperature ? I've seen an array of them detecting 10um, at liquid nitrogen or helium. I forgot which.

Rene

--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

...

200 THz antennas? What happens if you point a big array of them at the sun?

Cheers! Rich

Thanks, all. They work fine at room temperature, but are about a factor of 2 better at 77K due to less thermal broadening of the Fermi level--the tunnelling barriers are only about 0.2eV, compared with a few volts for more typical systems. Room temperature (25 meV) thus smears out the tunnelling behaviour more with these devices than, say, Al/Al2O3/Al. To be useful in computers they have to work at about 100 C, of course, and they seem to do so, with further graceful reduction of performance. We hope to be able to adjust the barrier heights up slightly to compensate--we've already improved the room temperature performance by about 3x that way.

The key is that the tunnel junction sits across a travelling wave region, so the device isn't dominated by RC effects--at 200 THz, you're down by 3 dB with an RC time constant of 0.8 femtoseconds, which is a bit challenging in a lumped-element circuit. Using a travelling-wave device means that we can avoid the problem--we don't have to drive all the capacitance at once. (We're worrying about single-digit attofarads here.) This is just like an analogue oscilloscope's vertical amplifier.

The capacitance seen by the optical signal doesn't take into account the pads and so on, because the antennas are designed so that the pads are essentially ground at optical frequency--the antenna arms are about 3/4 wavelength long on the side without the pads, and 1/2 wavelength on the pad side. (I realize this is hard to visualize--I'll post a link to the writeup when it's done.) We'll publish the first announcement in Optics Express, and follow it up with an Applied Optics paper, probably, or maybe Nature if we can get the performance really juiced soon enough. There's a paper on last year's results on my web page,

They were extrapolations from DC I-V measurements to infrared detection, which sounds ridiculous but actually seems to work for these gizmos.

The baseband speed limits are considerably higher than quantum detectors, because the capacitance is so small--basically just the pads, which are around 1 um square--maybe 400 aF. Photodiodes and TIAs are limited by the RC bandwidth on the electrical side--the BW of a vanilla TIA is about sqrt(f_RC *f_T), so to get a decent speedup you have to burn a lot of TIA power. It's really a power vs bandwidth issue that is driving the on-chip optics effort. I think you'll see lots of optical interconnect inside servers by the 35 nm node, and maybe on-chip optics by the 22 nm node.

Cheers,

Phil Hobbs

Is there any reason why these won't work a little bit higher in frequency, in the visible?

The performance of the metal antennas is already tanking by 1.5 um--metals are very dispersive in the visible, and none of them is very metal-like. The waveguides are made of silicon, which of course is opaque below about 1.2 um. You really need the high-index material to get good coupling to the metal--with glass guides, the mode just splits in two, one half going into space and the other into the substrate.

On the other hand, you'll certainly get *something* out of them, just probably not at reasonable quantum efficiency.

Cheers,

Phil