DISQUS

enantiomer2000 · 5 months ago

this tech seems even further away than some of the other potential silicon transistor replacements although it seems like it could be more powerful. so are we thinking like 25 years out or so?

GoatGuy · 5 months ago

Yep.

With significant other problems: today silicon features are on the 45-450 nanometer scale (not much chip real-estate is patterned larger, exceedingly few things are smaller. Trend is down. Layering is getting deeper. Tolerances tighter. Thermal physical alignment problems increasing.)

Optical features MUST minimally be on the order of several wavelengths of light. Just the nature of it. Cite not current photolith getting down to 45 nm. Tricks aplenty used, with all kinds of nonlinear materials at play. I don't see real devices, optimized, much below 500 nm, including the waveguides carrying the photons. They need to be very smooth, very tight tolerances. Nanoengineering helps, but again the "vias" and "gates" themselves must irreducibly be greater than the wave-function of the photons themselves. Think of it as "wires need insulation", and so do photons.

The other somewhat elephantine issue is that the molecular switching is accompanied itself by significant energy absorption. Energy-absorption becomes "heat", becomes "aggregate heat", becomes "dissipation". Limits the leveraging to 3D, especially when (necessarily!) tens of billions of optical gates are "computing" whatever they've been intended to compute. Then deleveraging the first paragraph's wavelength issue ... 10,000,000,000 opticals, with vias, taking up 1 sq. micron each ... 1,000,000 to the mm² ... becomes 10,000 mm² ... let's say 10 deep, 1000 mm², or a chip roughly 30mm on a side (somewhat more than an inch).

Except. That we didn't take into account all the vias, busses, routing planes, spectral separators (gratings), various i/o gantries, and more. The 30 mm² on a side chip gets stretched at LEAST 3x in all dimensions. 100mm on a side. Wow ... that's one BIG chip.

The upside might be that the optical components don't depend on big slivers of single-crystal silicon (though I doubt that they'd ever be freed from that substance as a substrate, if for no other reason that much of teh I/O to the chip could also be electrical, which although postulated as a nuiscence, is in fact the unalterable reality of the transition period of "electronics" to "photonics".

25 years?

I don't think even close - and I'm not sure that except for the most specialized uses, the stuff ever becomes mainstream. By the time 25 years passes, silicon (and its successors) will easily be at the 5 nanometer level, a factor of 100 more compact than today (which is already a factor of 1000 smaller than even the most optimistic optical, which frankly will "hold" rather than improve, due to physics). The gap widens. The dream continues. The cloud of gas continues to attract the naive. Optical might be faster, but certainly not in running down traces. 5 nanometer semiconductors are going to be switching on single-digit femtosecond scales (silicon-germanium already is at 0.198 ps = 198 fs for specialized 22 nm devices). at 3e10 cm/sec = 3.3e-11 sec/cm = 33e-12 or 33 ps/cm ... 3.3ps/mm ... 3.3fs/micron ... then deep nanometer semiconductor will be switching at light-micron scales, OR THE SCALE OF INDIVIDUAL PHOTONS within 25 years.

'Cuz, after all ... a photonic circuit can never (ever!) switch faster than the time it takes for photons to arrive, and then leave. Probably quite a bit longer, actually - to do the phase change switching. Molecules don't like rearranging their orbitals faster than the 10-20 femtosecond scale.

GoatGuy