Future of Classical Computing Hardware: Plasmonics
A recent issue of Science had several interesting papers possibly relevant to the future of classical computing hardware. I'll try to review them one at a time over the next few days.
Ozbay wrote a review of plasmonics. Surface plasmons (SPs) are electromagnetic waves that are confined to the region near a metal/dielectric interface, theoretically much smaller than the (vacuum) wavelength of the light. You can't quite build a standard wire or even waveguide for SPs; proposed structures include an arrangement of nanoscale gold dots on the surface, so that the total layout creates a waveguide or mirror. Ozbay says that work on interfacing external ("normal") optics to plasmonics is proceeding well. However, plasmonics is being touted as a possible answer to the difficulty of intra-chip interconnect bandwidth and real estate, and that field still needs work (signal losses are apparently a big problem). In Ozbay's words,
[W]hen a lot of data need to travel from one section of a chip to another remote section of the chip, electronic information could be
converted to plasmonic information, sent along a plasmonic wire, and converted back to electronic information at the destination.
Unfortunately, the current performance of plasmonic waveguides is insufficient for this kind of application, and there is an urgent need
for more work in this area. If plasmonic components can be successfully implemented as digital highways into electronic circuits,this will be one of the "killer applications" of plasmonics.
But what I don't understand is how this will solve our bandwidth problem. Ozbay mentions that a fiber optic interconnect can carry >1000 times as much data as an electronic interconnect, but I'm not following either the logic behind that or why it translates directly to plasmonic waveguides on a chip. The signalling necessary to convert between plasmonic waves and electronic state is inherently limited in speed. Is it that a plasmonic wire supports faster switching because longer electronic wires have increased capacitance and slower switching times? It's not propagation time, which won't be much different. The structures are not smaller; if anything, they may be larger, and would seem to allow fewer layers of interconnect, so I don't think it's spatially more efficient. Frequency division is one possibility, but I think the plasmonic structures have to be optimized for a particular wavelength. I don't get it. I'm going to have to follow some of the references and figure this out...
The article includes discussion of all-plasmonic chips, plasmonic light sources, and plasmonic nanolithography. The lithography I don't quite grok, either, but a silver superlens helps focus (using near-field effects) lithographic exposure to features much smaller than a wavelength. That looks very promising.
Overall, a fascinating and possibly fundamental shift to how we move data, but it appears to be a long ways to commercial use yet.