You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
It's a ways out as a product, but offers a pathway to the future. This might be a fundamental turning point, I just wonder how long. I figure five years, any other thoughts on a time frame appreciated.
The rise of the LinLET, maybe I should trademark that right now. No doping is a big plus. Power consumption at scale a question. Their point about using multiple color lasers could be huge. (Intrinsity Fast14 for light?) Managing the light sources might require significant MEMS advances along the lines of the TI DLP technology scaled way up for billions of interconnects. Five years seems a bit optimistic, but 10 doesn't seem outrageous to have a working processor.
I do find the applicability limited. The gate is controlled by optical but output is electrical so you can't connect the output to an input of a next stage. So it does not seem compatible with current std. cell digital flow.
The 80s called, they pointed to the "LED" and "optocoupler" entries in Wikipedia. Lots of issues in scaling it up, like fanout and process steps and switching speeds and domain management and EDA tools and debug and probably a dozen more. I would not presume this instantly becomes competitive with today's advanced CMOS, but it would likely be more of an SSI > MSI > LSI > VLSI path replayed in optical domain as these issues get worked out. Of course, it could also be an ECL replay, fast but very finicky and not worth the trouble.
I can barely spell graphene. One of the advantages of this optical process is no doping, and from what I know about graphene, that goes in the opposite direction with lots of careful doping required.
Staf's point was a good one. The cost for "a" optical transistor might be pretty low, but the cost for "a billion" optical transistors might be high.
After a little more thought I do think that the proposed transistor is not well fit for digital applications. I think if you need to talk about a lens needed for each transistor, it will take up simply take too much space.
A missing element is an electrical to optical convertor; e.g. with a material that emits light when a current runs through it and just printed as a line on top of the presented structure you can make a transistor. If additionally the amount of light is proportional to the current and the current is also proportional to the amount of light you have something that represents a bipolar transistor behaviour. For digital this would still not be ideal as it depends on static currents but it may be an ideal analog transistor.
You can do the same thing with a field emitter(typically tungsten or molybdenum, but also CNT). The tip of a field emitter has a typical radius of 1 nm, and it is readily triggered by a laser pulse. At one point it was thought that field emission is rather slow, but it is extremely fast [See Fig. 7 in International Journal of Quantum Chemistry 70 (1998) pp 703-710] and more recent papers by others. The turn-on time can be about 1 femtosecond.
mhagmann, Thank you for the information, sometimes it takes lots of people to put the pieces together and welcome to SemiWiki. I hope you enjoy the discourse on everything as much as I do.
