Howard Chu (hyc@symas.com) on 3/17/08 wrote:
---------------------------
>That's a non-sequitur. Nobody said it had to be CMOS-based. E.g. graphene...

You were talking about CMOS whether or not you were aware of it. Let us go back to the statement which spawned this dialogue:

"Considering all the news articles about terahertz transistors recently,
it's probably premature to say we've hit a permanent clock speed limit."

The only sort of news you hear about terahertz transistors falls under one of the following four categories:

1. Theoretical models for the f_t or f_max of NMOS devices
2. Theoretical models for the f_t or f_max of GaAs BJTs or InP HBTs
3. Theoretical models for the f_t or f_max of some kind of HEMT or pHEMT
4. A supercooled, one-off version of one of the above

Now, no one cares about GaAs, InP, BJTs, HEMTs, or pHEMTs for any sort of digital logic. This is because all of those circuits consume incredible amounts of static power, which is simply not viable. If it were viable, we'd just use ECL, but we don't. Thus, the only viable THz-transistor technology you could have been talking about are NMOS devices on a CMOS process.

You might be angry at this point and want to scream out, "no one said it has to be CMOS technology!!" Well, the problem here is that there simply isn't any news about THz graphene/ballstic transistors. To the best of my knowledge, there aren't any papers which report f_t or f_max figures for graphene/ballistic transistors. There are papers which report theoretical electron mobility values of 15 000cm^2/(V*s) for graphene, but nothing about actual transistor performance. There are the researchers at Rochester who seem to think they can build single-electron, ballistic transistors, but I haven't seen any results.

I would love to be proven wrong here and shown high-performance carbon transistors, because then I can have more optimism towards alternate transistor materials and structures. I believe that a change of materials/structures is fundamentally necessary towards the continued frequency scaling of large-scale, digital integrated circuits. However, until someone can actually make these devices in more than single-unit quantities, I will remain pessimistic towards the future of digital VLSI.