also, what's with all these SEMI-conductors?
We need proper good old fashioned conductors, not this half-assed semi shit. That'll make your computer thingies go faster.
The world might still be 20 years from the end of Moore's Law, but the hunt for technologies to replace semiconductors is going on right now. A group from Michigan Technological University is offering one such alternative: a quantum tunnelling transistor that operates at room temperature. The culmination of work begun in 2007 …
Doing it at room temperature and with single electrons is a huge step forward.
Note that's not really "transistor" action. What you've got is conduction by quantum tunneling. You have a potential difference between 2 contacts and current flows. That's a description of any conductor
But you've done it through a material which is basically an insulator.
What you need is some way to control that conductivity. The 3rd terminal crossing the BNNT.
Given the very one dimensional nature of the conduction it would seem that a BNNT chip would be a series of nanotube pieces spliced together or criss-crossing each other, as opposed to just laying the switches on top of a semiconductor as happens with current chips.
Manufacturing that structure (in bulk) is going to need people who are much smarter than me. Perhaps some kind of biomemetic or emulsion technology for the quantum dots?
Thumbs up for room temp and a very different but still electrical system.
And from the announcement:
Theoretically, these tunneling channels can be miniaturized into virtually zero dimension when the distance between electrodes is reduced to a small fraction of a micron.
So they've got a way to go yet before they can beat the current state-of-the-art 14nm processes.
The 14nm feature size is just the size of one of the gates on a fin-fet. Intel calls this tri-gate technology as they use 3 fins to form one gate. Total size from jut looking a at a image from the intel web site is quite a bit larger, the foot print per gate is about 140 nm by 140 nm (excluding the spacing between gates).
2d area wise they are about the same. It comes down to heat dissipation and cost of manufacturing (as well as if these structures can be stacked.
" I suppose this is going to make Haswell look like computing by making marks on a stick with a knife."
Not anytime soon...
But it does raise the question about what software we'll be using if the technological progression keeps on seeing processing power doubled every eighteen months. Time for a new ground up OS, anyone? Or will we keep buying scrub ups of Windows, Linux, and various *nix derivatives with technological roots in the 1990's?
"Or will we keep buying scrub ups of Windows, Linux, and various *nix derivatives with technological roots in the 1990's?"
I think the problem is one of "compatibility".....a new grounds up OS might not allow companies and individuals to easily transfer their data sets from the "old" platform to a new one.....and in the meantime, those companies that have made the transition will then find that their data isn't compatible with their suppliers and/or customers.
I'm sure provision can be made for this, but it'd be a huge upheaval.... :-(
"I think the problem is one of "compatibility".....a new grounds up OS might not allow companies and individuals to easily transfer their data sets from the "old" platform to a new one"
I'm unconvinced. Digital data is usually very easy to convert. Several decades ago, in my first job after graduation I had half the job of cobbling together the data from a Honeywell DPS6 to an ICL VME mainframe. Both used their own proprietary file formats and languages, and the link was done in half a day, and it worked, despite my involvement.
There's other compatibility issues of retraining the meatsacks and changing the infrastructure in whatever way is appropriate, and I'd guess that people and cost will continue to be barriers to change.
Yea! Ledswinger - I remember doing a piece of demo software to convert ICL punch card data to Honeywell DPS6 ASCII.
Then there was the intractable problem of the "currency character" code: £ in the UK, $ in the US, etc.
Oh happy days of incompatibility!
Is it any better today?
Things haven't been doubling every 18 month at a given price for a while (*). Most of the recent work has been concentrating on reducing power budgets - so a machine I buy today is only slightly faster than 2 years ago, but it uses about half the power.
(*) ESPECIALLY when you factor in outboard components. Consider that the latest generation of Haswell PCs are 20% more expensive than the Ivy Bridge ones they're replacing.
> But it does raise the question about what software we'll be using if the technological progression keeps on seeing processing power doubled every eighteen months.
You can be sure that Windows will still be trying to do essentially the same thing as Windows 95 (run programs in a graphical environment) but it will inexplicably require a 4-core 1THz Intel quantum CPU with 10TB memory to do it
But processing power isn't doubling every 18 months. Ivy Bridge to Haswell in some cases you only see around a 10% increase; that is a far cry from doubling. Moore's law says the number of transistors double every 18 months. Not all of those transistors are going to the CPU execution unit(s) but to these days GPU, cache, new features like AVX extensions etc.
If you examine what the OS does, you will see it is an interface between a vaguely von Neumann architecture and what humans want to do with computers.
Changing the technology of the Von Neumann architecture doesn't affect the OS at all.
Oddly enough, after a, million years, through wood, iron steel and rubber developments, wheels are still round..
Sorry, but that won't happen until the smallest PC available will be able to render full-size triple HHHHD over any screen resolution with an infinite amount of objects, in real-time.
When we get to that level of computing power, we might pause for a week or two before considering just what we could start thinking about (double the 200k x 160k screen size ! Twice the amount of objects to render ! Half the power requirements ! ...) to improve our computing power even more.
They say the dots are 3nm wide to ensure they only carry a single electron at a time (which is quite clever) but they seem to be saying the large dimension needs to be in the micrometre range to make it work.
Not good news if you want to push for the ultimate density in devices
That would suggest switching nodes would need to be > x Micrometres apart. I hope I'm wrong on this one.
Note that while this uses quantum effects it's not really the common usage of the term "quantum computer", qubits, all solutions generated simultaneously, no registers, no programming language.
OTOH it is the sort of thing which you could put in a chip carrier and solder to a circuit board, which mfgs like a lot
Sorry folks but I just don't see any of those weird and wonderful devices beloved of SF computers any time soon.
The world might still be 20 years from the end of Moore's Law, but the hunt for technologies to replace semiconductors is going on right now.
It would be rather stupid to wait until we ran out of ways to improve current technology before starting to research alternatives...
Oh, but wait, that's what people think we should be doing with the internal combustion engine, isn't it?
This quantum transistor is already bigger then transistors in circuits being made at Intel and IBM. There is nothing new about room temperature tunneling the Zener Diodes been doing it for 40 years commercially.
The article says it can get down to single atoms- but the nanotube is inherently limited to tens of atoms or more so the sketch made is very flawed at delivering the solutions promised.
How about the details of why this is special?
It occurs to me that a 3-D matrix "positronic matrix" of these with some sort of memristor coating ie TiO2 embedded in a composite with room temperature superconducting buses for the data nodes could be used as a very advanced neural net based on quantum principles.
Such a device would be far more advanced than anything we have at the moment and could learn thousands of times faster than the best FPGAs.
Said room temp. superconductors might end up being based on carbon nanotubes doped with barium or similar, or possibly a barium/lead compound resembling a coaxial cable at the nanoscale.
AC/DC 6EQUJ5 and no I am not "Skynet", honest :-)
Biting the hand that feeds IT © 1998–2022