The Death of the Silicon Computer Chip

Stony Stevenson sends a report from the Institute of Physics' Condensed Matter and Material Physics conference, where researchers predicted that the reign of the silicon chip is nearly over. Nanotubes and superconductors are leading candidates for a replacement; they don't mention graphene. "...the conventional silicon chip has no longer than four years left to run... [R]esearchers speculate that the silicon chip will be unable to sustain the same pace of increase in computing power and speed as it has in previous years. Just as Gordon Moore predicted in 2005, physical limitations of the miniaturized electronic devices of today will eventually lead to silicon chips that are saturated with transistors and incapable of holding any more digital information. The challenge now lies in finding alternative components that may pave the way to faster, more powerful computers of the future"

Re:Unlikely

[geekoid]

hmmm, I trust the people I know on the floor more then someone whose job it is to say things that maintain consumer confidence.

It would be a stock hit to say "We will be replacing silcone in x period of time if X is any longer then 'right now'.

Some new technologies solve those problems. Technologies in the 'we hobbled something together proof of concept stage, not the I wrote this down on paper stage.

Some of it is impressive, whether or not there will b a practical way to mass produce it is another thing. If not, I can imagine a time in the future where only large entities that can afford 500K a chip will be using them. Or anyone at home that can afford the latest electron microscope, laser, super cooling.

meh, I'm just glad the MHz war is pretty much subsided and we are FINALLY focusing on multi-core.

Re:I'll...

[iamhassi]

"I doubt silicon will be going anywhere anytime soon - its simply too affordable."

Agreed. Besides, they've been saying this since the 90s, that silicon can't possibly get any faster and it'll be replaced very soon.

I call BS. They had 350 gigahertz silicon chips 2 years ago [news.com]:
"At room temperature, the IBM-Georgia Tech chip operates at 350GHz, or 350 billion cycles per second. That's far faster than standard PC processors today, which range from 3.8GHz to 1.8GHz. But SiGe chips can gain additional performance in colder temperatures....SiGe chips, the scientists theorized, could eventually hit 1 terahertz, or 1 trillion cycles a second."

I think silicon is safe for awhile longer.

Re:I'll...

[scubamage]

You make some good points and I can't really argue them. As the die sizes continue to get smaller, silicon wafers must be more and more pure because tinier artifacts in the wafer can cause issues in the manufacturing process and thats going to be pretty unavoidable. However it also means that more dies can be stamped onto each wafer which should negate the number that are lost. I was meaning more that even if computer hardware is replaced with something else, things which need lower grade integrated circuits are still going to use silicon. I mean, you don't need a 1thz processor for a car's ECU, or for a garage door opener. And as more and more appliances become "smart" more things are going to need lower end chips - so I highly doubt that silicon is going anywhere. Maybe not for pc's, but everything else that is just starting to get 'wired' silicon is going to be around for a VERY long time.

Re:I'll...

[Beetle B.]

Like all new technologies the REAL cost is the in manufacturing and the cost goes down once we've manufactured enough of it to refine the process until we know the cheapest and quickest ways to do it.

Cost is not the main problem with nanotubes.

Nanotubes have a certain chirality - denoted by (m,n) with m and n being integers. Those two numbers define the properties of the nanotube (e.g. if m-n is a multiple of 3, the nanotube is metallic - otherwise it is semiconducting). They also determine the radius.

So far no one has come up with a way to get a nanotube of a certain chirality. They just synthesize many nanotubes and then pick manually the ones they want - if it exists in the sample. Until they can do this, the nanotube industry will not become a reality.

Silicon Scaling

[wilsonjd]

Silicon scaling will run out. We will reach a point where we can no longer make working circuits any smaller, but it will NOT be in the next four years. 45, 32, 22 nm circuits are already in the lab. 16nm (which may be the limit,) is expected to be in production by 2018 (10 years from now.) After 16nm, quantum tunneling may be a problem. http://en.wikipedia.org/wiki/16_nanometer [wikipedia.org]

Intel thinks we may hit the limit by 2021. http://news.zdnet.com/2100-9584_22-5112061.html [zdnet.com]

Homebrew graphene transistors

[the_kanzure]

SciAm is running an April 2008 article on graphene, so here are my notes on graphene fabrication [heybryan.org]. This is pretty neat, and worth some amateur experimentation. You can make the AFM/STM for ~$100 USD [heybryan.org]. As for graphene, there are some instructions on that page for chemically synthesizing it, or just use pencil graphite and write over a piece of paper. Another cool idea is figuring if we can use mechanical force to use a very thin pencil tip to write a circuit. JohnFlux in ##physics on freenode mentions that resistors could be used as a poor man's piezo, just heat up the metal (or perhaps pencil) and it will move. It will move very slowly. But a start.

Re:I'll...

[smackt4rd]

That 350GHz chip is probably much simpler and easier to build than a CPU, but the fact remains that it'd be incredibly difficult to just try and switch from Si to some other semiconductor and be able to build something cheap. We're already starting to see the manufacturers switch their architecture to multi-core CPU's. I think that's alot more practical than trying to switch to an exotic material.

Re:I'll...

[Anonymous Coward]

Why does it have to be smaller all the time? Would it be that bad to double the size of the wafer to add twice the transistors at this point?

Re:I'll...

[Beetle B.]

The issue is not so much that some are being "wasted". The problem is selecting the ones you want. How do you automate that? You have a process that gives you lots and lots of nanotubes. How do you automatically filter out the ones you want? That's been the problem since day 1, and has not been resolved.

Re:I'll...

[camperdave]

My admittedly limited understanding of carbon nanotubes is that they are self producing. What I mean by that is that if one stable tube diameter is 20 atoms, and another stable tube diameter is 30 atoms, then the 20 atom tube is going to continue to grow as a 20 atom tube. It won't spontaneously widen out to a 30 atom tube. If that is the case, then all you would need is a few seed nanotubes, and the right conditions.

Re:I'll...

[imgod2u]

That's for RF chips and RF signals. Silicon Germanium (SiGe) is the material and the 350 GHz signal being propagated is a sine wave with the FET being kept in the linear region. Digital signals are much more difficult to get to 350 GHz.

To give you an idea, in a mixed signal BiCMOS chip where the digital components are standard CMOS and there's a SiGe layer on top for the RF circuits, the RF transistors are capable of amplifying an input sine wave all the way to the multiple tens of GHz. In the same process, the peak switching frequency for a digital signal is around maybe 5 GHz for an inverter let's say.

Digital switching has many properties that an analog signal does not. It has a knee frequency and rise/fall time requirements. It has to switch above and below the detection threshold (and regenerate the signal through each gate rather than propagate any low signal levels). Being able to have a transistor propagate a 350 GHz sine wave is orders of magnitudes easier (and possible) than having a flip-flop operate at 350 GHz using the same process technology.

Re:I'll...

[default luser]

One of the problems you have is that gate volume is approaching thousands of atoms. This is a problem because certain regions need to be doped in order to make the silicon do it's job.

What is the problem, you may ask? Well, just look at the Wikipedia entry you linked. Even doped silicon is still %99.999999 pure.

So, you have your gate that is thousands of atoms in volume, and dopant concentration that is in the 1 million to 1 billion ratio...so what is the likelyhood that your gate is going to contain that one dopant atom you need in the lattice for best performance? Without that dopant, the performance of your gate suffers, and may not even work at all.