Hacker News Reader

Comment by jonnathanson | Hacker News Reader

If we look at Moore's Law as it was originally stated, i.e., w/r/t transistor density on chips, then yes, it is up against a serious obstacle in the near future. That obstacle is quantum physics. Below a certain nanometer count, you're working at truly atomic scale, and Heisenberg effects start to kick in.

Quantum computing may solve this issue, but realistically, probably not at a pace quick enough to keep Moore's Law operating on track the way it has been historically. More likely, we'll hit a plateau for awhile and eventually shatter it. When that happens, computers will be very different machines from what they are now. The shift from transistor-based computers to quantum computers will be akin to the shift from vacuum tubes to transistors.

In the meantime, we're probably just going to load up and more and more parallel processors.

View on HN

Moore's Law doesn't explicitly say anything about density:

The most popular formulation is of the doubling of the number of transistors on integrated circuits every two years.

(WP http://en.wikipedia.org/wiki/Moores_law#Other_formulations_a...)

So the density needs to increase only of the size if the size integrated circuit cannot increase.

A more realistic direction is up. Or, 3d integration. When the manufacturing gets mature, this gives a large immediate boost, because in two chips of size n/2 that are bonded together, the maximum distance between any two circuit elements is smaller than in a single chip of size n. After that, it will allow scaling by increasing layers.

Agreed, but we should note that the fundamental problem is not transistor density or quantum physics, but heat dissipation. Going 3D makes smaller process nodes less important, but efficiently removing heat from a die stack is a real challenge. Sandwich a 130W TDP processor die between two other similar die and the heat dissipated per cm^2 is tremendous.

IANAEE but is there some way to integrate peltier junctions into a 3D microchip design so that the heat all gets funneled out of the CPU in a regulated and controlled manner?

Not my area of expertise, but I believe most research today is on microfluidics, introducing microchannels into 3D interconnect layers alongside the TSVs to cool the device. For example, http://deepaksekar.weebly.com/uploads/1/9/8/6/1986457/3d_mic...

> The shift from transistor-based computers to quantum computers will be akin to the shift from vacuum tubes to transistors.

The shift to quantum computers will be much larger, more akin to the shift between the Jacquard loom and Turing equivalent languages.

Transistors were (initially) an incremental improvement over tubes, which were themselves an incremental improvement over relays. Admittedly these "increments" were order of magnitude jumps. But you still had the same thing - it was smaller, faster, more efficient but still fundamentally a mere digital switch.

Of course chances are good that we'll miss what quantum computers are really capable of (superposition) and just use them as better single-atom switches.

I sort of expect that we'll also switch over to some material with higher electron mobility at some point if we can't keep shrinking features fast enough for silicon to maintain its lead. That'll just be another spurt of performance improvement after a N year interregnum, though.

Some responses to various assertions in your post:

- transistor density: only when taking 2D density into account. There is a whole third dimension to start using. Although it won't scale exponentially in sheer numbers, it may scale by introducting new topologies.

- 'Heisenberg effect': single atom transistors have been built and uncertainty in the observation of its state is not the problem. The Heisenberg uncertainty principle is not the issue here.

- Quantum computing will not solve the issue. It can solve some specific classes of problems spectacularly fast. For most common problems, it's not even (significantly) faster in theory.

- Nobody knows what kind of quantum computers will be technologically possible. There is no reason whatsoever to assume that quantum computing will ever replace 'regular' computing. My personal prediction is that it will track the record of fusion power for the decades to come: i.e. no practically useful results. Experiments in my university tended to go awry if someone in across the hall received a mobile telephone call. They invested quite some time and energy into shielding the experiments, to little avail.

And a minor nitpick:

- "quantum physics" is a theory. The obstacle isn't "quantum physics", it's the experimentally observed granularity of reality and what you can reasonably build out of those granules. "Quantum physics" is the problem as much in this realm as "Quantum physics" is the reason you can't make anything you want out of legos.

Your points are well taken, but the minor nitpick is a bit unfair. I was using the phrase "quantum physics" as shorthand for what, I suppose, I could have more specifically described as "...the mechanics and dynamics of matter at a subatomic level, at least according to theory."

I'm a layperson, and you are a physicist, and there is no question that you understand these domains a lot more clearly and thoroughly than I do. And there is no question that my usage of certain terms is going to strike you as imprecise. That said, I believe there can be some leeway in the usage of terms to make them more immediately communicable to an audience of mostly laypeople. Smart laypeople, sure, with a greater basic understanding of these topics than your average person on the street. But, by and large, not an audience of trained theoretical physicists.

When I said that "quantum physics" is "the problem," I did not mean to reify quantum theory. Then again, I don't think most readers of my post would assume I was trying to do so. While your nitpick is fine, in as much as I wasn't as specific as I could have been, it seems a lot of quibbling over a semantic triviality.

Reply on news.ycombinator.com