Is Moore's Law Based on Knowledge? Or is it Recursive

[maverick_starstrider]

Hi, here's my question. If all our technology were to suddenly disappear but all our knowledge and designs were to remain, would our progress back to where we were, in terms of chip design, follow Moore's Law? In other words, could one just make a chip with speed X provided one had the blueprints (and laser lithography or whatever), or does one need to start with chip Y (which is extremely slow) and that is needed to make a machine that can make chip Y*2 which is needed to make a chip that is Y*4, etc. Just curious.

 

[chiro]

Hi, here's my question. If all our technology were to suddenly disappear but all our knowledge and designs were to remain, would our progress back to where we were, in terms of chip design, follow Moore's Law? In other words, could one just make a chip with speed X provided one had the blueprints (and laser lithography or whatever), or does one need to start with chip Y (which is extremely slow) and that is needed to make a machine that can make chip Y*2 which is needed to make a chip that is Y*4, etc. Just curious.

Do you mean something like if all the physical manifestations of the technology (CPU chips, lasers and so on) and the equipment used to make them disappeared, but all the textbooks, diagrams and information in some linguistic form didn't?

 

[maverick_starstrider]

Do you mean something like if all the physical manifestations of the technology (CPU chips, lasers and so on) and the equipment used to make them disappeared, but all the textbooks, diagrams and information in some linguistic form didn't?

Yes. In a nutshell, do you NEED a pentium 100 to build a pentium 200, etc.?

 

[Ryan_m_b]

I think I can answer your question. If you were to form a society that was roughly similar to that of the world in the 1950s Moore's law may appear again.

It's important to remember that Moore's law is not really a law, it was an observation by Moore that transistor count doubles every two years (IIRC) that the computer industry then took on as a running goal. What has allowed the trend to continue for so long is the huge commercial applications in the market and leading to greater investment in how small transistors can be made.

It is possible that society could have developed differently and at some point jumped to chip X without making chip Y (perhaps because there was little incentive for computer science at first but lithography improved independently).

 

[maverick_starstrider]

I think I can answer your question. If you were to form a society that was roughly similar to that of the world in the 1950s Moore's law may appear again.

It's important to remember that Moore's law is not really a law, it was an observation by Moore that transistor count doubles every two years (IIRC) that the computer industry then took on as a running goal. What has allowed the trend to continue for so long is the huge commercial applications in the market and leading to greater investment in how small transistors can be made.

It is possible that society could have developed differently and at some point jumped to chip X without making chip Y (perhaps because there was little incentive for computer science at first but lithography improved independently).

I understand that over-arching reason for Moore's Law but I don't understand the deeper why. WHY does it take 2 years to double the number of transistors? What is the limiting factor? Is it simply lithography technology?

 

[Ryan_m_b]

I understand that over-arching reason for Moore's Law but I don't understand the deeper why. WHY does it take 2 years to double the number of transistors? What is the limiting factor? Is it simply lithography technology?

I don't think there's a firm reason, there's probably a whole host of factors including the need for better lithography. The two year time frame is a business practice that has been very good at marshalling resources to try and achieve the goal by that time.

Moore's law isn't really that precise, looking at this chart we can see that there have been various times where transistor count hasn't doubled (see Pentium II to Pentium III) and times when it has more than doubled (see 8085 to 8086).

[URL]http://upload.wikimedia.org/wikipedia/commons/0/00/Transistor_Count_and_Moore%27s_Law_-_2011.svg[/URL]

In general the two year target holds because out of all the computer companies striving to get a chip out that's twice as good as the last one of them is bound to make it. Of course this isn't going to go on forever and we've probably only got a decade or less before we see the end of Moore's law.

 

[Proton Soup]

if we had to start over, we wouldn't even have the tools available to view the designs for the latest computers. things would progress faster given what we now know, but there is still an element of having to build the tools to build the tools to build the tools...

 

[f95toli]

I understand that over-arching reason for Moore's Law but I don't understand the deeper why. WHY does it take 2 years to double the number of transistors? What is the limiting factor? Is it simply lithography technology?

It is a combination of technological factors (lithography, materials, tooling etc). But that it is (or rather was) two years is also due to the way Intel and co operates, it is how they organize their projects (Intel has very strict time tables from R&D to market). 2-3 years is a common running time for projects in large companies.

Also, in at least one case Intel demonstrated a CPU a bit prematurely, simply to make Moore's law "valid" for one more year,

 

[Newai]

Uh, I thought Moore's Law was about the costs, not the power.

 

[Ryan_m_b]

Uh, I thought Moore's Law was about the costs, not the power.

It's about the number of transistors you can fit on a chip. Consequently the price of the same concentration of transistors will half. E.g. if 1000 transistors per chip cost £1000 then in two years 2000 transistors per chip will cost £1000.

 

[diazona]

The most sensible (in my opinion) interpretation I've heard is that Moore's law is simply an empirical observation. There isn't any particular reason that it takes about two years to double transistor count, other than that that's the pace that computer chip development happens to proceed at.

 

[apeiron]

If we had all today's knowledge and we were back in the 1950s, we could of course shortcut Moore's law. So the law is partly about the accumulation of new knowledge.

But there is also the economic and market context. Each new generation of smaller circuitry demands larger, more expensive, factories. So you have to have a customer base that pays for things to happen.

With today's knowledge and yesterday's markets, it would be back to bespoke defence systems at first most probably. Smart kit for the few.

On the other hand, if we could wipe away the Intel microprocessor design legacy, commercial chips might in fact leap way ahead of where they are now due to better, more streamlined, processing architectures. So we would perhaps be doing better than Moore's law (more parallel processing, for example).

So Moore's law is basically incremental because a lot of manufacturing lessons have to be learned to achieve each step up in miniaturisation. But also because the whole market has to grow to support the industrial effort. You can't run ahead of the software developers too far, and you can only leave behind legacy features gradually.