OK one more Colwell-ism and then I am headed for a three and a half day weekend, even if Moore's Law complely collapses Saturday morning. Here it is:
Colwell on post-CMOS entrepreneurs: "Watch out for opportunities if general-purpose processors stall out...however please don't design something you can't program. We've done that too many times and its embarrassing."
Part of the problem with Moore's Law is that there is an inverse law that goes along with it: "For every increase in computing capability, there is a slightly larger than proportional increase in computing demand." That's why, while for a long time, big iron computers got smaller and smaller, but now the the big iron (super computers and data centers) can now be so large as to dwarf the old "giant" computers.
I hate to bring up the K acronym (KGD) but the way that most companies are pushing along at Moore's rate is by stacking and vertical growth. Thus, KGD comes into play. KGD has been one of the persistant "on the horizon" big things for 30 years. I work for a BI and Test company and we are seeing significant movement in WLBI. This is the way to keep the Moore's law train rolling.
I like Duane's comment about the engine. "Silicon is nearing that wall. What's out jet engine?" I thought the cognitive computer chip was interesting emulating neurons, synapses, dendrites, and axons in the brain's neural networks using special-purpose silicon circuitry." It's stil silicon though.
'Corelets' Prime Cognitive http://www.eetimes.com/document.asp?doc_id=1319173
I could see it with some sort of CVD, but the problem would be keeping a crystal structure in the build. Unless you're hysically pushing atoms around, there's no guarantee to keep a planer, monocrystaline surface, and that's required for the semiconductor to work.
Garcia - I grew up in Kelso, Washington, on the West coast of the USA. Our little airport had the manufacturer of one of the most famous and closest to working flying cars back in the 1950s / 60s. I think he built like two or three of them. No one has done any better than "build two or three." The current ones are still just hype as far as I can see.
Rick, but if you back up the scale a bit might it become possible for higher-level manufacturers to work with them directly? I keep thinking about Google / Facebook / Amazon designing their own servers. Why not move to the next level as well?
Colwell notes that exponentials such as Moore's Law are rare in electronics and life in general. Say goodbye to one but don't expect another ride that fast for awhile. It was the best roller coaster we may see.
Curently we are incredibly invested in a "Silicom" ecosystem. There are things we can do, like using silicon-germanium alloys that will boost speed and reduce power at the same technology node -- and we have thinks like the new FinFET technologies.
But eventually I have no doubt that we will move to graphene-based devices (or something else ... who knows?)
@Rick: They tore it apart and found two of the transistors had no wires going to them.
I remember an old computer in the UK -- thsi was a "big boy" -- not a personal workststion -- it came running with one clock speed -- when you needed to upgrade, you could buy another motherboard that ran at twice the clock speed.
The man woudl come from the factory, take your motherboard out, and plug the new one in, and leave. As soon as he was out of sight, he would flip the "2X" switch on your old motherboard -- this was the one he would install in the next company's computer...
Here's what I'm talking about - In the early day's of aviation, piston engine / propeller drive aircraft got progressively faster. Engines got bigger, with increased horsepower. The last days of the piston engine saw some incredibly complex monsters.
But the piston and prop hit a pretty solid wall. Fortunately, the jet cam along at about the same time. It's completely different. Different principles and different implementation. It's not even measured the same. Piston in horsepower and jets in thrust.
Silicon is nearing that wall. What's out jet engine? Graphene? Quantum? Something else?
The pocketbook issue is you can use those extra transistors to create something that compels people to give you significantly Moore money every two years. Without the extra transistors, you may only get a little more money. Electronics becomes less exciting for consumers, Wall Street. Oh oh
Maybe the questions isn't "will Moore's Law stop?" so much as it is "What will replace it?" Just because at some point Si can't handle any smaller geometries doesn't mean something else won't pick up the wagon and allow continued improvements in performance. We may just have to change the scale.
Another Colwell classic: Like's learning moments: He had his first transistor radio, a six transistor model. His friend had an eight transistor model and he was jealous. They tore it apart and found two of the transistors had no wires going to them.
@Rick: People treat Moore's Law like the dying Tinkerbell in Peter Pan. Just believe hard enough and it will come back to life.
LOL I remember in 1990 when we moved to the 1 micron node ... people predicted the end of Moore's law .. they said 0.5um (500nm) was as far as we could go ... then we got to 500 nm, then lower, and lower and lower ... and we're still plunging on...
Regarding the topic of the day, I'm not so sure it is that big a deal. We are still working out what can be done with customized SoC's. It actually might be a good thing to slow down the process changes to let us catch up.
Blog Make a Frequency Plan Tom Burke 17 comments When designing a printed circuit board, you should develop a frequency plan, something that can be easily overlooked. A frequency plan should be one of your first steps ...