Moore's Law Isn't Dead, But Needs a New and Broader Interpretation: Intel's Top Exec (wired.com) 82
On Sunday, Intel held a five-hour event, where 100 attendees from startups, venture capital, and tech giants drank in semiconductor-themed cocktails and detailed explanations of how sand is processed into silicon chips. It was a celebration of how exponential upgrades from the chip industry have propelled progress in technology and society over the past 50 years -- and an argument that the party's not over. From a report: "It's going to keep going," said Jim Keller, a semiconductor rock star who joined Intel last year as senior vice president of silicon engineering, and a cohost of the event. "Moore's law is relentless," he added, referring to the 54-year-old assertion by a former Intel CEO that the number of transistors that could be fit onto a silicon chip would double on a predictable schedule.
Intel still dominates the market for server chips that power cloud computing, but its two most recent generations of chip technology arrived late. [...] "The working title for this talk was 'Moore's law is not dead but if you think so you're stupid,'" he said Sunday. He asserted that Intel can keep it going and supply tech companies ever more computing power. His argument rests in part on redefining Moore's law. "I'm not pedantic about Moore's law talking just about transistors shrinking -- I'm interested in the technology trends and the physics and metaphysics around that," Keller says. "Moore's law is a collective delusion shared by millions of people."
Keller said Sunday that Intel can sustain that delusion, but that smaller transistors will be just one part of how. On the conventional side, he highlighted Intel's work on extreme ultraviolet lithography, which can etch smaller features into chips, and smaller transistor designs based on nano-scale wires due to arrive in the 2020s. Keller also said that Intel would need to try other tactics, such as building vertically, layering transistors or chips on top of each other. He claimed this approach will keep power consumption down by shortening the distance between different parts of a chip. Keller said that using nanowires and stacking his team had mapped a path to packing transistors 50 times more densely than possible with Intel's 10 nanometer generation of technology. "That's basically already working," he said.
Intel still dominates the market for server chips that power cloud computing, but its two most recent generations of chip technology arrived late. [...] "The working title for this talk was 'Moore's law is not dead but if you think so you're stupid,'" he said Sunday. He asserted that Intel can keep it going and supply tech companies ever more computing power. His argument rests in part on redefining Moore's law. "I'm not pedantic about Moore's law talking just about transistors shrinking -- I'm interested in the technology trends and the physics and metaphysics around that," Keller says. "Moore's law is a collective delusion shared by millions of people."
Keller said Sunday that Intel can sustain that delusion, but that smaller transistors will be just one part of how. On the conventional side, he highlighted Intel's work on extreme ultraviolet lithography, which can etch smaller features into chips, and smaller transistor designs based on nano-scale wires due to arrive in the 2020s. Keller also said that Intel would need to try other tactics, such as building vertically, layering transistors or chips on top of each other. He claimed this approach will keep power consumption down by shortening the distance between different parts of a chip. Keller said that using nanowires and stacking his team had mapped a path to packing transistors 50 times more densely than possible with Intel's 10 nanometer generation of technology. "That's basically already working," he said.
Re: (Score:1)
What kind of talk is that?
Re:intel isn't dead (Score:4, Funny)
PININ' for the FJORDS?!?!?!?
What kind of talk is that?
Pinning is when a process is to be executed on a specific processor on a multi-process systems.
FJORDS is a complex Assembly instruction: Float Join Orthogonal Register Decimal Shift
Re: (Score:2)
I've been waiting decades for hardware decimal support to make a comeback, well done Intel! Well done.
Re: (Score:2)
It's just resting
No, its just fucked, the same way Motorola was fucked when it failed to keep up with its competition.
To be clear here, Intels competition isnt AMD because Intel used illegal practices to force AMD to spin off its competition.
The rent-a-fabs are Intels competition. The industry is semiconductor fabrication. Thats what Intel does, and they are now fucked at it. No magic bullets this time.
Re: (Score:2)
Nobody buys Intel microprocessors, they're overpriced garbage. People buy cheap random ARM micros, or they buy quality from Texas Instruments.
Re: intel isn't dead (Score:1)
If it wasn't dead we would have now processors running at 40GHz at 0.07V supply voltage. That's by extrapolating the golden age of technology scaling (0.5u->0.13u). The root causes for the slow down is the transistor gain and subthreshold slope which prevent us from reducing the supply voltage further. High voltage means high current, relatively large routing parasitics, high power and low speed. OK for memories but not CPUs. There were some improvements along the way but 0.7V is still a far cry from 0.0
Moore's law is not dead (Score:1)
It just smells that way.
Re:Moore's law is not dead (Score:5, Insightful)
Shrinks no longer mean the cheaper die that they once meant due to the more die/wafers. The equipment and processes that need to be procured and developed have costs that are high enough that they not just eviscerate any savings due to the higher output, but also end up costing just as much, if not more. Only reason to shrink now is for greater performance or more power savings (due to decreased Vdd levels).
So yeah, Moore's law is very much dead: you don't get half the price every 18 months anymore.
Re: (Score:2)
Moore's Law was never exclusively about Dennard scaling. It only concerns the cost per transistor no matter how that is achieved. So it includes scaling, greater total area, and packaging improvements.
"That's basically already working" (Score:3)
And with that Keller broke the bullshitometer.
Re: (Score:3)
"Jim Keller, a semiconductor rock star..."
Perhaps we stop reinterpreting things, maybe? (Score:5, Insightful)
Re: (Score:2)
Re: (Score:3)
To be fair, on behalf the Bible, I'll guarantee your proposed methodology will fail personally and permanently within a few decades.
Re: (Score:3)
The New Testament basically replaced all the laws from the Old Testament, and it has been going like that for a long time.
So it is a good analogy. We need a new law, but we'll still have to learn about Moore's Law in history class, and whenever people aren't paying attention some pundit will slip in some deprecated bullshit right from Moore. Just like they do with the bible.
Re: (Score:2)
The New Testament basically replaced all the laws from the Old Testamen ... and I wager: not for Christians either.
It did not for Jews or for Muslims
Re: (Score:2)
Here you mean Christians in tribal allegiance only, who idolize the Christ guy as a symbol of their collective power to repeal Roe v. Wade, so long as you can simply go about your day ignoring every damn thing he ever said, or is recorded as having said, or is construed as having said by moral sages of the Early and High Middle Ages so pleasing to God that he helped jiggle their quill elbows.
Jesus could equally well serve as a tri
Re: (Score:2)
Muslims split from Christians, so no you're wrong. Learn some history.
sounds like (Score:2)
sounds like "Moore's law is running out of gas" talk
so maybe two more generations to reach max 2D density, and I'm not believing they can "stack 50 layers" over something with the area of a CPU with decent yield, the probability of a fatal flaw becomes too big.
We'll be needing a radically different kind of tech than silicon semiconductors come 2023 - 2025
Re: (Score:2)
so maybe two more generations to reach max 2D density, and I'm not believing they can "stack 50 layers" over something with the area of a CPU with decent yield, the probability of a fatal flaw becomes too big.
Meh, they seem to do well with stacking NAND layers for SSDs so I don't think that's the problem. But they're already having a hard time spreading the heat from a single slice, I don't see any way they could keep a cube within reasonable operating limits. Maybe if they tried an entirely different mode of operation where you have tons of cores running at very low speed.
Re: (Score:2)
The problem with 3D stacking is heat dissipation.
Doing 2D in 3D ... (Score:2)
Re: (Score:2)
but they're speaking of short paths between components into the z axis, this is not the same as plugging things into a common bus
Re: (Score:2)
the physics and metaphysics (Score:1)
Cool! Intel is going to make magic processors!
Re: (Score:2)
Or at least magic words that can mean anything.
Intel must be scared shitless (Score:3, Insightful)
Holy shit, the bullshit from Intel just keeps getting worse. There was a time when their BS wasn't nearly so bad (early 80's). Itanium and Netburst and Athlon FUD were terrible but this is some next level horseshit.
Any halfway competent engineer can plot the number of gates per chip. It was bad enough when gate leakage current put a clamp on clockspeed. It was even worse when decreasing feature size meant decreasing speed. But that they had to roll out someone of the competence of JK to try and keep the orchestra playing while the investors run for lifeboats is scary.
I truly hope that we find a better family than CMOS, but CMOS has been AMAZING. There are other technologies out there but Intel has delayed the shrink from 10nm how many times now? Comhine that with the almost monthly significant security flaws and it's not just like they drove the ship into the iceberg, it's like they drove the ship into the iceberg and tried to plug the holes with gelatin, then pasta and finally bean paste.
Re: (Score:2)
eh, Intel's 10nm is the same as the competitors (samsung, TMSC) 7nm density (both of which are marketing terms anyway)
Intel has the highest density too.
A better article about the "Silicon 100 Summit" (Score:3)
https://hexus.net/tech/news/in... [hexus.net]
"Teeeechnically" (Score:2)
-Self proclaimed CEO of Intel, living out of a cardboard box next to the defunct companies former headquarters.
Bullshit! (Score:2)
Moores law is alive and well. (Score:2)
It just isn't alive and well for general purpose CPUs.
It is killing it over in video card land, however.
It was misstated. (Score:3)
if you think so you're stupid (Score:2)
'Moore's law is not dead but if you think so you're stupid
Our marketing department is not dead but if you think so you're stupid
That's not Moore's law. (Score:2)
"Moore's law is relentless," he added, referring to the 54-year-old assertion by a former Intel CEO that the number of transistors that could be fit onto a silicon chip would double on a predictable schedule.
Moore's law states that the number of gates in the processor will double on a predictable schedule. It's not about how many you can physically fit on a wafer. It's about how many gates there will be in the finished CPU product. They don't even have to be on the same die! Using Chiplets (ala AMD) lets Moore's law proceed apace without having to fit all the transistors onto one surface.
The above quoted text from the summary came directly from TFA itself, so you can blame the author, and/or the editor at Wired
Re: (Score:2)
So a double socket motherboard means that Moore's law is still applicable as they're both part of the same CPU product.
Well, no. In theory, if the two chips were not designed to be able to work alone but only as a team, you might reasonably call that one CPU. But nobody does that. If you're going to go to all the trouble to put it into another package, you might as well make it be able to work independently. In fact, making it work independently simplifies the packaging and the signaling, so that's something you'd do naturally.
The closest thing to that we've really had was math coprocessors. They were on the board separatel