Transistors Will Stop Shrinking in 2021, Moore's Law Roadmap Predicts (ieee.org)

Moore's Law, an empirical observation of the number of components that could be built on an integrated circuit and their corresponding cost, has largely held strong for more than 50 years, but its days are really numbered now. The prediction of the 2015 International Technology Roadmap for Semiconductors, which was only officially made available this month, says that transistor could stop shrinking in just five years. From an article on IEEE: After 2021, the report forecasts, it will no longer be economically desirable for companies to continue to shrink the dimensions of transistors in microprocessors. Instead, chip manufacturers will turn to other means of boosting density, namely turning the transistor from a horizontal to a vertical geometry and building multiple layers of circuitry, one on top of another. These roadmapping shifts may seem like trivial administrative changes. But "this is a major disruption, or earthquake, in the industry," says analyst Dan Hutcheson, of the firm VLSI Research. U.S. semiconductor companies had reason to cooperate and identify common needs in the early 1990s, at the outset of the roadmapping effort that eventually led to the ITRS's creation in 1998. Suppliers had a hard time identifying what the semiconductor companies needed, he says, and it made sense for chip companies to collectively set priorities to make the most of limited R&D funding.It still might not be the end of Moore's remarkable observation, though. The report adds that processors could still continue to fulfill Moore's Law with increased vertical density. The original report published by ITRS is here.

Comments

In other words, Moore's law will continue

[acoustix]

We hear the same bullshit every 2 years. Moore's law has nothing to do with the SIZE of the transitors. It has to do with the number of transistors on the chip and, to a lesser extent, the density of the transistors. Arranging the transistors vertically and horizontally will allow the law to continue.

Re:In other words, Moore's law will continue

[tlhIngan]

In actual reality, most of Moore's law has stopped 6-8 years ago. Just compare a midrange CPU from back then with one from today in actual performance. Not so much of a difference.

And Moore's law has never been about performance. Just transistor density.

General logic like what makes up the computation portion of a CPU don't need Moore's law at all - the transistor density is so low, they generally fab tons more transistors that sit around doing nothing. This way when a bug is found, they can revise the metal layers and put some of those spare transistors to use. This easily saves half of the masks they need to re-do, so at a $100K each per mask, it could mean spending under a million dollars over a couple of million dollars.

Instead, Moore's law is closely followed by memory manufacturers, because the denser the transistors, the more memory available. This applies for bot flash and RAM - 6-8 years ago you probably had a machine where 8GB of RAM is considered high end for a PC. Nowadays, 64GB is often the high end for a PC. As well, 120GB of SSD storage was considered luxury. Nowadays, you can get 480+GB for less money than that 120GB SSD, and it's not just SATA2, but SATA3. Or even PCIe.

There are two things in IC fabrication - you have "pin limited" and "silicon limited" designs. Similar to how in programs, you have "I/O bound" and "CPU bound". "Pin limited" ICs mean the overall functionality and design is limited by the number of pins your package supports. Even with 1000+ pins in modern packages, that still limits what you can do. Whereas in silicon limited designs, the limit is how much area your design takes up - more area means higher costs due to less dice per wafer, as well as higher chance of die defect. Memory devices are area limited - the pin counts of modern RAM and flash devices is low, but the area is high. Moore's law increases the storage density so you can have more storage in the same area.

It's why SSDs have a hard time catching up to HDDs (at least with raw storage) - SSDs improve with roughly Moore's law. HDDs have been improving (storage wise) faster.

In fact, most of the millions and billions of transistors in your CPU aren't used for logic processing - probably 90% of those transistors are memory related - caches, on board memory, etc. Because those are dense. SRAM cells are typically 6T (6 transistor) designs, so if your CPU has 16MB of cache, that's 96M transistors right there and then just in the storage array. Even more fascinating is that those 95M transistors will probably occupy less area than one of the major processing units on the same chip which may be only 1-2M transistors.

Re:What about heat dissipation

[Areyoukiddingme]

Admitted, I'm just another guy debating a topic I don't know much about, but won't layering components on top of each other result in massive heating issues? I mean, the heat from each layer has to go somewhere, right?

Yes. That's why IBM, among others, has been fabricating cooling capillaries into chips. They're experimenting with inter-layer liquid cooling through tubes just a few microns wide, imitating physical shapes found in the smallest of blood vessels to keep the fluid moving.

Re:Molecular computing

[K. S. Kyosuke]

I love Stanislaw Lem's concept of "the last generation computer". It may have been tongue-in-cheek in the time he wrote Fiasco (when the much-hyped "fifth generation" was "the Next Big Thing") but the concept feels increasingly relevant these days.

"This was a computer of the 'last' generation--last, because no other could have greater calculating power. Limits were imposed by such properties of matter as Planck's constant and the speed of light. Greater calculating ability could be achieved only by the so-called imaginary computers, designed by theorists engaged in pure mathematics and not dependent on the real world. The constructors' dilemma arose from the necessity of satisfying mutually exclusive conditions to pack the most neurons into the smallest volume. The travel time of the signals could not be longer than the reaction time of the components; otherwise, the time taken by the signals would limit the speed of calculation. The newest relays responded in one-hundred-billionth of a second. They were the size of atoms, so that an actual computer had a diameter of barely three centimeters. A computer any larger would be slower. The Hermes' computer did indeed take up half the control room, but that was for its peripherals: decoders, hierarchic assemblers, and so-called hypothesis generators, which, with the linguistic modules, did not operate in real time. But decisions in critical situations, in extremis, were made by the lightning-swift core, which was no bigger than a pigeon's egg."