[kurthr]

I wish this wasn't so laughable. What this really shows is that IBM has high margins for their chips, and a long standing tolerance for absurdly low yields and long wafer turn times. That they announce "production" before any of the other fabs is more likely do to PR needs than real technical advancement. https://arstechnica.com/gadgets/2015/07/ibm-unveils-industry...

EUV has been available for years and no doubt TSMC and SS have their own EUV test chips at 7nm (comparable to Intel 10nm), but the EUV equipment business will be validated by large orders and $ spent, not on prototype silicon. We aren't there yet (but probably will be in a year or so- they've gone from 100W to 150W in the last 9months and need to hit 200-250W).

Will we have 1-3nm transistors? Yes. Will they be commercially viable? Probably No.

Moore's law ends when the CFOs decide it's not worth building another multi-$B factory on schedule based on net expected return... which already happened over the last 4 years. Sorry, downvote at will.

[jackals]

That 200-250W is just just for minimum commercial viability, and that's the best that can be hoped for with the current drive lasers. Industry calls for four time that power, i.e. close to 1kW for sustainability.

[davidf18]

I suspect that one way or the other the military and spy agencies will be funding the further development for national security and weapons reasons. It may not be known, but they are (or will be) funding either directly or indirectly.

[Animats]

The US military hasn't been a significant driver of electronic technology since the 1980s. They just don't have the volume to affect the technology path. Military electronics tends to lag commercial, because the life cycles are longer. The USAF's big electronics problem is obtaining supplies of obsolete parts.

The pure military stuff tends to be in the sensor space. DARPA has some projects to build much better accelerometers and gyros on ICs. They want cheap inertial navigation to back up GPS.

It's possible to reach much higher densities by writing an IC with an electron beam. This is slow, but useful for one-off jobs. DoD is known to be using that.[1] "DoD foundries make a wide variety of custom chips in small quantities - the exact opposite of commercial practice." Interestingly, the USAF is doing this partly because they don't trust off-the-shelf chips not to have "backdoors".

[1] http://www.multibeamcorp.com/PR_MB20170525.htm

[filereaper]

IBM's previously owned East-Fishkill Foundries have long been part of the DoD Trusted Foundries list. [1] (Slide 8)

IBM split off and kept the R&D part of the foundries and sold the manufacturing aspect to GlobalFoundries.

[1]http://www.acq.osd.mil/se/briefs/2016-TrustPolicy-Baldwin.pd...

[undersuit]

>Will we have 1-3nm transistors? Yes. Will they be commercially viable? Probably No.

Why not? We can only stack our layers so much, eventually the 1-3nm range of transistors will become useful to give technology another squeeze before having to learn how to layer a bizarre number of layers.

[microcolonel]

28nm is still more expensive today than 40nm ever was, twice as expensive, per wafer, as 40nm currently is. The size of the elements is still somewhat decreasing, but the price per transistor is actually increasing at this point.

[kurthr]

Although there is hope that EUV will help control design rules, the current expectation for developing a 10nm design is in the $200M range. It's expected to double at 7nm and again at 5nm. Try amortizing that over anything but an iPhone, Samsung, or PC volumes.

[undersuit]

I'm just arguing that lacking an alternative to silicon transistors, we'll hit limits in die stacking and architecture optimizations that will make < 3nm lucrative for some applications.