Is 7nm the Last Major Node?

Megalith

24-bit/48kHz
Joined
Aug 20, 2006
Messages
13,000
A combination of technical difficulties, narrowing improvements in performance, and increasing costs is making the future of scaling ambiguous: despite the inevitable onset of 5nm, many companies cannot find a compelling reason to go down to 7nm, and emerging markets are expected to cling to older nodes instead. The brute force approach to shrinking no longer works, and some believe that the entire semiconductor industry may require a reset.

…it’s simply getting harder to design, inspect and test chips at advanced nodes. Physical effects such as heat, electrostatic discharge and electromagnetic interference are more pronounced at 7nm than at 28nm. It also takes more power to drive signals through skinny wires, and circuits are more sensitive to test and inspection, as well as to thermal migration across a chip. All of that needs to be accounted for and simulated using multi-physics simulation, emulation and prototyping.
 
If it is then never sell a 7nm chip it'll become a collectors item :)
 
Ya ya ya, we'll find a way don't worry, there's plenty of smrt people working there.
 
"Scaling" has been difficult since the 8088 and even the Z80. It's going to break at some point. Let's be honest though, every generation is accompanied by cries of "DoOoOm".

Truth be told, the pattern will be broken by something better, not by its own difficulties.
 
Necessity is the mother of invention. Maybe we will push towards more effective use of the current silicon first. Mitigation of heat may be the stop gap measure we need.
 
Silicon is nearing it's limits. Other materials will have to be used.
 
I really dislike how the press tends to represent the issues at hand as we push forward with smaller nodes.

Design of smaller nodes is not the issue. We still have humans perfectly capable of designing products at smaller nodes, but the complication of the circuitry is a large burden. Our designs get more complicated every day, the node size goes down resulting in increased circuit density and complications due to sheer number of features being designed. Computer aided design will aid with this to some extent, but still isn't up to par with an expert designer tuning his own designs.

The biggest obstacle is getting those designs turned into lithography masks. We already design and manufacture products that exceed the capabilities of photo lithography. We got around this by implementing new patterning techniques using multiple masks to create features far smaller than that limit. Each one of those masks costs money though. Every additional masking step in a manufacturing process also increases the production time of the product, not to mention delaying the turnaround time between tape out and production testing while the masks are created and implemented. As feature size decreases, we have to keep adding more masks to the process, and alignment starts becoming a huge issue. A "tiny" offset of one of the several masks used in production might be acceptable at one node size, but could result in severe defects at a lower node size. This is why we already have chip manufacturers trying to find new ways to avoid adding even more masks due to cost of both time and capital investment in them.

We can push beyond 7nm without EUV litho, but enthusiasts aren't the ones making the choices that make this possible. The technology will only advance when it makes sense to the fabs from a cost investment/return aspect. So while the problem is represented as a technical limitation, money is the true hold up.

EUV isn't our fabrication Messiah. Ignore, for a moment, the issues with production time requirements that currently keep it from going mainstream, or the required downtime of cleaning the vaporization chamber once it's dirtied up with vaporized metal (required very often btw). EUV has a built in technological birth defect. The edges of EUV cuts can be very rough. That roughness isn't something we can currently control, as we haven't developed a way to restrict individual protons to a tight enough beam to prevent it from occurring. That roughness might appear minor at one node, but as the node size shrinks in EUV, the roughness stays the same size, meaning your minor roughness gets turned into cut walls that look like jagged canyons compared to the circuit traces themselves. We also experience some rather annoying issues arising from quantum physics when using EUV. Those photons have a very high energy, and they don't always stop where they hit the substrate due to quantum tunneling.

The true way forward is in the past. We had started developing X-ray lithography in the early 90s/late 80s, but the tech got shelved due to patent bullshit between the major financial backers of the x-ray litho development. they couldn't come to an agreement over royalties and profit sharing resulting from the tech. It's no coincidence IBM is such a big backer of EUV, they were also a major backer of x-ray litho before the other patent holders got hostile. x-ray litho also required a pretty damn large cyclotron to product the x-rays, but EUV will require enough space to basically require a two floor basement below the fab rooms for equipment, making the cyclotron size a moot issue now. X-ray litho can push to sizes that would require some huge advances in substrate technology as well.

So smaller than 7nm nodes require:
$$$$$$ investment from the fabs
More advanced computer chip design advances due to design complexities and densities
Taking EUV out behind the woodshed and putting it down humanely as I'm tired of it being promoted as the fabrication Jesus to solve all issues. Seriously, why push forward with a litho tech that's only going to work for a couple nodes before it's inherent problems stop it dead in its tracks.
Finding a way to breath life back into x-ray lithography. The feature sizes possible far exceed anything EUC can or ever will offer. A cyclotron can run continuously, a EUV vapor chamber has to be taken down very frequently for cleaning.

TLDR: $$$$$ = smaller than 7nm. Patent shenanigans are denying us the best way forward.
 
…it’s simply getting harder to design, inspect and test chips at advanced nodes. Physical effects such as heat, electrostatic discharge and electromagnetic interference are more pronounced at 7nm than at 28nm. It also takes more power to drive signals through skinny wires, and circuits are more sensitive to test and inspection, as well as to thermal migration across a chip. All of that needs to be accounted for and simulated using multi-physics simulation, emulation and prototyping.

Are we talking quantum superposition at this point??
 
Silicon is nearing it's limits. Other materials will have to be used.
They're approaching atomic limits with any material at 7nm. Other materials won't fix process scaling.

Other materials could maybe increase clock speeds or reduce power requirements or maybe even improve yields, so they're absolutely worth looking into, but they won't do a thing for process scaling
 
I think that worry was mentioned in the article 15 years ago.

And now we're almost there. You must not understand that we're getting to the point that we're only using a handful of atoms to make a transistor.

This isn't some stupid conspiracy theory. Traditional transistors can't keep shrinking the way they have. Sure we'll have alternatives, but they'll be hellaciously difficult and expensive. We're talking about the manipulation of matter at the scale of individual atoms and even a small variance will wreak havoc.
 
Keep in mind this is purely for digital circuits. Analog (power supplies, audio amplifiers, battery management, etc...) lags WAYYY behind, not because we can't make circuits smaller, but because the amperage and voltage requirements force larger structures.
 
Last edited:
I really couldn't care. With the current computing power that we have today everything is possible, the only reason for further miniaturization is to make biological machines (life), but silicon is definitely not the right path for that.
 
I would imagine (fantasy) micro integrated nano peltier pipes and massively stacked IC's. If you can't go horizontal, go vertical.
 
I really couldn't care. With the current computing power that we have today everything is possible, the only reason for further miniaturization is to make biological machines (life), but silicon is definitely not the right path for that.

Your argument is valid for most desktop users, but not for high end users, and especially not for high end enterprise or scientific workloads.
 
And now we're almost there. You must not understand that we're getting to the point that we're only using a handful of atoms to make a transistor.

This isn't some stupid conspiracy theory. Traditional transistors can't keep shrinking the way they have. Sure we'll have alternatives, but they'll be hellaciously difficult and expensive. We're talking about the manipulation of matter at the scale of individual atoms and even a small variance will wreak havoc.

Yeah you are right. I don't have to understand it. That's other peoples job.;)

C'mon boffins..work it out!!!!
 
I just remember seriously getting into the hardware game back in the mid- to late-90s and reading computer magazines that prognosticated that we would never be able to get past the .12 micrometer (120 nanometer) process level due to physical difficulty.
 
Back
Top