[yetihehe]

> I don’t understand how superconductors will make phones into supercomputers.

Currently transistors and connections use a lot of power just to transfer some bits, superconducting wires and maybe transistors would help.

> Also, how does this help fusion become possible?

Stronger magnets which you don't need to cool with liquid helium will help achieve better plasma confinement.

[p1mrx]

We've already passed the point where superconductors need liquid helium:

https://en.wikipedia.org/wiki/ARC_fusion_reactor

"The most probable candidate material is yttrium barium copper oxide, with a design temperature of 20 K, allowing various coolants (e.g. liquid hydrogen, liquid neon, or helium gas) instead of the much more complicated liquid helium refrigeration chosen by ITER."

[kickopotomus]

Superconductors may reduce power consumption but they don't do anything to improve transistor density.

[spaethnl]

Before the node-size race we also had the clock-speed race. Eventually it was common for processors reach 2-4ghz, and after that the clock speed gains stopped being practical because as you increase clock speed you also increase energy requirements and heat.

I think the implication is that clock-speed could start increasing again. It would probably require a completely new manufacturing process, but if we assume this superconductor is legit, perhaps an older process could manufacture it.

If so, maybe we could have (just spitballing here, I have no idea) 28nm super conducting CPUs that run at a 1thz instead of 4ghz. That would be quite an improvement over today's CPUs, even with fewer transistors, I think.

There are other losses and limitation in increasing clock-speeds aside from just resistive losses, but I think they are a significant part of the current bottleneck. Other losses involve transistor switching losses, and inductive losses but I don't really know the details, and I think those details change with superconductors.

[kickopotomus]

We stopped chasing clock speeds because of the physical timing limitations of gate and signal propagation. Not because of heat. Suppose you are using a 5GHz clock. Every cycle is 0.2ns. Light can only travel 6cm in that time. Electricity propagates a little slower through a conductor (and even slower through silicon). So if you are using some insanely fast clock, you are just wasting cycles waiting for signals to move across the chip.

[yetihehe]

Current processors are no longer synchronous, each part now works asynchronously and there can be several instructions waiting to be completed at once, average Intructions Per Clock are already over 1, so there is no problem waiting a little more for signal to propagate.

[l33t233372]

That’s really interesting. It seems like this technology could be huge