Hacker News Reader

Comment by FlowingRiver | Hacker News Reader

Interesting. LogicFolding is something that seems so obvious in retrospect and yet somebody has to be the first to do it.

I did find it neat that it ends up actually reducing overall heat despite the multiple layer wafers as signals have less overall distance to travel as they move in layer space rather than across the chip.

View on HN

I suppose you could say it's an obvious thing to try but not to implement. I remember proposing something along those lines here >10 years ago but I didn't have a good comeback for how you'd address the heat dissipation problems, or enough know-how to justify trying to research it more seriously.

Pretty much, that was why I was surprised that power use declined doing this. The reduced of signal resistance has a greater off sets than the additional thermal layers.

But my shoot from the hip thoughts are that this might end up being used more for mid range chips than high end. higher clock frequencies would probably compound the thermal laying beyond standard resistance and buffers.

It moved the complexity from lithography to alignment

That is it, this is fundamentally a 7nm process leveraged potentially to a 1.4nm (claimed) scale.

Makes you wonder what is possible when applied this to much smaller node sizes. Interesting times ahead.

It seems fundamentally easier to do the alignment in the lithography step as you only have get it correct once in the mask instead of correct for every chip every time.

But how can they patent this, as 3D chips have been built for decades already?

I mean, it is nice that they found out some properties (lower heat than expected), but you cannot patent that, I suppose.

It is probably like a lot of patents, some very specific details on design and the technology of interconnects between layers.

Chips have been multiple layers since the beginning practically but it was also laid out in a somewhat 2D fashion of ALU here, FPU there, cache somewhere else, but now they are stacking functionality so that is isn't so discrete, in a 2D space it would get a little fuzzy.

It isn't mind blowing ground breaking stuff but it is very cool. Evolutionary not revolutionary.

Hardcore engineering vs. a novel scientific discovery.

Perhaps given the difficulty of acquiring and (using?) ASML machines, Huawei is looking for alternative routes to get there. Innovative nevertheless.

Reply on news.ycombinator.com