Ask HN: So, did Moore's Law die or not? | Hacker News Reader

Ask HN: So, did Moore's Law die or not?

5 pointsuejfiweun2 comments

In the 2010s there was a lot of doom and gloom about Moore's Law being dead, and the need for new paradigms to speed up computers in the future. For instance, the following article from Technology Review:

https://www.technologyreview.com/2016/05/13/245938/moores-law-is-dead-now-what/

However, it seems that this may have changed recently. ASML's EUV lithography machines have been said to be able to reignite Moore's law. It's visible from a number of sources:

https://www.google.com/search?q=asml+euv+moores+law

However, there's a lot of hype and bullshit out there, and I can't tell what's legit or not, so I turn to the knowledgeable people of HN. Is Moore's Law actually dead? Are these EUV lithography machines really such an innovation that they have extended Moore's law? If not, then how DOES computing hardware get faster from here, and what does the answer mean for programmers?

Moore's law is about chip complexity, not speed. Generally, we're getting down to feature sizes so small that quantum tunneling and cosmic rays flipping bits are issues.

So, in that aspect, I think we're done with that ride. However, they are figuring out how to stack silicon dies directly on top of each other, with reliable interconnect methods. If this keeps up, you might see a few more doublings of the numbers of transistors in an IC package.

Most transistors in a computer are in memory, and most of those are waiting to be accessed. If we can build computing fabric that handles data directly, like an FPGA on steroids, there might be a few more doublings of performance to be gained there as well.

Disclaimer: I have not worked in the semiconductor industry, this is my old fart, making a reasonable guess about the world from all he's seen over the decades, opinion.

The pop-sci version of Moore's law, where old programs run magically faster on new hardware, has been dead for a while. Shrinking the size of a chip now has issues with current leakage, so the energy savings from shrinking aren't as easy either.

Moore's law is only an observation, so we'll see if these new devices extend Moore's law in hindsight. The way forward for faster programs does seem pretty clear though: more parallelism. That can be using vectorization, using more cores, or using coprocessors.

How programming languages address parallelism is very important now, much more important than back when the pop-sci version was true.

Reply on news.ycombinator.com