Hacker News Reader

Comment by baq | Hacker News Reader

When you look at what it takes to successfully manufacture chips at these feature sizes you start to believe in the anthropic principle, or maybe I should call it semiconductoric. Veterans in the field can hardly believe its at all possible.

> Why just twice as fast? Once you know that something is doable, it works and the rough direction things can get done really really fast.

I agree on the first part. The second part is full of details. The mirrors have to be just right. The lasers have to be just right. The timings have to be just right. There's a million other things that have to be just right and the path to get there may not be very parallelizable.

View on HN

Sure its still hard but unlike those who did it first, those coming after don't have to guess as much and spend resources on dead ends.

This is not Taliban getting into semiconductors, this is Japan who is already into semiconductors but lost its edge in this particular field but its leading in some other fields. Therefore they already have a considerable know how on mirrors and laser that they will have to improve on. IMHO the catch up will be much quicker and much cheaper than what the Europeans and Americans spent on to explore the path leading to what we have today.

OTOH you have Intel which never stopped R&D and they still have lots of trouble catching up; arguably they've been trying for almost a decade now, five years if you're being generous. (And they buy tools from ASML! Which other posters explain will also be the case here.)

IMHO Intel's case is more nuanced, they were milking their position until Apple did something extraordinary with TSMC. They are also a giant corporation who has to re-organise and restructure, therefore can't move as fast as a newcomer. Intel's case is a business case, more than an engineering case. Just as Nokia.

The point still stands: they have the infrastructure, the experience, the logistics and the supply chain all secured and they're still at least a couple years behind if not more. It's extremely optimistic to say that it's going to be easier to start from scratch.

Could be, even engineering projects with well defined requirements and a mature tech sometimes fail.

> they were milking their position until Apple did something extraordinary with TSMC

I mean, that's true from a business perspective.

But they were trying very hard to make the next process node work. They went from a process upgrade every other design generation (the famous "tick-tock") to just tock-ing at 10nm because they couldn't get the next process generation working.

That was very much an engineering problem.

Intel was receiving continuous beatings from AMD (Ryzen/Epyc) a few years before Apple came up with M1. 2016/2017 vs 2020. I don't think Apple and Intel are in much competition: Intel doesn't make laptops and lacked Apple's moat when they did, and Apple doesn't sell (amd64) server CPUs.

>Sure its still hard but unlike those who did it first, those coming after don't have to guess as much and spend resources on dead ends.

In many domains they do. Once the original creators are retiring or not involved, there are tons of tacit knowledge at every step of the way, from the high level abstract design to specific quirks of some manufacturing equipment, that's very hard to build back.

Catching up is usually easier than pushing the frontier.Once you know that it's possible, it becomes a matter of iteration.

So it will take less than sixty years. Yes, I can believe that.

There's a lot of FUD around the subject too. Take naysayers with the same degree of skepticism you take the optimistic folk.

Judging from the actual industry state, and how many have been left behind in that domain, I'd rather believe the naysayers.

Until now it's just an announcement and some money thrown at the problem, so less than vapor.

Reply on news.ycombinator.com