Moore's law has an economic component, i.e. "double transistors for the same price", which is falling apart faster than the density promises. New nodes are expensive and have massive fixed per-chip design costs.
Which makes soldered chips in mobile devices (and NUCs/small PCs) all the more insane. But it's good for business - people will just buy a new one anyway...
Hopefully socketed chips on desktops will never go away, maybe recycling/reuse businesses will become more common (manufacturers will try and stop that with some sort of locks, no doubt) or we'll see the return of sockets on laptops (I can only dream).
Clock speeds stopped increasing exponentially in the mid 2000s[1].
Transistor density keeps increasing exponentially.
Power per operation keeps decreasing exponentially.
Marginal cost per transistor keeps decreasing exponentially.
Cost to build a new fab keeps increasing exponentially.
The number of leading edge fab companies might have stopped decreasing exponentially, but since we're down to 3 it's hard to tell.
[1] Frequency wasn't mentioned in Moore's 1965 paper but the phrase "Moore's Law" wasn't coined until 1975 after Dennard published his results and the phrase has historically always been ambiguous as to exactly what was doubling periodically when Dennard Scaling still held.
To indirectly answer your question. There is nothing in the Semiconductor sector that measures at 2x every 24 months. Whether that is Performance, Cost or Transistor Density.
And to throw another question, what exactly is Moore’s law? Because in modern day journalism and conversation, the term "Moore’s law" means transistor improvement.
Sorry, just to be clear did Intel announce they would be moving to measuring stuff in angstroms, and that they first measurement would be 20 angstroms. Also known as.... 2nm.
You really have to hand it to that marketing department.
It's sort of interesting as a human tendency. You very rarely hear anyone in the industry refer to 0.25u as 250nm, but below 2 at 0.18u you do occasionally hear it referred to as 180nm. By the time you're at 130nm, it's quite rare to hear it referred to as 0.13 and at 90nm you never hear 0.09um. This wasn't some industry planned marketing thing... it just seems reasonable. It's convenient that there are angstroms as well, since they are close atomic sized (Silicon atom is ~0.2nm - lattice parameter 0.54nm).
Intel’s marketing dept was so cool in 90’s. Blue man group, bunny suit commercials, the jingle, Pentium ads. Even the USB inventor rockstar commercial was super cool in mid 2000’s.
Things went downhill after Paul Otellini left and BK was brought it. Paul missed the mark on mobile chips, but BK is the reason for Intel’s malaise in last 10 years. He destroyed the company much the same way that HP exec gutted out the company in early 2000’s.
I miss Pentium. What a cool name. I heard Marketing folks did an internal survey and it was suggested by an employee. Good judgement and brand acumen.
-- I'd be very curious to know who was running marketing at Intel in the 90s - agreed it was amazing - but I recall Otellini joining in the early 2000s? --
On the technology side, it lines up with the switch to GAA from FinFET. That 2nm/20a process is just a stopgap before 1.8nm/18a, and has more to do with 1.8nm than any of the old processes.
Seems TSMC is sticking with nm for their naming, even for 1.4nm.
22 comments
[ 2.2 ms ] story [ 56.9 ms ] threadSeems like Moore is still correct? Maybe with ARM & AMD? Is it doubling though or is the tail end starting to dwindle?
Hopefully socketed chips on desktops will never go away, maybe recycling/reuse businesses will become more common (manufacturers will try and stop that with some sort of locks, no doubt) or we'll see the return of sockets on laptops (I can only dream).
Transistor density keeps increasing exponentially.
Power per operation keeps decreasing exponentially.
Marginal cost per transistor keeps decreasing exponentially.
Cost to build a new fab keeps increasing exponentially.
The number of leading edge fab companies might have stopped decreasing exponentially, but since we're down to 3 it's hard to tell.
[1] Frequency wasn't mentioned in Moore's 1965 paper but the phrase "Moore's Law" wasn't coined until 1975 after Dennard published his results and the phrase has historically always been ambiguous as to exactly what was doubling periodically when Dennard Scaling still held.
To indirectly answer your question. There is nothing in the Semiconductor sector that measures at 2x every 24 months. Whether that is Performance, Cost or Transistor Density.
And to throw another question, what exactly is Moore’s law? Because in modern day journalism and conversation, the term "Moore’s law" means transistor improvement.
You really have to hand it to that marketing department.
Things went downhill after Paul Otellini left and BK was brought it. Paul missed the mark on mobile chips, but BK is the reason for Intel’s malaise in last 10 years. He destroyed the company much the same way that HP exec gutted out the company in early 2000’s.
I miss Pentium. What a cool name. I heard Marketing folks did an internal survey and it was suggested by an employee. Good judgement and brand acumen.
Seems TSMC is sticking with nm for their naming, even for 1.4nm.
https://www.realworldtech.com/intel-4/