The world according to Moore’s law, where the number of transistors that can be fabricated on an integrated circuit doubles every two years, is expected to keep marching on for some time ahead.
Intel recently presented their chip fabrication process technology plans out to the year 2022, where process nodes out to 4nm are expected.
This transistor revolution, with densities at near-zero cost each, is going to cause a shift in how the industry evolves, and how it defines itself.
It is different going forward because it used to be that once an application was written and then compiled to run on x86, its performance scaled up automatically, with Moore’s law delivering faster CPU clocks, more local memory, faster DRAM, etc.
For single-threaded applications, for single threaded processor designs, Moore’s law was an enabler. And that worked extremely well for decades.
But the performance of office applications like Excel, Word, email, Explorer and Powerpoint reached the level of good enough on x86 chips years ago. The transistors used to run them are being commoditized.
An Atom x86 CPU today with just ~ 50M transistors on a 25mm^2 die (45nm process) handles these basic functions more than good enough.
So what value or experience can a system deliver when it can use a chip with more than 100 billion transistors that costs as much to manufacture as a pair of cheap running shoes? It will be able to run most of today’s applications and the operating system in just a tiny corner of the chip.
The future of value-added non-commoditized semiconductors belongs to applications that need to be solved with parallel programming. Only these can employ hundreds of billions of transistors, that enable tens of thousands of local compute cores, and scale to massively networked virtual parallel compute networks.
Moore’s law is no longer the friend of x86.
Those days are gone forever.