Neural microchips will become real within the next two decades. We expect that by 2035, it will become the norm for most AI systems. 20-30 trillion (possibly upwards of 100 trillion) transistor architectures using 3D stacking and photonics will form part of the artificial brains we expect to see. Handling most of the light and quick work on system, while inference centers (the datacenters of today) will push 1 zettaflops by the end of the 21st century.
We suspect that 10 angstrom will be the absolute limit for silicon based chip components. Beyond this, new breakthroughs in photonics will need to materialize. The angstrom era of neural chips will mean huge advancements and open up the possibility of ASI, which together with 3D stacking will mean trillions of transistors, and 10+ exaflops will be commonplace.
