The Unsustainability of Moore’s Law - by Charles Rosenbauer

Roughly every five years, the cost to build a factory for making such chips doubles, and the number of companies that can do it halves. 25 years ago, there were about 40 such companies and the cost to build a fab was about $2-4 billion. Today, there are either two or three such companies left (depending on your optimism toward Intel) and the cost to build a fab is in excess of $100 billion. Project these trends forward another ten years and you can expect a single factory to cost nearly half a trillion dollars, and the number of companies that can do it should drop to less than one.

人類これ以上 CPU 作るのムリなのでは、という気分になるが、

Highly parallel chips designed from the ground up for extremely high defect tolerance could perhaps be manufactured with a vastly lower-quality, and by extension vastly cheaper chip fab.

 I’m currently raising some pre-seed funds, aiming to build high-performance, high-efficiency, general-purpose CPUs designed from first principles. I’m also increasingly convinced that, with the right compromises and specialization, a startup-scale fab may be viable and is at least well-worth the experiment.

という求人への長いフリなのだった。Cerebras が似たようなことを言っていましたね。

100x Defect Tolerance: How Cerebras Solved the Yield Problem - Cerebras

字数の大半はチップ細微化の不可能性を長々と説明していて、そこは面白い。