Gordon Moore’s 1965 observation (revised in 1975) that the number of transistors on a chip doubles roughly every two years. It is an economic and manufacturing trend, not a law of physics, and it held for decades on the back of lithography.
What is actually slowing
Two things are often conflated and should be kept apart:
- Density is still improving. New device structures (GAA nanosheets, CFET (Complementary FET), Backside Power Delivery (BPD)) keep packing more transistors in.
- The economics and the energy stopped paying. Cost-per-transistor has flattened below roughly 7nm, the number of foundries at the leading edge has collapsed to a handful (see Sovereignty Semi Bifurcation), and the per-watt gains ended when Dennard Scaling broke.
So “Moore’s Law is dead” is the wrong headline. Transistors still shrink; what stopped paying is the cost and power curve. EUV and its successors keep density moving but do not restore the free lunch (see Beyond Euv Not In 2020S).
Why it matters here
Once density gains no longer translate into cheaper, cooler compute, architecture has to do the work that process used to. That is the opening for “post-Moore” approaches, and the binding constraint that those approaches target is data movement and the The Memory Wall, not transistor count.