1965Gordon Moore notes that the number of components per chip has doubled every year
1975He revises it to a doubling about every two years
DensityMore transistors per area means more function and, historically, a lower cost per transistor
Node names"7 nm" or "3 nm" became marketing labels, not a measured length on the chip
Moore's law is an observation about economics and density, not a law of physics.
THE CURVE
Fifty years on a log plot
A straight line on a log chart is steady exponential growth. From about 2,300 transistors in 1971 to over 100 billion on the biggest chips of the 2020s is a factor of roughly 50 million, or about 25 doublings.
Twenty-five doublings is what turned a calculator chip into a pocket computer.
DENNARD'S RECIPE
Shrink everything together
L, tox, V ÷ κ → power per area unchanged
In 1974 Robert Dennard showed that if lengths, oxide thickness and voltage all shrink by κ while doping rises by κ, each transistor gets faster and cheaper in power, and a chip's power per area stays constant. For about thirty years that meant every generation was both denser and faster, at no extra heat.
The voltage has to scale too. Without it, the heat explodes.
THE POWER WALL
When the free ride ended
Voltage floorCutting Vth further would blow up the subthreshold leakage, so VDD stalled near 1 V
HeatPower per area started rising with each generation, around the mid-2000s
Clock speedStalled around 3 to 5 GHz, and designers turned to many cores and special accelerators
New leversStrain, high-k metal gates, FinFETs and nanosheets (Part 4) keep improving each generation
Shrinking stopped being free, so engineers started reshaping the transistor itself.
CHECK THE IDEAS
Moore, Dennard or the wall?
Moore counts transistors, Dennard held the heat constant, and the wall is what came after.
UNIT 10 STUDY COMPLETE
📉
Ready for the Fab Challenge?
You've followed fifty years of shrinking, and seen the point where shrinking alone stopped being enough.