Gordon Moore was an engineer who wrote a short article in 1965 about how many components could be packed onto one chip. He noticed the number was doubling on a regular schedule and expected that to continue. He later helped found Intel. The pattern he described was named after him.

The cheap half of it is. Transistors still shrink, but slowly, and the cost of each one has stopped falling at the leading edge. Chips carry on improving through better design, packaging and specialised blocks. So the old schedule is broken while progress continues.

does 3 nanometre mean the transistor is 3 nanometres wide

No. The number is a name for a generation of manufacturing rather than a measurement of any feature on the chip. Two companies using the same number can build quite different transistors. To compare fairly, look at how many transistors fit into a square millimetre.

why did processor clock speeds stop rising

Because power stopped falling as transistors shrank. Smaller transistors leak current even when switched off, and leaked power becomes heat. A chip that cannot lose its heat cannot run faster, so clock speeds settled at a few gigahertz. Makers added more cores and special blocks instead.

What is Moore’s Law, and is it over?

Imagine a shop that sells maize flour. Every two years the same money buys you a bag twice as big. Nothing else changes. The shop is the same and the flour is the same. Only the amount doubles.

You would plan your life around that shop. You would put off a big purchase, because next year is cheaper. Whole businesses would grow up assuming the doubling carries on.

That is what happened with transistors for decades. Moore’s Law is the name for the pattern. It is worth knowing what it really said, because most of what people repeat about it is wrong.

Where the idea came from

In 1965 an engineer called Gordon Moore wrote a short article about the young business of putting several components on one piece of silicon. He counted what had happened so far and drew a line forward.

The count had been doubling about every year. He guessed it would keep doing so for another decade. Ten years later he looked again and changed the figure to a doubling roughly every two years. Moore went on to help found Intel, and the industry took his line as a target.

You will often hear the figure given as 18 months. That version came from other people in the industry, and it mixed in speed as well as counting. The number Moore himself settled on was about two years.

What Moore actually said

Read the claim closely, because it is narrower than most people think.

Moore was counting transistors on one chip, at the size where each transistor was cheapest to make. He was making a point about money. Past a certain chip size, faults creep in and the cost per transistor climbs again. His law is really about that sweet spot moving year by year.

Three things he did not say are worth listing.

He did not say chips would get twice as fast every two years.
He did not say this was a rule of physics. It was an observation about an industry.
He did not promise it would last forever. His first article looked only ten years ahead.

That last point matters. Because the industry treated the line as a plan, it became partly self-fulfilling. Companies set budgets by it. Suppliers built machines to hit it. The forecast pulled the work along behind it.

Why it held for so long

Doubling the count means shrinking each transistor. A transistor is a tiny switch built from semiconductor material. Shrinking it used to give three gifts at once.

You got more transistors in the same space. You got faster switching, because a smaller switch fills with charge sooner. And you got less power for each switch, because the voltage could come down too. The chip could run quicker without running hotter.

That third gift had its own rule, worked out in the same era. Engineers call it Dennard scaling. Power per square millimetre stayed flat as the parts got smaller. This is the quiet engine that made the whole thing feel like magic. Chips got denser, faster and cooler on one schedule.

What broke

The third gift went first. Around the middle of the 2000s, voltage stopped coming down in step with size. Transistors began to leak current even when switched off, and leaked power turns into heat.

What happens when a chip cannot shed its heat? It cannot run faster. That is why clock speeds stalled at a few gigahertz and have barely moved since. A desktop chip from twenty years ago runs at a broadly similar clock to one sold today. Almost every gain since then came from somewhere else.

Makers responded by putting several cores on one chip instead of one faster core. That helps when software can split a job into parts. Plenty of software cannot.

The second thing to break was cost. Shrinking now needs machines and steps so expensive that the price of each transistor has flattened at the leading edge. Cheap transistors were the whole point. When they stop getting cheaper, the law stops mattering, even if the count still creeps up. You can see why inside a chip fab, where each new generation of printing gear costs more than the last.

Why the nanometre numbers mislead

A chip today may be sold as a 3 nanometre or 2 nanometre part. Here is the awkward truth. No part of that transistor measures 3 nanometres.

These numbers were once real measurements of a physical feature. Over time they drifted into being names for a generation of process, kept alive because buyers expect the next number to be smaller. Two makers with the same number on the label can build quite different transistors.

Density is the honest measure. Ask how many million transistors fit into a square millimetre. That figure still rises. It rises more slowly than the labels suggest.

So is Moore’s Law over?

Split the question in three, because the honest answer differs for each part.

Are transistors still shrinking? Yes, slowly, and with new shapes. Designers now wrap the control gate all the way around a thin sheet of silicon to hold the leakage down. The steps between generations are smaller than they once were.

Are they getting cheaper? At the leading edge, mostly no. That half of the law has stopped.

Are chips still getting better? Yes, and quickly. However, the gains now come from design rather than from shrinking. So the plain answer is this. MOORE’S LAW AS A PROMISE OF CHEAP DOUBLING IS OVER. Progress in chips is not.

What is taking its place

Four ideas now carry most of the improvement.

Specialised blocks. A circuit built for one job beats a general one by a wide margin. That is why your phone carries an NPU for neural network work.
Chiplets. Rather than one big chip, makers build several small ones and join them in a single package. Small pieces have fewer faults.
Stacking. Memory and logic are now built upward as well as sideways, so signals travel a shorter distance.
Faster memory. Feeding the chip is often the real limit, so a lot of the gain comes from moving data less.

Notice the shift. The old game was making one thing smaller. The new game is arranging many things well.

What to check

Look up the clock speed of a laptop or phone from about ten years ago. Then look up the clock speed of a new one. The two numbers will sit close together, often within a few tenths of a gigahertz.

Now compare the core counts, and look for the special blocks listed in the spec sheet. The gap between those two comparisons is the end of Moore’s Law, written in plain numbers on a page you can check yourself.

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.