Steve Blank The End Of More – The Death of Moore's Law

A version of this article appeared for the first time in IEEE spectrum.

For most of our lives, the idea that computers and technology would become better, faster and cheaper each year was as assured as the rising sun every morning. The story "GlobalFoundries Stops All 7nm Development" does not sound like the end of this era, but for those who use an electronic device, it certainly is.

Technological innovation will take a different direction.

GlobalFoundries was one of three companies that manufactured the most advanced silicon chips for other companies (AMD, IBM, Broadcom, Qualcomm, STM and the Ministry of Defense.) The other foundries are Samsung in South Korea and TSMC in Taiwan. Now, there are more than two who are chasing the edge of the attack.

This is a big deal.

Since the invention of the integrated circuit about 60 years ago, computer chip manufacturers have been able to package more transistors each year on a single piece of silicon. In 1965, Gordon Moore, one of Intel's founders, observed that the number of transistors doubled every 24 months and that he would continue to do so. For 40 years, the chip industry has managed to achieve this goal. The first integrated circuits of 1960 had about 10 transistors. Today, the most complex silicon chips have 10 billion. Think about it. Silicon chips can now contain billion times more transistors.

But Moore's law ended ten years ago. Consumers simply have not received the memo.

No more Moore – Technological innovation at the end of the process
The chips are actually "printed", not with a printing press, but with the lithography, using exotic chemicals and materials in a "fab" (a chip factory – the factory where are produced fleas). Packing more transistors in each generation of chips forces the manufacturer to "reduce" the size of the transistors. The first transistors were printed with lines 80 microns wide. Today, Samsung and TSMC are striving to produce chips offering functionality of several tens of nanometers. This is a reduction of about 2,000 against 1.

Each new generation of chips that reduces the width of the line requires manufacturers to invest huge sums in new chip making equipment. While the first mills cost a few million dollars, the current plants – those that are pushing the forefront – exceed $ 10 billion.

And the explosive cost of the plant is not the only problem to pack more transistors on the chips. Each reduction in the width of the chip line requires more complexity. Features must be placed precisely at the exact locations of each layer of a device. At 7 nanometers, this requires up to 80 separate mask layers.

Moore's Law was an observation on process technology and economics. For half a century, he has driven the aspirations of the semiconductor industry. But the other limitation to packaging multiple transistors on a chip is a physical limitation called Dennard scaledas the transistors become smaller, their power density remains constant, so that the energy consumption remains proportional to the surface. This fundamental law of physics has created a "Power Wall" – an obstacle to clock speed – that has limited the frequency of the microprocessor to about 4 GHz since 2005. That is why the clock speeds of your microprocessor have stopped to increase with leaps and bounds 13 years ago. And why the density of memory will not increase at the speed we saw a decade ago.

This problem of continuous reduction of transistors is so difficult that even Intel, the leader of microprocessors and, for decades, the absolute reference in advanced technology, has had problems. Industry observers have indicated that Intel has encountered several obstacles on the way to next-generation models of 10 and 7 nanometers, and is now behind TSMC and Samsung.

This combination of spiral manufacturing costs, technological barriers, power density limits and diminishing returns is why GlobalFoundries has thrown the towel on new decreasing line widths. It also means that the future direction of innovation on silicon is no longer predictable.

It is the end of the beginning
Finishing putting more transistors on a single chip does not mean the end of innovation in computers or mobile devices. (To be clear, 1) the tip will move forward, but almost imperceptibly from one year to the next and 2) GlobalFoundaries will not stop, they will simply not be those who push the tip 3) Existing installations can make 14nm chips of the current generation and their expensive tools have been paid. Even the oldest plants at 28, 45 and 65 nm can make a ton of money).

But what this means is that we are at the end of guaranteed annual growth of computing power. The result is the end of kind of innovation we've been used to for 60 years. Instead of faster versions of what we used to see, device designers now have to be more creative with the 10 billion transistors they need to work with.

Let's remember that human brains have 100 billion neurons for at least 35,000 years. Yet we have learned to do a lot more with the same computing power. The same goes for semiconductors – we will find radically new ways to use these 10 billion transistors.

For example, new chip architectures are coming (multicore processors, massively parallel processors and silicon dedicated to automatic / artificial learning and GPUs like Nvidia), new methods of packaging chips and chips. memory interconnection, and even new types of memory. . And other designs advocate extremely low energy consumption and others at very low cost.

It's a brand new game
So what does this mean for consumers? First, high-performance applications that require very fast computing locally on your device will continue to migrate to the cloud (where data centers are measured by the size of the football field), thanks to the new 5G networks. Secondly, even though the computing devices we buy will not be much faster with today's standard software, new features – facial recognition, augmented reality, stand-alone navigation, and apps that we have not even thought of – come from the # 39, new software. using new technologies such as new screens and sensors.

The computer world is entering a new and unexplored territory. For desktops and mobile devices, the need for an "essential" upgrade is not a matter of speed, but a new feature or application.

For chip makers, all the rules are wrong for the first time in half a century. There will be a new set of winners and losers in this transition. It will be exciting to watch and see what comes out of the fog.

Lessons learned

  • Moore's law – doubling every two years the number of transistors that can fit on a chip – is over
  • Innovation will continue in new IT architectures, chip packaging, interconnects and memory
  • 5G networks will transfer more and more high-performance consumer computing needs to the cloud
  • New applications and hardware other than processor speed (5G networks, displays, sensors) will now generate consumer device sales
  • New winners and losers will appear in mainstream devices and chip vendors

Filed under: Technology |