The take-away
- Gordon Moore, the co-founder of Intel, predicted in 1965 a doubling every year in the number of components per integrated circuit. This rhythm is now actually slowing down.
- Future solutions embrace multicore structures: multiple small pieces of silicon that are specialised for specific tasks will be built together like Lego-blocks.
The rate of progress since the first silicon transistor* in 1947 has been enormous, with the number of transistors on a single chip growing from a few thousand in the earliest integrated circuits to more than two billion today. Integrated circuits are important in the running of numbers of tiny transistors present in the electronic components. Carried by Moore’s law (a reference to Gordon Moore, the co-founder of Intel, who expressed it for the first time in 1965), that predicted the doubling density of transistors on a microprocessor every two years, these integrated circuits grew quickly.
Nevertheless, electronics based on silicon now face a challenge: the latest components in circuits measure just 7nm wide. By reducing the size further, its behaviour becomes unstable and difficult to control. Meeting this challenge may require rethinking how we manufacture devices, or even whether we need an alternative to silicon itself. EPFL’s expert Andras Kis explains how the industry will face these new hurdles regarding silicon.
Technologist: Why has silicon been the most suitable material for the development of microprocessors?
Andras Kis: Silicon has a good combination of interesting electronic properties. First of all, there is a huge amount of it. We can find silicon for example in stones or sand. And then, it’s easy to process or modify it for electronic purpose. It can also be easily oxidized, therefore optimizing the properties and the components of the transistors.
T: How has Moore’s law enabled technological progress in recent years?
AK: In the previous decades, Moore’s law helped to set the tempo for the industry. But now it is no longer the case because the whole the industry is slowing down. And we can see some problems.
T: What kind of problems?
AK: All the problems are linked to Moore’s law. To follow the tempo, you need to take all the components of a transistor and shrink them to be smaller, working along the same principles but being capable of working faster. It is necessary to regulate the heat dissipation to avoid a malfunction. Then there are also some fundamental technological limits. When you make the transistors smaller, it gets harder to control them, and they become less efficient. So basically, there are some problems with the dynamic power and the static power of the transistors.
T: So it’s all about the size?
AK: In average the industry is stuck at 14nm. Currently, the transistors are small and comparable in size to large molecules or proteins. Intel is still stuck at 14 nm but recent iPhone’s have a 7nm transistor size. Other companies try to produce some 10nm or below. If you want to go further, you can’t go much below 2 or 3 nm. We are then running out of materials and atoms to do it. So it’s the physical end of it.
T: Are we at the end of doubling the density of transistors on a microprocessor every two years?
AK: The industry is not at the end of this era. However, we may well be in something like five years. It could be possible to shrink a little bit the transistors, but not much further. And then there is the economic limit. New processor technologies often require new production facilities and their price increases as the transistor size shrinks. Instead of doubling the density, companies want to find clever ways to go around this problem, such as imaging new architecture for the processors.
One thing that is already being used is the multicore architecture. This can be taken one step further. Instead of carving new processors from silicon as single chips, semiconductor companies assemble them from multiple smaller pieces of silicon – known as chiplets – that will be specialised for some tasks and then will be built together like Lego-blocks.
T: What are the alternatives to silicon for chips?
AK: The scientists thought about the germanium in the beginning. This material had better electronic properties, but it doesn’t have a natural property of oxidising which makes production using silicon easier. There are many alternatives but the industry is not ready to exploit them. Alternatives such as diamond and graphene are not effective. Graphene is not a semiconductor so you couldn’t switch off the flow. Diamonds are isolators so electric components won’t work with it. Carbon nanotubes could be an alternative, but there is a fundamental problem related to producing them with controllable properties.
T: Can or will silicon disappear from the chips in the future or will it still have a role to play?
AK: Silicon is not dead and probably never will be. The industry is more focused on doing better with what they have instead of looking for other alternatives. In terms of profitability and improvements, growth will be less important than before.
T: Does the new generation of microprocessors represent the potential to build a new industry in Europe?
AK: Currently, Europe can’t compete with the United States, China or Asia. All the big players of the semiconductor and microprocessor industry are there. What Europe could do is to leapfrog its competitors by searching and developing new technologies or discovering new materials. It is a risky strategy but if Europe wants to be, one day, a leader, it is probably the only way to do.
T: Which applications of the future will require particularly powerful microprocessors?
AK: The new microprocessor’s architecture could improve the efficiency of the Internet of things or artificial intelligence devices. In this scenario, Europe, and more precisely Switzerland, could take advantage of it in the watchmaking industry. Both could take advantage from local expertise and developments of new materials for electronics.
*A transistor is a semiconductor device integrated in a processor