Law-abiding industries

A chemist made some startling predictions 40 years ago that have driven the semiconductor and electronics industries ever since. Katharine Sanderson met Gordon Moore

Forty years ago Gordon Moore wrote a magazine article that revolutionised the way we live today. We now take it for granted that our computers, mobile phones, home entertainment systems and other gadgets will become obsolete almost as soon as we have bought them. We expect the advances in processing power and technology to move on apace. And it seems that we can expect this progress to continue. But to Moore this has been no surprise. He saw it coming.   

Cramming more components onto electric circuits  turned out to be an almost prophetic article. In it Moore predicted that integrated circuits on silicon wafers heralded the future for the electronics industry. He also foresaw the widespread presence of home computers, and the future impact of mobile communications. Moore’s predictions have driven the semiconductor and electronics industries and the progress of the company that he went on to create: Intel.   

That article in Electronics spawned what we now know as ’Moore’s law’. In 1965 Moore noted that since they were introduced, the number of transistors on an integrated circuit had doubled about every two years. He predicted that this exponential trend would continue, increasing processing power and bringing an associated decrease in costs of electronics.     

Sticking to the law   

His predictions came true and the semiconductor industry looks set to keep his law alive for some years to come, with Intel planning to launch a processor called ’Montecito’ later this year - with 1.7 billion transistors on a single chip.   

In 1965, Moore was director of R&D at Fairchild Semiconductor, on the US east coast. Fairchild was making some of the first integrated circuits on silicon chips. Moore’s expertise was in chemistry, with a PhD in physical chemistry from the California Institute of Technology. ’I was a chemist at one time,’ Moore, now 76, told Chemistry World, ’but since then I think they’ve completely changed chemistry so that I don’t really understand it anymore’. He understood well enough then to leave Fairchild in 1968 and form Intel with Robert Noyce.   

Chemistry helped Moore choose the original materials for making integrated circuits: ’We used to say we were very careful when we picked our materials initially. We looked at the relative abundance of the elements. Oxygen was the most abundant but it was a gas. If you move down to silicon you find a solid to work with. So we chose silicon as the semiconductor. And silicon and oxygen in just about the ratio they exist in nature make SiO₂, so that was a good insulator. The third most common element was aluminium so that was what we used as a conductor.’   

Since then more elements have made an appearance in the devices that have been made: 51 elements in all. ’Since we’ve moved into a new generation of Moore’s law we’ve seen an explosion in the periodic table,’ says Patrick Gelsinger, senior vice president and general manager of Intel’s digital enterprise group. But, as Moore predicted in 1965, silicon still provides the scaffolding for all the devices.   

’If we think about communications systems that we use today, they all rely on silicon microelectronics technology,’ says Elsa Reichmanis, director of the materials research department at Bell Laboratories, part of Lucent Technologies. ’There is no question that the business that Moore started has expanded and is having an impact on every aspect of our lives,’ she adds.   

Reichmanis, at Bell labs, develops lithography for making devices on silicon wafers. A resist, usually a photo active polymer, is coated onto the silicon wafer, patterned and developed. Other layers are added and developed in the same way to build a device. Photo lithography is used - light of a chosen wavelength strips the resist from the wafer to leave behind the desired pattern. And herein lies a problem that could halt any continuation of Moore’s law. As devices get smaller and smaller, the wavelength of the light used to etch out their features is going to be larger than the features themselves. Lithography, not smaller devices, will be the limiting factor.   

Patterning problems   

Equally, the molecules that are used to mask the areas needed to stay behind on the wafers are going to cause problems. ’We need to start thinking about how we’re going to fabricate new materials and how we develop processes to deposit many individual layers on state of the art devices today,’ says Reichmanis.     

’There will be 30 or more discrete layers, and the thickness of many of those layers may only be a few nanometres. So as we start having to pattern features that small, the individual sizes of the molecules we are using as masks for those resists are approaching the size of the features we’re trying to pattern. I’m not sure how we can do that through traditional subtractive processing.’   

An alternative is self-assembly, or the ’bottom up’ nanotechnology approach - building devices molecule by molecule, atom by atom. Moore remains sceptical, ’I think nanotechnology is looking at the wrong end of the problem,’ he says. ’You can make some interesting little transistors but that’s not the main deal. You’ve got to connect a billion of these together to make a function.’     

He doesn’t see any alternatives to current fabrication methods appearing in the near future. ’People talk about quantum computing and DNA computing. To me those are pretty far out. I look at quantum computing and say "gee, that’s a marvellous way to show how non-intuitive quantum mechanics is". It’s so far from the point that you can use it to solve real problems that I don’t see how they can make the transition. DNA computing is somewhat the same way. There may be some ways of doing self organisation that we’ll come up with, that’s a nice concept - anything that gets around having to push the lithography that much further would be useful - but I don’t quite see what they are.’   

Donald Fitzmaurice, at University College Dublin (UCD), Ireland, and principle investigator at the centre for research and adaptive nanostructures and nanodevices (CRANN) based in Trinity College Dublin, works closely with Intel on developing the next generation of materials. ’It’s a very conservative industry,’ he says. ’People like Moore would be sceptical about the potential of bottom up approaches at the moment, but so would most scientists and technologists working even in the area. They see it as a huge challenge. It’s absolutely not clear that they can deliver on it yet, but it is certainly one of the approaches that has to be explored because it’s one of the few approaches around that could conceivably offer a solution.’   

Reichmanis is another bottom-up proponent: ’In 10 to 20 years I think we’ll need to think about additive processing that incorporates self-assembled and controlled growth,’ she says.   

Fitzmaurice agrees with Moore that linking tiny devices will be a major headache. ’The key issue is organisation,’ he says. He is counting on nature to help find the solution. ’Biology is able to assemble and organise vast numbers of complex nanostructures to achieve even more complex function.’ At CRANN and UCD, Fitzmaurice is looking at the information encoded in large biological molecules like DNA and proteins and using it to control the process of building nanoscale wires or switches from nanoparticles.     

Nanotubes show potential   

He’s also using carbon nanotubes to do similar work. This is an area where Moore does see potential, especially because nanotubes are such good conductors. ’If we can figure out how to incorporate nanotubes it would help get around one of the major problems we have of making things smaller and smaller,’ he says. This is the problem of increased resistivity that comes when devices are made smaller.   

Single crystal nanowires - silicon nanowires or carbon nanotube structures - will be introduced into mainstream processor technology in these new, smaller devices, says Fitzmaurice. ’You may well have a part of your desktop computer that is built from carbon nanotubes or nanowires.’ And the timeframe for this is, in his opinion, six to eight years.   

Moore’s law isn’t just about semiconductor chemistry and electronics. It’s about economics. The chicken-and-egg question often arises: is Moore’s law an observation on how the industry is developing or has Moore’s law become the driver for the industry? Is Moore’s law a self-fulfilling prophecy?     

Fitzmaurice sees the push to keep Moore’s law going as a real market driver. ’Moore’s law is both a technological and economic law,’ he says. ’Technologically it means that you double the number of transistors on a wafer every 18 months, but in effect you then halve the cost. So that means your computer does more but costs you less money.that is why Moore’s law is so important to the industry.’     

The semiconductor industry is one of the most successful global industries in terms of cooperating to compete. The process of getting a new production line for each new development - at the moment that is 65nm devices, with the future goals of 45nm and 22nm devices already mapped out - is highly evolved. ’Everyone has to agree to do the same thing and then compete to do it. If everybody was bringing in competing standards and competing products there would never be compatible equipment to fit out the entire production process,’ says Fitzmaurice.   

Making things smaller means everything gets better at the same time. Higher performance is possible with lower power and costs go down. Moore thinks this drives the competition in the industry. ’If you’re behind a generation you have both a performance and a cost disadvantage,’ he says. ’That’s a pretty tough competitive position, so participants recognise that they have to stay very close to the leading edge.’     

Long-lasting law   

The question on everyone’s lips is - when will Moore’s law end? Gelsinger, and Intel, plan to see it continue well into the future. ’We believe that Moore’s law is alive and well,’ says Gelsinger.     

Rodney Brooks, from the Massachusetts Institute of Technology and chief technical officer of artificial intelligence company iRobot, recently opened a celebration of the 40 year anniversary of Moore’s law organised by the US Chemical Heritage Foundation. He posed the questions: do we need faster computers? And do we even need Moore’s law anymore? Brooks’ answer was an emphatic ’yes’. ’We will find uses for faster computers,’ is his prediction. Gelsinger has the same idea: ’we believe that if we build it, they will come.’     

Reichmanis thinks that as people are given better devices, they want even better ones, ’We have a continuing thirst for new technology that has us developing more and more. That then is driving research and development. And the fabrication processes rely on chemistry.’   

’Moore’s law is as much about an attitude to your market as it is about technology,’ concludes Fitzmaurice. ’We’re going to continue to provide enhanced performance at reduced cost for our customers. And Moore’s law is one way of ensuring that you do that.’   

As for Moore, as he watches his law set off into its fifth decade, thoughts must return to those graphs he plotted by hand for that article 40 years ago, ’I could see things were going to change and I wanted to try to get that message across. And the fact that the projection proved at all accurate is an accident I think.’ It’s tempting to believe him, but for someone with such instinctive insight, this statement will have to be put down to modesty.