Article - Issue 8, May 2001
Magnetic nanotechnology and computers
Dr Russell Cowburn
The microprocessors and disk drives in our computers are becoming ever more powerful and efficient. But the physical properties of materials will eventually limit further advances and we will have to find alternative technologies. Russell Cowburn introduces us to the fascinating world of magnetic nanotechnology and proposes that our computers might one day use magnetism, not electricity, to process information.
The silicon microchip has been one of the most impressive feats of engineering ever performed by mankind. The number of transistors on a chip has steadily doubled every 18 months for the last 25 years. So steady has been the growth that the phenomenon has even acquired the status of a pseudo-scientific law called ‘Moore’s Law’, after Gordon Moore, the founder of the computer giant Intel. It is this exponential growth which has caused the remarkable increase in capabilities of modern computers.
I am old enough to remember seeing as a child the launch of the ‘amazing’ ZX81 home computer by Sir Clive Sinclair. Others will remember when computers filled entire rooms. In 1943, Thomas Watson, the chairman of IBM, is reported to have said, ‘I think that there is a world market for maybe 5 computers’. How far from the truth he was! Today’s mobile phones carry more computing power than one of IBM’s early room-sized computers. The much-discussed ‘Information Age’ is most certainly upon us, and has only been possible because of Moore’s Law.
Running in parallel with the engineers’ increasing ability to process information has been the technology of information storage. Hard disk drive capacities have also shown an exponential growth during the past 40 years. Processing and storage must grow together: as computers become more powerful, programs and data sets become larger.
But how much longer can the growth continue? Mathematicians like to mock exponential laws like Moore’s Law by calculating the number of years before a chip contains more transistors than there are particles in the Universe! Clearly the growth must stop one day, but when? It is notable that both the microprocessor and data storage industries are each beginning to worry about their future roadmaps as they encounter increasingly difficult technical problems. A recent New York Times article entitled ‘Chips forecast to hit a big barrier’ quotes an Intel scientist: ‘These fundamental issues have not previously limited the scaling of transistors … there are currently no known solutions to these problems.’ He added that this was ‘the most difficult challenge the semiconductor industry has ever faced’. International conferences on data storage are similarly filled with papers entitled ‘Ultimate limits in data storage’ and the like.
Most observers agree that current techniques can be used for another 10 years at least. After that, the future is less certain. What is reasonably clear is the nature of the technical problems which will limit further growth of the microprocessor and disk drive as we currently know them. In both cases, size is the issue. The transistors which make up microprocessors have two key parts: the channel, through which all of the transistor’s current passes, and the gate, which controls the flow of current through the channel. For the transistor to work, the gate must be electrically insulated from the channel. One of the consequences of reducing the size of the transistor in order to pack more on to a chip is that the thickness of the insulation between the gate and the channel (called the ‘gate oxide’) must also reduce proportionately. Fundamental physics of materials, however, puts a minimum allowed value on the thickness of gate oxide before it stops being an insulator and becomes a conductor. When this happens the transistor stops working. The New York Times article was actually triggered by a report in the journal Science that Intel researchers had found the limit of gate oxide thickness.
It may be that the structure of the transistor can be modified to overcome this particular problem. Silicon-on-insulator (SOI) technology may provide one means of doing this. Nevertheless, more fundamental problems are also awaiting solution. The first is to do with the number of electrons carrying signals through the transistor channel. Channels are made from semiconductors, which means that the density of carriers (number of electrons or holes per cubic metre of material) is relatively low. Consequently, a small transistor only contains a small number of electrons. The danger is that the number of electrons becomes too small. When this happens, the electrical current no longer flows steadily but rather in a staccato start–stop fashion, thus greatly increasing the electronic noise on the device. Noise means errors and errors mean a useless chip. It is unlikely that problems of this nature can be overcome simply by changing the transistor structure.
A further basic problem concerns power dissipation from the chip, which results in the microprocessor becoming hot. Significant engineering effort has always gone into removing the thermal power as efficiently as possible from the transistors to prevent them from overheating. As the transistors become smaller and smaller, the power density increases and the challenges in heat management become tougher.
Size is as much of a problem in the data storage industry. Information is stored on a hard disk by locally magnetising a small region, or domain, of a magnetic material. The reason that I can buy an 18 Gigabyte hard disk drive today whereas 5 years ago I could only buy a 1 Gigabyte drive is simply because the domains have been made smaller. The disadvantage of reducing the domain size is that thermal fluctuations of the magnetisation become relatively more important. The worst fear of most data storage engineers is a phenomenon known as the ‘super-paramagnetic limit’ in which thermal fluctuations become so great that the data simply evaporate.
We say that one man’s poison is another man’s tonic, and this is certainly the case with the physics of small sizes. Nanotechnology delights in understanding and using phenomena which uniquely occur in small (nearatomic) objects. The very fact that so many of the current challenges facing the computer industry are associated with the unusual behaviour of small transistors and domains means nanotechnology is perhaps very well placed to offer solutions.
As a researcher, I am interested in the question: can we use magnetic nanotechnology to help make better information processing and storage devices? In conjunction with Professor Mark Welland’s group at Cambridge University, my research group at Durham University has been researching nanometre-size magnets, or simply ‘nanomagnets’ (see Figure 2). (1 nanometre equals 1 millionth of a millimetre.) One of the most striking things that we have learned about nanomagnets during the course of our experiments is that their properties can be engineered to an astonishingly high degree of precision, simply by adjusting their size and shape. Equally importantly, and in stark contrast to electronic transistors, nanomagnets become more predictable and function more reliably the smaller one makes them.
This important discovery has led us to propose a new type of microprocessor that is based entirely on magnetism. Instead of using electrical currents controlled by transistors, it processes information by exchanging magnetic fields between nanomagnets. And so, whereas a conventional microprocessor signals the number 1 by a high voltage and the number 0 by a low voltage, our magnetic chip uses a North pole to represent a 1 and a South pole for a 0. This area of research is still at a very early stage and many fundamental questions still need to be addressed in the research laboratory before a product could be made.
Nevertheless, there are at least two encouraging indications for the future of this new technology. The first is a parallel development in the semiconductor industry called MRAM (Magnetic Random Access Memory). Pioneered by companies such as IBM, Siemens, Motorola and Hitachi, MRAM integrates tiny elements of magnetic metals into computer memory chips. The memory chip then performs in the same way as a conventional memory chip, except that it retains its memory when the computer is powered down – a property known as non-volatility. The most immediate advantage of MRAM to the consumer is that computers will no longer require a lengthy boot-up time when switched on in the morning. MRAM is now a mature research field and is entering a production development phase. We take encouragement from this because MRAM (if successful) will establish firmly the concept that microelectronics can be significantly enhanced by the incorporation of magnetic materials. Once the memory functions of a computer have thus been enhanced, it is a smaller conceptual leap to imagine that the microprocessor could similarly benefit.
The second encouraging indication of the future viability of magnetic microprocessors comes from an experiment which we recently performed in Cambridge University Engineering Department. Figure 3 shows a microscope image of a small magnetic logic device which we made and successfully tested. The device performs a logical AND function between two digital signals. The AND function is one of the basic functions which any microprocessor must be able to perform.
Is there a tangible advantage in using magnetism instead of electricity to perform computation? It appears that there is. We see magnetic chips potentially outperforming their electronic cousins in two key respects. Firstly, the magnetic dots that form the magnetic chips can be made to be very small indeed – about 40,000 times smaller than the transistors in today’s chips. In one sense, because they exploit quantum physics (exchange interaction), the smaller they are the better they work. That means that 40,000 times more processing capability can be packed into the same size chip. Secondly, it turns out that the magnetic chips use very little energy. An approximate analogue can be drawn from everyday life: a torch bulb runs a battery down, whereas a bar magnet works without being plugged into anything. Although some power will be needed, we estimate that it will only be one millionth of today’s electronic chips’ requirement. In practice, that means laptop computers with very small batteries. The days of carrying around heavy batteries are numbered!
Information processing is not the only candidate likely to benefit from magnetic nanotechnology. The hard disk drive industry is looking increasingly closely at a number of potential solutions from nanotechnology. The sensors which read back magnetic information from the storage disk must detect ever smaller magnetic fields. The conventional method for doing this used a phenomenon called AMR (Anisotropic Magneto-Resistance) which simply means that the electrical resistance of a piece of magnetic material can be changed by applying a small magnetic field to it.
In 1988, two European researchers (Albert Fert from France and Peter Grünberg from Germany) discovered a much more powerful magnetic effect called Giant Magneto-Resistance (GMR). Whereas AMR causes resistance to change by 2%, GMR can cause resistance changes of 50% or more. The effect could most usefully be found in metallic layers of only a few nanometres’ thickness. Within approximately 10 years, the new physics became mainstream engineering; virtually every hard disk drive now shipped contains a GMR sensor.
The disk itself which carries the data domains is also being looked at carefully by nanotechnologists. One of the limitations of current disks is the precision with which the shape of a data bit can be defined. The magnetic domains invariably have jagged edges which introduce errors into the read-back process. One possible solution is to use nanotechnology to break the disk surface up into a vast array of magnetic nanostructures and to store one bit of data in each structure. Figure 5 shows an example of a section of so-called ‘patterned media’ which we have fabricated in our laboratory.
There is still a great deal of fundamental research to be done before many of these technologies will become commercially available. Interest in nanotechnology is, however, undoubtedly increasing. More importantly, one senses that industry is really beginning to believe that nanotechnology will offer tangible commercial advantages. GMR, using nanometre thick layers, is already a mature industrial product; MRAM is on the verge of leaving the research laboratories and entering production lines. Other magnetic nanotechnologies such as the magnetic microprocessor are still at a very early research stage. Engineers are now being offered tools which will allow them to manipulate materials with unprecedented precision. The 21st century promises to be an interesting one.
both the microprocessor and data storage industries are beginning to worry about their future roadmaps as they encounter increasingly difficult technical problems
R.P. Cowburn (2000). The attractions of magnetism for nanoscale data storage. Phil. Trans. R. Soc. Lond. A 358, 281–301.
R.P. Cowburn and M.E. Welland (2000). Room temperature magnetic quantum cellular automata. Science 287, 1466– 1468.
R.P. Cowburn (2000). Property variation with shape in magnetic nanoelements. J. Phys. D 33, R1–R6.
Dr Russell Cowburn
Nanoscale Magnetism Group, University of Durham
Russell Cowburn has been a Lecturer in Physics at Durham University since October 2000. Before this he was a Fellow of St John’s College, Cambridge. He specialises in research into nanotechnology. When not having fun with high-power laser beams, Russell listens to Brahms and, with his wife, is an active member of a local church. Email: firstname.lastname@example.org