Re-inventing The Chip

At first sight, Jun Yao, a PhD student at Rice University in the US, is an unlikely person to change the foundations of the global semiconductor industry. After all, the industry is composed of giants such as Intel and Samsung, and spends more than $30 billion a year on research and development (R&D). The chip industry is seemingly mature, with no major technology revolutions happening in the past few decades. It is not known to have adapted and commercialised many path-breaking technologies from the universities for decades. What can a university student offer an Intel now?

Yao has just announced, along with his professor James Tour, the fabrication of a tiny transistor using silicon oxide. His work is counter-intuitive; silicon oxide is normally used for insulation, and not as part of electrical circuits. "It took me a year and a half to convince my colleagues that this was happening," says Yao. People in industry are even more sceptical than academics about commercial technology.

But these are unusual times for the semiconductor industry. It had built its reputation, not to speak of immense wealth, on an ability to keep shrinking the transistor relentlessly for five decades. The leading chip companies know that if there isn't a wealth of technology options to choose from soon, they will not be able to keep shrinking the transistor as they have done so far.

A Switch Of Hope
The Rice invention suggests a fascinating possibility: a less sophisticated switch that could let chip companies go to small dimensions and into 3D. And it was only one of a series of inventions in the past two years that suggests either high-density storage or high performance computing within a few years. A day after Rice University published its research, Hewlett-Packard (HP) announced that it is only three years away from commercialising a completely new circuit element — a part of an electrical circuit — called memristor. This could, says HP, give us exceptional memory and logic materials. We don't know how these inventions would stand the test of the industry, and still less how they would compete against similar technologies. But they are at the top of a growing list of technology options to replace the seven-decade-old transistor in the near future.

Other inventions in the world's laboratories include nanodot storage, the junctionless transistor, spintronics chips, and so on. There was no such list even two years ago. When taken individually, they may seem like blue-sky research, but when taken together they provide the industry with a range of options within five years. Says Stan Williams, senior fellow at HP and founding director of its Information and Quantum Systems Lab: "Because existing memory and storage technologies are facing huge challenges going forward, there has been an explosion of interest in finding new technologies to compete with or eventually replace them."

Of The Shrinking Chip
Soon after the first silicon transistor was developed in the 1950s, the semiconductor industry started a process that was to last nearly six decades: continuous scaling. The industry did not seem very innovative to outsiders — Intel, for example, did little research and its R&D consisted almost exclusively of development —but its ability to keep shrinking the transistor without pause for five decades fuelled innovation in other industries and economic growth throughout the world.

MOORE'S LAW

WHERE In computing WHO Formulated by Gordon Moore, co-founder of Intel, in 1965 WHAT The number of transistors that can be placed inexpensively on an integrated circuit has doubled about every two years. This has continued for more than half a century and might stop by 2015

Moore's Law, formulated by Gordon Moore in 1965, predicted the number of transistors on a chip will double every two years. Industries such as IT, life sciences, banking and energy worked Moore's Law into their roadmaps and strategies. However, semiconductor companies cannot shrink the transistor forever. Physicists know that when you reach very small dimensions, currents — and thus information stored in chips — leak and atomic particles behave in strange ways. Engineers know that — and have found out in practice — that it is not easy to dissipate the heat if you make the chip compute too fast. The industry somehow managed these problems with some smart engineering. Clever workarounds usually end at some time.

Semiconductor companies built big businesses based on Moore's Law, by charging a high premium on new chips and making huge profits. When Moore's Law ends — and it is predicted to end soon, although no one knows precisely when — this business model also would end. More importantly, innovation in many other fields would slow down considerably. Developments in gene sequencing, climate modelling, energy research and many others would not be possible without continuous improvements in computing performance. So, the world needs Moore's Law to continue. For this, chip companies are developing several alternative strategies, all of which involve increasing the performance without shrinking the transistor.

In the next two years, Intel and AMD would improve the performance of microprocessors by using more cores — the heart of a microprocessor — in one chip. The experimental cloud chip, launched by Intel early this year, contains as many as 48 cores, compared to a maximum of 12 cores in current servers. A large number of cores in a chip can be problematic; these cores have not been originally designed to work in tandem within a single chip. Many industry observers feel the core of the processor needs to be optimised, or redesigned, for massive parallel processing. It would also need a paradigm shift in software development.

The other options being tried by the industry include things such as optical interconnects that will speed up communication inside a chip (light travels fast), 3D stacking, keeping the memory close to the processor, and so on. None of these approaches is revolutionary, but they help continue increasing chip performance for a while. But after 5-7 years, the chip industry may need to reinvent the chip. The new set of discoveries could address this combination of problems. "We have been in the scaling business for a long time now," says Suman Datta, professor of electrical engineering at Penn State University and a former Intel engineer. "We are going to see a big shift in technology soon, but it is difficult to say which one will win."

After the industry exhausts the current materials and architectures, it would look at new ones. The first new materials would be used in memory. One material called phase change memory has already started replacing flash, and this process will accelerate soon. "Phase change memory will completely replace flash in five years," says Chung Lam, IBM's lead phase-change memory researcher. Even this could face competition from other memory materials soon. Memristor could become a serious competitor. The Rice switch is another. Says Mike Mayberry, director of components research at Intel: "These two examples are interesting, but not yet proven to be better than other dense-memory proposals that others have investigated."

Take memristor. It is now being developed as a memory storage technology. It would first start replacing flash and then replace the hard drive, if the technology is proven commercially. Memristors can store 10 times more information while using one-tenth of the energy consumed by flash memory. So it is an advance on two fronts rather than one. But as HP demonstrated this year, it can also do the job of a logic circuit, now performed by the transistor. A memristor is a simpler entity than the transistor, and they can be made very small without resorting to complex and costly production processes. Semiconductor firms are still sceptical as the technology is yet to be commercially proven.

More Optionss
Here are a few more examples of new chip technology options. Three months ago, Roger Narayan, a professor at North Carolina State University, developed a chip based on nanodots or nanoscale magnets; one square inch of a nanodot chip could hold one billion pages of a book. Earlier in the year, researchers at the Tyndal Institute in Ireland fabricated the world's first junction-less transistor; this device is supposed to be particularly suitable for chips less than 10 nanometres in diameter. Researchers at University of California in Berkeley have made chips using iron nanoparticle and carbon nanotubes that reportedly can store information for a billion years. Another promising line of work includes the use of magnetic materials, and the use of spin as well as the charge in the electron.



At IBM, engineers have been developing a technology that they call racetrack memory. Racetrack memory uses electron spin to store data. A hand-held device — like an MP3 Player — that uses this method can store a few thousand movies. IBM thinks the technology would be commercialised by the end of this decade, but breakthroughs are coming rapidly in this space. For example, a month ago, a University of Kansas researcher developed a method for rapidly reading the spin of electrons; such a method is critical in spintronics, an exciting new area that uses the spin of the electron rather than its charge to store information. Last month, University of Ohio researchers demonstrated a plastic spintronics device that can store information.

So the next few years will be unusually exciting for a semiconductor technologist, as several technologies compete for industry glare. Whatever wins in the end, one thing is certain: the semiconductor industry is due for big technology changes this decade.

p(dot)hari(at)abp(dot)in