In the past 20 years alone, computing performance has seen a dramatic increase of about 10,000 times. This increase has, of course, been due to advances in processor technology, where performance for price has doubled every 18 to 24 months.
The impact has been profound, driving increases in productivity and efficiency for individuals, small businesses, and major corporations, while spawning innovations in medicine, defense, entertainment, and communications.
But now, this rate of processor improvement, known as Moore’s Law, is approaching the physical limits of the technologies inherent in current computer chips. It has been known all along that there are serious technical issues associated with packing more transistors and connections onto silicon chips.
For instance, the wavelengths of light used for critical steps in chip manufacturing have fundamental limitations, as does power management.
These limitations may be roadblocks for continued improvement of silicon chips made in traditional ways, but other innovative technologies and processing approaches are being developed that have the potential to keep processor advances in line with Moore’s Law.
We will examine three of them:
1. New strides in silicon chip technology2. Circuits made from the nanotech “wonder material” graphene3. Circuits based on a new substance called molybdenite
Several promising developments in silicon technology have been discovered as it moves to the nanoscale. One of them is a new technique in photo-lithography invented by researchers at MIT and the University of Utah.1,2 It enables the production of complex shapes and finer lines on chips, delivering additional leaps in computational power from conventional slivers of silicon.
Photolithography is the process of transferring the circuit paths and electronic elements of a chip onto a wafer’s surface using light sources. With traditional photolithography, features on chips are limited to being larger than the wavelength of the light used. The new technique has produced features that are one-eighth that size.
More than most, Intel’s success has relied on the relentless advance of Moore’s Law. With perhaps only 25 years left, Intel and others are scrambling for the next technological paradigm.
Although this feat been achieved previously, this is the first time it’s been done using equipment that is suitable for quick, inexpensive manufacturing processes. The technique relies on less-expensive light sources and conventional chip-manufacturing equipment, putting the cost on par with that found in current chip-making plants.
Massively parallel computing is another path being promoted as a means of pushing the limits of silicon-based processors in order to continue the persistent pace of doubled performance.3
This type of hardware and software configuration enables simultaneous processing of multiple computing activities. Currently, parallel computing is only being exploited for narrow sets of scientific and engineering applications. In fact, the hardware and software is common in video cards and supercomputers with hundreds or thousands of processors.
Expanding this technology to a broader set of applications will require new algorithms, programming models, operating systems, and computer architectures. Many industry experts argue that these challenges must be addressed so parallel computing can become a viable option for improving computer performance, longer term, without abandoning silicon.
§§§§§§§§§§
However, the promise of parallel computing is dimmed by energy management, when many processors are placed on a single chip, and by the lags involved in communications among chips.
Researchers at Rensselaer Polytechnic Institute discovered that stacking nanoribbons of graphene can boost the material’s ability to transmit electrical charges. The discovery further supports the idea that graphene could one day replace traditional copper as the best material for interconnects that transmit data and power within computer chips.
The latter problem is already being extensively researched by engineers addressing the bottlenecks in communication between individual super-chips in today’s supercomputers. Light can be faster than electrons moving though conductors when it comes to inter-processor communication. But, speed and bandwidth are at issue, especially when large quantities of data are transmitted to and from a computer via fiber optics.
The information must be translated into electronic signals for computers to use, and then translated back for retransmission. This process is slow and also makes the data susceptible to cyber-attack, in addition to requiring expensive equipment.
Researchers at Purdue University have recently created a passive optical diode that consists of two tiny silicon rings that are about one-tenth the width of a human hair.4 Because this diode can be easily integrated into industry standard CMOS computer chips and requires no external assistance for transmitting signals, it could eliminate the need for translation, paving the way for faster, more powerful, and more secure super-computers containing thousands of connected processors. And, over the longer term, this could filter down to massively parallel mainstream servers.
§§§§§§§§§§
Furthermore, as the size of each silicon-based processor shrinks, so must the copper pathways, called interconnects, that carry electricity and therefore information between the transistors and other components within a processor.
This size reduction of the interconnects causes several problems, including running less efficiently, using more power, generating more heat, and being more prone to permanent failure. These failings are significant impediments to continued performance gains for silicon computer chips.
To meet this challenge, researchers are trying to replace the copper with another alternative. One material that holds promise is graphene.
While silicon in layers less than two nanometers thick can become unstable, oxidize and quickly deteriorate, Molybdenite (i .e., Molybdenum disulfide or MoS2) can be laid down in sheets just three atoms thick. Recently, researchers at the Laboratory of Nanoscale Electronics and Structures (LANES) built the first Molybdenite integrated circuit. It has a natural semiconductor band gap and its based on an abundant, naturally occurring mineral. Molybdenite is also flexible, potetentially paving the wave for bendable computers.
This substance is made up of carbon atoms, arranged in a sheet, one atom thick, appearing much like a nanoscale chicken-wire fence. In essence, it is a single layer of the graphite found in a pencil.
A team of researchers at Rensselaer Polytechnic Institute has discovered that stacking several thin graphene ribbons on top of each other enhances their ability to transmit electricity, enabling the structure to act as an interconnect.5
Significantly, graphene does not exhibit the same negative characteristics that copper does as interconnects shrink. As copper nanowires shrink, electrons travel sluggishly, which creates a lot of heat, causing atoms of copper to be dragged along with the electrons.
Consequently, electrical resistance increases, the movement of electrons is decreased, and computer speed and performance is degraded.
While single layers of graphene cut into ribbons exhibit other problems, researchers recently discovered that these can be eliminated by stacking the graphene interconnects four to six layers thick. As a result, the researchers are highly optimistic about the use of graphene to replace copper interconnects. It appears that stacking graphene will prove to be a viable method for mass production.
Similarly, we are overcoming barriers that have so far prevented graphene use as a viable alternative to silicon itself. One such issue is a characteristic called band gap, which is an energy gap that prevents electrons from flowing freely. Graphene sheets have no “natural” band gap.
On the other hand, semiconductors like silicon and germanium, function because they exhibit narrow band gaps. That is, the band gap is large enough for the semiconductor to act as an insulator until an increase in electric field causes electrons to jump the gap, switching the semiconductor’s state from off to on. Without band gap, there is no control of the circuitry. So, without a band gap in graphene, it cannot be used to fabricate transistors.
Researchers at Rensselaer Polytechnic Institute believe they have resolved this issue.6 They were able to create a band gap in graphene by exposing it to humidity. By controlling the amount of water absorbed, they are able to precisely tune the band gap.
This ability to turn graphene into a semiconducting material makes it an extremely viable candidate for use in a new generation of transistors, diodes, nanoelectronics, and nanophotonics.
§§§§§§§§§§
At the same time, IBM has taken another approach to using graphene in circuitry.7 The company created integrated circuits on 200-millimeter wafers coated with an atom-thick layer of graphene. To create electrical gates, deep trenches were filled with tungsten before the graphene sheets were applied.
Although these new circuits are not ready for commercialization, IBM’s work is some of the first outside academia that has shown tangible results with graphene circuits.
§§§§§§§§§§
However, graphene is not the only substance getting attention today as an eventual mainstream alternative to silicon chips. Molybdenite is a relatively abundant, naturally occurring substance that can also be easily manufactured by reacting sulfur with the metal molybdenum ? and it definitely has the potential to overshadow graphene as the basis of the next epoch in information processing.
One of the key advantages molybdenite holds over silicon is that it can allow the size of transistors to shrink even further. Silicon cannot be made into layers any smaller than two nanometers thick. Thinner layers can allow a chemical reaction to take place that oxidizes the surface and compromises the electronic properties.
Conversely, molybdenite can be formed into layers that are only three atoms thick, which is three times smaller than the silicon layers; and yet, even at this size, it is stable, with conductivity that is easy to control.8
It’s this characteristic that gives it a crucial advantage over graphene. Molybdenite naturally exhibits band gap, making it suitable for use in electronics, whereas graphene must undergo adaptations to gain band gap.
Furthermore, physical flexibility is an attractive characteristic of molybdenite. So, it could readily lead to computers that roll up, or electronic devices that could be applied directly to the skin.
Even though it’s on the cutting edge of technology, integrated circuits have already been produced using molybdenite. This work, at the Laboratory of Nanoscale Electronics and Structures, has shown molybdenite’s advantages over silicon in terms of smaller scale, less energy consumption, and greater flexibility.9,10
There are many questions that still must be answered before molybdenite circuits can be produced on a large scale. However, it now appears to have a promising future in helping to keep the price-performance curve of Moore’s Law intact for the next decade and beyond.
Based on this trend, consider these forecasts:
First, processor price-performance advances will keep on track, in accordance with Moore’s Law, for at least the next decade.
An explosion of information technology in virtually every aspect of business and personal life, and our increasing dependence on it, will demand the continued improvement. No scientific show-stoppers are expected. Ultimately, the marketplace will decide whether advanced forms of silicon, graphene, molybdenite, or some other material emerges as the mainstream technology of the 2020s. But, without development of these alternatives, advances in information technology will stall, and with it the economic potential of the 21st century.
Second, the commercial potential of extending Moore’s Law will depend on parallel performance improvements in related technologies.
Extraordinary advances will also take place in data storage density, random access memory performance, and network bandwidth ? all critical for seamless, “anytime, anywhere” information access. In the not-too-distant future, many kinds of problems that remain unsolvable by conventional computers will be off-loaded to cloud-based quantum computers, a topic the Trends editors have previously examined.
References List :1. Physical Review Letters, November 11, 2011, Vol. 107, Iss. 20, "Breaking the Far-Field Diffraction Limit in Optical Nanopatterning Via Repeated Photochemical and Electrochemical Transitions in Photochromic Molecules," by Trisha Andrew Ph.D., et al. ⓒ Copyright 2011 by the American Physical Society. All rights reserved. http://prl.aps.org2. For additional information, visit the Massachusetts Institute of Technology website at: http://web.mit.edu3. For more information about parallel computing, visit the National Academy of Sciences website at: http://www8.nationalacademies.org4. Science, January 27, 2012, Vol. 335, Iss. 6067, "An All-Silicon Passive Optical Diode," by Minghao Qi, et al. ⓒ Copyright 2012 by the American Association for the Advancement of Science. All rights reserved. http://www.sciencemag.org5. ACS Nano, August 23, 2011, Vol. 5, Iss. 8, "Effect of Layer Stacking on the Electronic Structure of Graphene Nanoribbons," by Saroj K. Nayak, et al. ⓒ Copyright 2011 by the American Chemical Society. All rights reserved. http://pubs.acs.org6. Small, November 22, 2010, Vol. 6, Iss. 22, "Tunable Band Gap in Graphene by the Controlled Adsorbtion of Water Molecules," by Nikhil Koratkar, et al. ⓒ Copyright 2010 by John Wiley & Sons, Inc. All rights reserved. http://onlinelibrary.wiley.com7. For more information about using graphene in circuitry, visit the IBM website at: http://spectrum.ieee.org8. MIT Technology Review, January 30, 2012, "Graphene Competitor Used to Make Circuits," by Patrick Cain. ⓒ Copyright 2012 by Technology Review. All rights reserved. http://www.technologyreview.com9. ACS Nano, December 27, 2011, Vol. 5, Iss. 2, "Stretching and Breaking of Ultrathin MoS2," by Andras Kis, et al. ⓒ Copyright 2011 by the American Society. All rights reserved. http://pubs.acs.org10. ACS Nano, December 27, 2011, Vol. 5, Iss. 2, "Integrated Circuits and Logic Operations Based on Single-Layer MoS2," by Andras Kis, et al. ⓒ Copyright 2011 by the American Chemical Society. All rights reserved. http://pubs.acs.org
