One vertical extension from last month's 2003 predictions is the microprocessor long-term development path.
Moore's Law has been consistent over the last 37 years or so, but to leverage it forward an equal number of years would be truly remarkable. Another tool, which may also prove useful in any future prediction, is to understand how quickly knowledge and technology have been assimilated, and then to apply that rate of advancement to the development potential of the microprocessor. Presented here are the short term, intermediate, and long term prospects of microprocessor development.
Anyone working in the computer industry will at some time hear about Moore's Law because of its ability to predict future processor transistor density, and thus, performance. In 1965, just four years after the first planar integrated circuit was discovered (not microprocessor), Dr. Gordon E. Moore with Intel had observed exponential growth in the number of transistors that could be manufactured on a chip. Dr. Moore went on to predict this exponential growth would continue. As it turned out, Intel has been able to manufacture microprocessor chips that at least doubled the number of transistors over a 12-month period or so, and yet the cost per transistor has dropped over time.
Intel illustrates this point and maintains a very informative table and chart of the history of the microprocessor at www.intel.com/research/silicon/mooreslaw.htm. The graph they present is a linear plot on a semi-log grid, reflecting the number of transistors-per-processor-type plotted across the year that the processor was released. The number of transistors has grown exponentially, but when plotted on a semi-log graph it creates a linear plot that makes it easy to extrapolate forward into future years. This extrapolation provides a good base for long-term predictions of processor performance.
Intel's short-term strategic plans for the microprocessor can be found at www.intel.com/ebusiness/products/roadmap.htm. Here they detail various families of computers and specify the type of microprocessor they hope to release within the first half of 2003. Intel does develop longer-term strategic plans but unfortunately, those are not available for general pubic viewing unless you are a business partner and/or have signed a non-disclosure agreement. However, other interesting roadmaps for all of Intel's products and technologies can be found at www.intel.com by typing the keyword "roadmap" into their search engine.
Now, what about longer projections for microprocessor development? What about a new breed of microprocessor that might be in development?
Microprocessor manufacturing technology has advanced in a similar way to that of mankind's knowledge. First starting slow and then progressing faster each year as new technology builds on old technology. Many have observed that mankind's technological achievements in the 19th century equaled all of the achievements of the previous 10 centuries or so. More recently, the technological achievements in the last twenty years has matched all of the achievements of the 19th century. Clearly you can see the trend; it is taking a shorter amount of time to match previous achievements.
In microprocessor manufacturing, the same is true. Manufacturing a semiconductor started slow and with simple technology. Transistor manufacturing requires etching of microscopic structures out of silicon and other materials. These structures were imprinted on light sensitive material using a photographic mask, and then later chemically etched to create the transistor structure. Originally, light was used to etch the photosensitive material. However, as these transistor structures became smaller, simple light did not produce clear and sharp structures because of the wavelength. So, the manufacturing process was shifted to shorter wavelength light. Ultraviolet proved to meet the requirements and served the industry for many years, but it, too, has run its course. Currently the industry is experimenting with x-ray and extreme ultra-violet or EUV (soft x-ray) because of the very short wavelengths offered by these lights. Consequently, the transistor structure, in recent times, has shrunk from .45 microns to .25 micron then to .13 microns and more recently to 90 nanometers. December 9th, 2002 IBM announced manufacturing a proof-of-concept transistor that measures 6 nanometers across. The following reference provides additional information about IBM's latest accomplishment. channels.netscape.com/ns/news/story.jsp?floc=FF-PLS-PLS&id=12090002000246059&dt=20021209000200&w=RTR&coview
To help one understand the significance of this small scale, a nanometer is 10 angstroms. A silicon atom has a diameter of 3 angstroms. Thus IBM's new transistor is just 20 atoms across. This is amazing and one could wonder where is the limit? environmentalchemistry.com/yogi/periodic/atomicradius.html.
The technique of manufacturing transistors using masks may soon reach its limit. So what manufacturing techniques could replace the masking process? The fundamental goal is not small transistor structures, but fast processing. It just happens that the smaller the transistor, the more that can be packed onto a chip, thus resulting in faster processing. One of the reasons this occurs is because the distance between transistors is reduced, so the time of communication between transistors is reduced, and thus the processor can operate at a higher clock speed.
3 Dimensional Processors
An extension of this reduction in communication time concept is a 3D structure. Researchers have found that the transistors are getting so small and that so many are getting packaged onto a 2 dimensional chip, that it takes longer to transmit a signal from one side of the chip to the other side than if they were to stack the components vertically, thus creating a 3 dimensional device. This is easier said than done. One Pentium processor requires at least 13 layers for the completed product, and that does not include the housing. A 3 dimensional processor would require approximately 26 layers, and perhaps more, to handle the heat dissipation. This is an interesting concept and the following reference describes the technique and some products that may be in the market place today. www.sciam.com/article.cfm?articleID=000BD05C-D352-1C6A-84A9809EC588EF21&pageNumber=1&catID=2
Another method to increase performance is to switch photons instead of electrons. A lot of research is being conducted on processors that switch packets of light or spectrums of light instead of streams of electrons. Once developed, these devices would be well suited for quantum computing. In other words in a few seconds, one of these computers could solve certain families of problems that would take a normal computer years (to decades) to complete. There are many obstacles to overcome before these computers will sit on your desk. Needless to say, their potential is exciting and certainly something to look for in the years to come. The following article reference will provide more information about quantum light computers. www.sciam.com/article.cfm?articleID=000211BF-48C2-1C6F-84A9809EC588EF21&pageNumber=1&catID=2
Not all processors have to be made with silicon semiconductors. There are other ways to build transistors and achieve fast processing speeds without using silicon. Recent advancements in carbon nanotubes are proving to be promising prospects to replace the silicon transistor. A carbon nanotube is a cylindrical tube composed of carbon atoms. These structures are quite small and would be suitable for high clock speeds, thus high performance. Researchers have been able to manipulate these structures to get a transistor type response. The following article describes some of the techniques used on nanotubes and their respective semiconducting state.www.sciam.com/article.cfm?articleID=0003B31B-F8FC-1C67-B882809EC588ED9F&pageNumber=1&catID=1
Another area of research makes use of molecules to produce a transistor effect. Here researchers place a specific type of molecule between two wire conductors, and then manipulate the molecule in such a fashion as to achieve a transistor effect. The molecules are very small and would be suitable for high clock speeds, thus high performance. Prototypes of these molecules have been manufactured and are proving to be a possible replacement to the silicon semiconductor. The following article also describes the techniques used on molecules to achieve a semiconducting state. www.sciam.com/article.cfm?articleID=00017C07-D8E3-1D07-8E49809EC588EEDF&pageNumber=1&catID=1
One of the most interesting fields of research, and the one that will yield the greatest benefit if it can be developed, is the quantum spintronics processors. Here the researchers use the characteristics of the electron spin to form the bases of a transistor circuit. As an electron spins, it has a magnetic axis that points up or down, depending on the orientation of the spin. Here an up-pointing spin could equal a 0 value state, and a down-point spin could equal a 1 value state. And as with the optical transistor, these quantum spintronic processors will have to ability to solve specific types of problems in a matter of seconds, while a normal computer would take years or decades. The following article describes the technique in greater detail. www.sciam.com/article.cfm?articleID=0007A735-759A-1CDD-B4A8809EC588EEDF&pageNumber=1&catID=2
As you can see, microprocessor technology is truly advancing at a breathtaking pace. Advancements in transistor reduction and the wide diversification in other types of processors are clear signs of accelerating returns. But these examples only project solutions for the intermediate future of microprocessors. So what can we expect to see in 20 or 30 years from now. It would be hard to predict the physical characteristics of processor that would be manufactured in two or three decades from now, but we can get a handle on what the performance might be. Several individuals have created semi-log graphs that plot the calculations-per-second of various processors plotted across time in years as to when the processor was introduced into the marketplace. Similar to the Moore's Law graph, this graph illustrates the capability of processors is increasing at an exponential rate.
One of the more interesting charts that has been created along this line was presented in an article written by Ray Kurzweil. He presented the typical calculations-per-second verse time chart but also drew horizontal lines at specific calculation-per-second points that represented the computational ability of an insect brain, a mouse brain, a human brain and the computational ability of all the brains in the entire human race. Apparently some neuroscientists came up with these figures by estimating the number of neurons in the respective brains and then determined the number of calculations that could be performed by that quantity of neurons.
The interesting point about this graph is that it illustrates that present day processors have the computational ability of an insect brain. After getting out the engineer scale and triangle, it appears that in the year 2010, processors will have the computational ability equal to a mouse brain. Then in the year 2030, processors will have the computational ability equal to a human brain. And then just 30 years later in 2060, a single computer will have the computational ability of all mankind. There was some disparity between the benchmark points extracted from his graph and comments made by Ray Kurzweil in his article, where he indicated processors would have the computation ability of the human brain by 2019 and a single computer would have the computational ability of all brains in the human race by 2029. In either event, this gives you a window as to when these events may occur. Ray Kurzweil's article can be found at: www.sciam.com/ article.cfm?articleID=0007EE6E-F71F-1C72-9B81809EC588EF21& pageNumber=1&catID=9
The last point to consider is the parallel advancement of machine intelligence that would develop concurrent with processor development. For years we have had computer-aided graphics (CAD), likewise we have had computer aided electronic design applications, similar to Electronic Work Bench. So far, these applications required a human to create designs.
However, as machine intelligence increases in capability, then it will be possible for computers to design the next generation of processors.
These initial machine creations may or may not be as good as the human designs, but as the machines build on the knowledge and experience, then it is not inconceivable that a machine-designed computer will exceed the capabilities of a human-designed computer.
Once this ball starts rolling, then we can see a new generation of accelerating returns for the machine-designed processors. For the second generation computer will design the third generation processor, and the third generation computer will design the fourth generation processor, and at each stage the next generation design will be much more advanced than its predecessor. At this rate, the design advancements could very easily go well beyond the comprehension of mankind.
To close, it might be humbling to ponder the opening quote Ray Kurzweil made in his article referenced earlier, "The accelerating pace of technological progress means that our intelligent creations will soon eclipse us-and that their creations will eventually eclipse them."
Ron Fenley worked as an engineer/analyst and retired in 1999. Ron moved to the country and now pursues his interest in computers, basic science and technology. Ron has been a computer enthusiast for 20 years and has been a HAL-PC member for about half that time. Ron can be reached at email@example.com