With simple rules but endless permutations, chess has fas- cinated millions of players for hundreds of years. When mechanical automatons became fashionable in the 18th century, onlookers were intrigued by “the Turk,” a chess- playing automaton. While the Turk was eventually shown to be a hoax (a human player was hidden inside), the devel- opment of the electronic digital computer in the mid-20th century provided the opportunity to create a true automatic chess player.
In 1950 Claude Shannon outlined the two basic strate- gies that would be used by future chess-playing programs. The “brute force” strategy would examine the possible
moves for the computer chess player, the possible replies of the opponent to each move, the possible next moves by the computer, and so on for as many half moves or “plies” as possible. The moves would be evaluated by a “minimax” algorithm that would find the move that best improves the computer’s position despite the opponent’s best play.
The fundamental problem with the brute force is the “combinatorial explosion”: Looking ahead just three moves (six plies) would involve evaluating more than 700,000,000 positions. This was impractical given the limited comput- ing power available in the 1950s. Shannon realized this and decided that a successful chess program would have to incorporate principles of chess strategy that would enable it to quickly recognize and discard moves that did not show a likelihood of gaining material or improving the position (such as by increasing control of center squares). As a result of this “pruning” approach, only the more promising initial moves would result in the program looking ahead—but those moves could be analyzed much more deeply.
The challenge of the pruning approach is the need to identify the principles of good play and codify them in such a way that the program can use them reliably. Progress was slow at first—programs of the 1950s and 1960s could scarcely challenge an experienced amateur human player, let alone a master. A typical program would play a mix- ture of reasonable moves, odd-looking but justifiable moves, and moves that showed the chess version of “nearsighted- ness.” By the 1970s, however, computing power was rapidly increasing, and a new generation of programs such as Chess 4.0 from Northwestern University abandoned most pruning techniques in favor of brute-force searches that could now extend further ahead. In practice, each programmer chose a particular balance between brute force and pruning-selection
In the 18th century the Turk, a mechanical chess player, astonished onlookers. Although the original Turk was a fraud (a small human player was hidden inside), the modern computer chess program Fritz 9 pays its homage by simulating its predecessor. (fRitz 9, chessbase gmbh, WWW.chessbase.com)
techniques. An ever-increasing search base could be com- bined with evaluation of particularly important positional features (such as the possibility of creating a “passed pawn” that could be promoted to a queen).
By the end of the 1970s, International master David Levy was still beating the best chess programs of the time (defeating Chess 4.7 in 1978). A decade later, however, Levy was defeated in 1989 by Deep Thought, a program that ran on a specially designed computer that could examine hundreds of millions of positions per move. That same year World Champion garry Kasparov decisively defeated the machine. In 1996, however, the successor program Deep Blue (sponsored by IBm) shocked the chess world by beat- ing Kasparov in the first game of their match. Kasparov went on to win the match, but the following year an updated version of Deep Blue defeated Kasparov 3 1/2–2 1/2. A com- puter had arguably become the strongest chess player in the world. As a practical matter, the match brought IBm invalu- able publicity as a world leader in supercomputing.
c
hessandai
The earliest computer chess theorists such as Claude Shan- non and Alan Turing saw the game as one potential way to demonstrate true machine intelligence. Ironically, by the time computers had truly mastered chess, the artificial intelligence (AI) community had concluded that mastering the game was largely irrelevant to their goals. AI pioneers Herbert Simon and John mcCarthy have referred to chess as “the Drosophila of AI.” By this they mean that, like the ubiquitous fruit flies in genetics research, chess became an easy way to measure computer prowess. But what was it measuring? The dominant brute-force approach was more a measure of computing power than the application of such AI techniques as pattern recognition. (There is, however, still some interest in writing chess programs that “think” more like a human player.) In recent years there has been some interest in programming computers to play the Asian board game go, where positional and structural elements play a greater role than in chess. However, even the latest generation of go programs seem to be relying more on a statistical approach than a deep conceptual analysis. Further Reading
Computer History museum. “mastering the game: A History of Computer Chess.” Available online. URL: http://www. computerhistory.org/chess/. Accessed April 28, 2007. Hsu, Feng-Hsiung. Behind Deep Blue: Building the Computer That
Defeated the World Chess Champion. Princeton, N.J.: Princ- eton University Press, 2004.
Levy, David, and monty Newborn. How Computers Play Chess.
New York: Computer Science Press, 1991.
Shannon, Claude E. “Programming a Computer for Playing Chess.” Philosophical Magazine 41 (1950): 314. Available from Computer History museum. Available online. URL: http:// archive.computerhistory.org. Accessed April 27, 2007.
chip
As early as the 1930s, researchers had begun to investi- gate the electrical properties of materials such as silicon
and germanium. Such materials, dubbed “semiconductors,” were neither a good conductor of electricity (such as cop- per) nor a good insulator (such as rubber). In 1939, one researcher, William Shockley, wrote in his notebook “It has today occurred to me that an amplifier using semiconduc- tors rather than vacuum [tubes] is in principle possible.” In other words, if the conductivity of a semiconductor could be made to vary in a controlled way, it could serve as an electronic “valve” in the same way that a vacuum tube can be used to amplify a current or to serve as an electronic switch.
The needs of the ensuing wartime years made it evi- dent that a solid-state electronic device would bring many advantages over the vacuum tube: compactness, lower power usage, higher reliability. Increasingly complex elec- tronic equipment, ranging from military fire control sys- tems to the first digital computers, further underscored the inadequacy of the vacuum tube.
In 1947, William Shockley, along with John Bardeen and Walter Brattain, invented the transistor, a solid-state electronic device that could replace the vacuum tube for most low-power applications, including the binary switch- ing that is at the heart of the electronic digital computer. But as the computer industry strove to pack more process- ing power into a manageable volume, the transistor itself began to appear bulky.
Starting in 1958, two researchers, Jack Kilby of Texas Instruments and Robert Noyce of Fairchild Semiconduc- tor, independently arrived at the next stage of electronic miniaturization: the integrated circuit (IC). The basic idea of the IC is to make semiconductor resistors, capacitors, and diodes, combine them with transistors, and assemble them into complete, compact solid-state circuits. Kilby did this by embedding the components on a single piece of ger- manium called a substrate. However, this method required the painstaking and expensive hand-soldering of the tiny gold wires connecting the components. Noyce soon came up with a superior method: Using a lithographic process, he was able to print the pattern of wires for the circuit onto a board containing a silicon substrate. The components could then be easily connected to the circuit. Thus was born the ubiquitous PCB (printed circuit board). This technology would make the minicomputer (a machine that was roughly refrigerator-sized rather than room-sized) possible during the 1960s and 1970s. Besides the PCBs being quite reli- able compared to hand-soldered connections, a failed board could be easily “swapped out” for a replacement, simplify- ing maintenance.
F
romic
toc
hiPThe next step to the truly integrated circuit was to form the individual devices onto a single ceramic substrate (much smaller than the printed circuit board) and encapsulate them in a protective polymer coating. The device then func- tioned as a single unit, with input and output leads to con- nect it to a larger circuit. However, the speed of this “hybrid IC” is limited by the relatively large distance between com- ponents. The modern IC that we now call the “computer chip” is a monolithic IC. Here the devices, rather than being chip
attached to the silicon substrate, are formed by altering the substrate itself with tiny amounts of impurities (a process called “doping”). This creates regions with an excess of electrons (n-type, for negative) or a deficit (p-type for posi- tive). The junction between a p and an n region functions as a diode. more complex arrangements of p and n regions form transistors. Layers of transistors and other devices can be formed on top of one another, resulting in a highly com- pact integrated circuit. Today this is generally done using optical lithography techniques, although as the separation between components approaches 100 nm (nanometers, or billionths of a meter) it becomes limited by the wavelength of the light used.
In computers, the IC chip is used for two primary func- tions: logic (the processor) and memory. The microproces- sors of the 1970s were measured in thousands of transistor equivalents, while chips such as the Pentium and Athlon being marketed by the late 1990s are measured in tens of millions of transistors (see micRopRocessoR). mean- while, memory chips have increased in capacity from the 4K and 16K common around 1980 to 256 mB and more. In what became known as “moore’s law,” gordon moore has observed that the number of transistors per chip has doubled roughly every 18 months.
F
uturet
echnoLogiesAlthough moore’s law has proven to be surprisingly resil- ient, new technologies will be required to maintain the pace of progress.
In January 2007, Intel and IBm separately announced a process for making transistors out of the exotic metal haf- nium. It turns out that hafnium is much better than the tra- ditional silicon at preventing power leakage (and resulting inefficiency) through layers that are only about five atoms thick. Hafnium transistors can also be packed more closely together and/or run at a higher speed.
Another approach is to find new ways to connect the transistors so they can be placed closer together, allow- ing signals to travel more quickly and thus provide faster operation. Hewlett-Packard (HP) is developing a way to place the connections on layers above the transistors them- selves, thus reducing the space between components. The scheme uses two layers of conducting material separated by a layer of insulating material that can be made to conduct by having a current applied to it. Although promising, the approach faces difficulties in making the wires (only about 100 atoms thick) reliable enough for applications such as computer memory or microprocessors.
Ultimately, direct fabrication at the atomic level (see
nanotechnology) will allow for the maximum density and efficiency of computer chips.
Further Reading
Baker, R. Jacob, Harry W. Li, and David E. Boyce. CMOS Circuit Design, Layout and Simulation. New York: IEEE Press, 1998. Saint, Christopher and Judy Saint. IC Layout Basics. New York:
mcgraw-Hill, 2001.
Semiconductor Industry Association. Available online. URL: http://www.sia-online.org/home.cfm. Accessed August 13, 2007.
Thompson, J. m. T., ed. Visions of the Future: Physics and Electron- ics. New York: Cambridge University Press, 2001.
chipset
In personal computers a chipset is a group of integrated circuits that together perform a particular function. System purchasers generally think in terms of the processor itself (such as a Pentium III, Pentium IV, or competitive chips from AmD or Cyrix). However they are really buying a
system chipset that includes the microprocessor itself (see
micRopRocessoR) and often a memory cache (which may be part of the microprocessor or a separate chip—see cache) as well as the chips that control the memory bus (which connects the processor to the main memory on the moth- erboard—see bus.) The overall performance of the system depends not just on the processor’s architecture (including data width, instruction set, and use of instruction pipe- lines) but also on the type and size of the cache memory, the memory bus (RDRAm or “Rambus” and SDRAm) and the speed with which the processor can move data to and from memory.
In addition to the system chipset, other chipsets on the motherboard are used to support functions such as graphics (the AgP, or Advanced graphics Port, for example), drive connection (EIDE controller), communication with exter- nal devices (see paRallelpoRt, seRialpoRt, and usb), and connections to expansion cards (the PCI bus).
At the end of the 1990s, the PC marketplace had chip- sets based on two competing architectures. Intel, which originally developed an architecture called Socket 7, has switched to the more complex Slot-1 architecture, which is most effective for multiprocessor operation but offers the advantage of including a separate bus for accessing the cache memory. meanwhile, Intel’s main competitor, AmD, has enhanced the Socket 7 into “Super Socket 7” and is offering faster bus speeds. On the horizon may be com- pletely new architecture. In choosing a system, consumers are locked into their choice because the microprocessor pin sockets used for each chipset architecture are different. Further Reading
Intel. “Desktop Chipsets.” Available online. URL: http://www. intel.com/products/desktop/chipsets/. Accessed June 6, 2007. “motherboards.” Available online. URL: http://www.motherboards.
org/index.html. Accessed June 6, 2007.
Walrath, Josh. “Chipsets Today and Tomorrow.” ExtremeTech. Available online. URL: http://www.extremetech.com/article2/ 0,1697,1845493,00.asp. Accessed June 6, 2007.