Контрольная работа по "Английскому языку"

Microprocessors in 2020

Every 18 months microprocessors double in speed. Within 25years, one computer will be as powerful as all those in Silicon Valley today

by David A. Patterson

_W

hen I first read the table of contents of this special issue, I was struck by how many articles addressed computers in the 21st century in some way. Unlike many other technologies that fed our imaginations and then faded away, the computer has transformed our society. There can be little doubt that it will continue to do so for many decades to come. The engine driving this ongoing revolution is the microprocessor. These silicon chips have led to countless inventions, such as portable computers and fax machines, and have added intelligence to modern automobiles and wristwatch-es. Astonishingly, their performance has improved 25,000 times over since their invention only 25 years ago. I have been asked to describe the microprocessor of 2020. Such predictions in my opinion tend to overstate the worth of radical, new computing technologies. Hence, I boldly predict that changes will be evolutionary in nature, and not revolutionary. Even so, if the microprocessor continues to improve at its current rate, I cannot help but suggest that 25 years from now these chips will empower revolutionary software to compute wonderful things.

Smaller, Faster, Cheaper

rpwo inventions sparked the computer revolution. The first was the so-called stored program concept. Every computer system since the late 1940s has adhered to this model, which prescribes a processor for crunching numbers and a memory for storing both data and programs. The advantage in such a system is that, because stored programs can be easily interchanged, the same hardware can perform a variety of tasks. Had computers not been given this flexibility, it is probable that they would not have met with such widespread use. Also, during the late 1940s, researchers invented the transistor. These silicon switches were much smaller than the vacuum tubes used in early circuitry. As such, they enabled workers to create smaller—and faster-electronics.

 More than a decade passed before the stored program design and transistors were brought together in the same machine, and it was not until 1971 that the most significant pairing—the Intel 4004—came about. This processor was the first to be built on a single silicon chip, which was no larger than a child's fingernail. Because of its tiny size, it was dubbed a microprocessor. And because it was a single chip, the Intel 4004 was the first processor that could be made inexpensively in bulk.

The method manufacturers have used to mass-produce microprocessors since then is much like baking a pizza: the dough, in this case silicon, starts thin and round. Chemical toppings are added, and the assembly goes into an oven. Heat transforms the toppings into transistors, conductors and insulators. Not surprisingly, the process—which is repeated perhaps 20 times—is considerably more demanding than baking a pizza. One dust particle can damage the tiny transistors. So, too, vibrations from a passing truck can throw the ingredients out of alignment, ruining the end product. But provided that does not happen, the resulting wafer is divided into individual pieces, called chips, and served to customers.

 Although this basic recipe is still followed, the production line has made ever cheaper, faster chips over time by churning out larger wafers and smaller transistors. This trend reveals an important principle of microprocessor economics: the more chips made per wafer, the less expensive they are. Larger chips are faster than smaller ones because they can hold more transistors. The recent Intel P6, for example, contains 5.5 million transistors and is much larger than the Intel 4004, which had a mere 2,300 transistors. But larger chips are also more likely to contain flaws. Balancing cost and performance, then, is a significant part of the art of chip design.

 Most recently, microprocessors have become more powerful, thanks to a change in the design approach. Following the lead of researchers at universities and laboratories across the U.S., commercial chip designers now take a quantitative approach to computer architecture. Careful experiments precede hardware development, and engineers use sensible metrics to judge their success. Computer companies acted in concert to adopt this design strategy during the 1980s, and as a result, the rate of improvement in microprocessor technology has risen from 3 5 percent a year only a decade ago to its current high of approximately 55 percent a year, or almost 4 percent each month. Processors are now three times faster than had been predicted in the early 1980s; it is as if our wish was granted, and we now have machines from the year 2000.

Pipelined, Superscalar and Parallel

Tn addition to progress made on the -L production line and in silicon technology, microprocessors have benefited from recent gains on the drawing board. These breakthroughs will undoubtedly lead to further advancements in the near future. One key technique is called

SILICON WAFERS today (background) are much larger but hold only about half as many individual chips as did those of the original microprocessor, the Intel 4004 {foreground). The dies can be bigger in part because the manufacturing process (one stage shown in inset) is cleaner.

pipelining. Anyone who has done laundry has intuitively used this tactic. The nonpipelined approach is as follows: place a load of dirty clothes in the washer. When the washer is done, place the wet load into the dryer. When the dryer is finished, fold the clothes. After the clothes are put away, start all over again. If it takes an hour to do one load this way, 20 loads take 20 hours.

 The pipelined approach is much quicker. As soon as the first load is in the dryer, the second dirty load goes into the washer, and so on. All the stages operate concurrently. The pipelining paradox is that it takes the same amount of time to dean a single dirty sock by either method. Yet pipelining is faster in that more loads are finished per hour. In fact, assuming that each stage takes the same amount of time, the time saved by pipelining is proportional to the number of stages involved. In our example, pipelined laundry has four stages, so it would be nearly four times faster than nonpipelined laundry. Twenty loads would take roughly five hours.

 Similarly, pipelining makes for much faster microprocessors. Chip designers pipeline the instructions, or low-level commands, given to the hardware. The first pipelined microprocessors used a five-stage pipeline. (The number of stages completed each second is given by the so-called dock rate. A personal computer with a 100-megahertz dock then executes 100 million stages per second.) Because the speedup from pipelining equals the number of stages, recent microprocessors have adopted eight or more stage pipelines. One 1995 microprocessor uses this deeper pipeline to achieve a 300-megahertz clock rate. As machines head toward the next century, we can expect pipelines having even more stages and higher dock rates.

 Also in the interest of making faster chips, designers have begun to include more hardware to process more tasks at each stage of a pipeline. The buzzword "superscalar" is commonly used to describe this approach. A superscalar laundromat, for example, would use a professional machine that could, say, wash three loads at once. Modern superscalar microprocessors try to perform anywhere from three to six instructions in each stage. Hence, a 250-megahertz, four-way superscalar microprocessor can execute a billion instructions per second. A 21st-century microprocessor may well launch up to dozens of instructions in each stage.

 Despite such potential, Improvements in processing chips are ineffectual un-

CLEAN ROOMS, where wafers are made, are designed to keep human handling and airborne particles to a minimum. A single speck of dust can damage a tiny transistor.

less they are matched by similar gains in memory chips. Since random-access memory (RAM) on a chip became widely available in the mid-1970s, its capacity has grown fourfold every three years. But memory spe'ed has not increased at anywhere near this rate. The gap between the top speed of processors and the top speed of memories is widening.

 One popular aid is to place a small memory, called a cache, right on the microprocessor itself. The cache holds those segments of a program that are most frequently used and thereby allows the processor to avoid calling on external memory chips much of the time. Some newer chips actually dedicate as many transistors to the cache as they do to the processor itself. Future microprocessors wffl allot even more resources to the cache to better bridge the speed gap.

 The Holy Grail of computer design is an approach called parallel processing, which delivers all the benefits of a single fast processor by engaging many inexpensive ones at the same time. In our analogy, we would go to a laundromat and use 20 washers and 20 dryers to do 20 loads simultaneously. Clearly, parallel processing is an expensive solution for a small workload. And writing a program that can use 20 processors at once is much harder than distributing laundry to 20 washers.