Clayton Hallmark | 08.08.2006 04:48 | Analysis
THE END OF MOORE'S LAW
This is a big deal, and there is enough data now to predict the time and the end result -- maximum number of transistors on a chip -- using statistical methods. I have done that in the graph.
Please scroll through the illustrations now to actually see the point of this article. You live under Moore's Law whether you know it or not. (See "MOORE'S LAW ENDING" GRAPH.) You will know -- feel, like an earthquake -- when it ends. Your computer and economy depend on Moore's Law, which says the maximum number of transistors on an IC chip doubles every 2 years. (See EXPONENTIAL-GROWTH GRAPH.) Herein, I predict that the ultimate number of transistors on a chip will be less than 10 billion.
There are physical obstacles to shrinking transistors further, but economics -- commoditization (cheap, generic computers) -- will get us first. (See TETRAD FIGURE ON SEMICONDUCTORS.) Believe it or not, the fickle American consumer will be demanding economy, mobility, and Internet-based services IN PLACE OF more computer performance and upgrades.
The emerging markets of India and China, half the world's population, will create a humungous demand -- and source -- for cheap computers. These Asian notebook PCs will cost less than $200 to own, and Americans will demand them, too, at grave danger to Intel, Microsoft, and Dell. (See PHOTO, $150-200 COMPUTER.) The WTO will try to slow this, but it can't stop it.
OPEN-SOURCE SOFTWARE MADE IMC POSSIBLE--WILL RISE AS MSFT AND INTEL FALL
"Windows is no good for inernational collaborative development, whereas Active is based on Linux free software (which is all developed internationally by volunteers, so it's ideal) so we set up a whole new Active project to develop the Active Sydney software into something that could be used for the Seattle IMC. It was a big rush to get it ready in time, and there was a lot of seat of the pants stuff to keep it running under the skyrocketing audience as the WTO protests took place."
This is how Matthew Arnison of Sydney, Australia, described events of 1999 in a 2000 interview. Not only the corporate media had to be bypassed to report on the 1999 protests, corporate software -- Windows -- did too. Fortunately, that is going to get easier as Windows dominance falls with the Wintel (Windows-Intel) monopoly. Moore's Law http://home.earthlink.net/~moores-law/ is closely tied to this monopoly, as it really is Intel's Law. It has always been their business plan more than anything else. The Wintel monopoly, every bit as much as WTO, should have been the object of protests long ago. http://home.earthlink.net/~moores-law/
THE END OF MOORE'S / INTEL'S LAW AND MICROSOFT'S LAWLESSNESS
Moore's Law, popularized by Gordon Moore in 1965, became the basis of Intel's business plan when Moore and Robert Noyce founded Intel in 1968; and long ago the semiconductor and computer industries (numerous companies) adopted it as their own business plan, or "roadmap." The law never was considered a real law that expressed the way things have to work, like Newton's law of gravitation or even the economics law of supply and demand. In the future, the market demand will be for a vastly larger number of processors (remember the masses of new users in the developing world) but, generally, NOT MORE POWERFUL AND COMPLEX ones. So the end of Moore's Law about IC complexity is at hand. Even though it was really just a rule of thumb and a planning tool, its end will have a huge impact on the world and on individuals like you. There might be another Tech Wreck. Some big companies might be destroyed. But then, finally, computing will be unshackled from the Intel-Microsoft combine (Wintel).
MOORE'S-LAW-ENDING GRAPH -- RED LINE
Refer to the graph. One dashed line shows transistors-per-chip doubling every 18 months, as often stated for Moore's Law -- though Moore himself never said 18 months. The alleged 18-month doubling has been extended, by various writers, from least-cost components to maximum achievable transistor count, to processor and computer power (rate of doing computing work). Conversely, 18-month halving has been applied to prices for computers and computing power. Actually, data, even for transistors-per-chip, do not fit this "curve." Where the 18 months probably came from was the average of Moore's original 1965 estimate of 12 months and his 1975 estimate of 24 months.
Another dashed line shows maximum transistors on a chip doubling every two years, just as Moore planned in 1975 and actually a fairly accurate representation for 1971 to 2006, based on the chips chosen. In 2006, Intel Corp. is spending billions of dollars in the midst of switching plants over to the 65-nanometer fabrication (manufacturing) technology "node" (step), from 90-nm. Basically such switches every two years keep chips moving up the complexity scale -- and following the "curve" more or less -- by allowing printing of smaller transistors and other features (e.g., 65 nm, by a "line + space" measure, instead of 90 nm) on silicon chips. http://home.earthlink.net/~moores-law/
At the same time, Intel has announced a $1-billion spending cut for 2006. Something's got to give since it is Intel's capital spending that propels complexification per Moore's Law (its traditional business plan).
THE RED LINE
To draw the red line representing the slower future pace of Moore's Law, I have performed a statistical analysis based on a logistic-regression model and past data for Intel chips. I am confident that statistics, not physics, is the way to look at the future of Moore's Law. We have plenty enough experience now to know what the outcome will be. I will explain later how IC complexity follows a logistic (natural-growth) S-curve, with a predictable limit, as in biology -- instead of the ever-upward curve (J-curve or hockey-stick curve) assumed by Moore's Law adherents. Logistic curves are useful for forecasting in many fields. No matter how much money Intel and the U.S. government throw at the problem of making more complex chips, the returns are in and we can now know approximately the limit of semiconductor complexity and when it will arrive (same as for peak oil production).
The graph suggests that the slowing of Moore's Law should soon become evident. The sure, visible signs will be product delays, especially in nanotechnology generations (nodes), from Intel, the proprietor of the law -- and other bad news (financial, market share, etc.) from Intel, Dell, and related companies.
ELECTRONICS INDUSTRIES GETTING TOO BIG FOR THEIR BREECHES--
FROM J-CURVE TO S-CURVE (See EXPONENTIAL-GROWTH GRAPH.)
Exponential growth, when plotted on a "normal" graph, produces a J-shaped (or a hockey-stick-shaped) curve. (The previous, Red Line graph did not have a "normal" vertical scale but a logarithmic one, explained below.) Please refer to the EXPONENTIAL-VERSUS-NATURAL-GROWTH GRAPH and the sketch on its left. The exponential rises slowly at first, then ever more steeply (rapidly), until it goes almost straight up. Moore's Law and the growth of the semiconductor and PC industries appear exponential like this so far, though Moore's Law growth is faster (has a steeper curve) than the industries'. Moore's Law will soon break down for semiconductors, and some leading companies will fail; but the industries will keep growing (on cheap commodity computers, embedded processors, and so on).
To the right of the J-curve, you see the future of semiconductor complexity and related industries. Note that the formula for the S-curve, or logistic, is different than the simple exponential formula of Moore's Law. The J-curve soon will turn into an S-curve, or logistic curve, representing natural growth. As a human approaches late adolescence, growth slows down and eventually stops, as shown on the right. The same is true of technologies and industries. Each of them has its own particular S-curve. The slowdown will come sooner for IC complexity than for the industries. http://home.earthlink.net/~moores-law/
Eventually, as immature industries grow to represent a few percent the national economy (GDP), their growth stops being internally generated and apparently exponential (stops being like Moore's Law). This soon will prove so for semiconductors and computing (and other industries based on microprocessors and thus Moore's Law) as it has for other fields. Their growth soon will show signs of secular (long-term or permanent) slowdown, showing that what many assumed was an exponential is really just another example of natural growth, following an S-curve (logistic). It will be many years before the growth of these industries stops, but soon that IC complexity growth per Moore's Law stops. Eventually, Moore's Law will be remembered as merely part of a natural-growth S-curve involving physics, economics, and Intel's planning. Verhulst (Pierre, originator of the S-curve equation for growth) will win out over Moore.
MEGATRENDS IN ELECTRONIC AND RELATED PRODUCTS
In his famous 1965 article in Electronics Magazine, along with his famous law Gordon Moore proposed "handy" "sale"-priced computers. The CARTOON from Moore's article (please see it above or below) shows them being sold from a stall or booth much as Dell does in malls, but the emphasis was on small size and low price. A few years later, Moore and Robert Noyce founded Intel (Integrated Electronics) with what came to be called Moore's Law as their business plan. They partnered in the 1980s with Microsoft in a computer version of the automobile horsepower race because that was more profitable than commodity computers (which Moore had proposed in the 1965 cartoon) and because the two companies, especially Microsoft, found they could manipulate US patent laws to create a Wintel monopoly.
With computing and information-processing technologies (computers, semiconductors, and other technologies subject to Moore's Law) representing 5 percent or so of the national economy (GDP, gross domestic product) and with hundreds of millions of Asian users coming along, the industry is too big for the two brands to control it. So we are approaching the COMMODITIZATION stage of the outline below.
Most products, including electronic ones, go through the same stages shown below. Electronic products, particularly, also undergo continual miniaturization, as denoted by MIN. below and shown later in the radio example.
CHARACTERISTICS OF ELECTRONICS COMMODITY (final stage above http://home.earthlink.net/~moores-law/):
A. LOW PRICE: Price is the main specification.
B. ASIANIZATION: Radio manufacturing to Japan in late 1950s. Computers to China in early 2000s. Elsewhere as wages go up in China. Cheap labor often beats automation for cost-cutting.
C. MINIATURIZATION: Saves on materials in product, fuel for shipping across the Pacific, electricity/battery use.
D. UTILITY: A must-have like the light bulb, which everyone uses for reading and writing without paying much attention to it. Standardization (II above) allows it to be connected anywhere in a national grid. Computers are becoming a utility like this: Their grid is the Internet, and computer programs are coming to be used on the Internet on a pay-as-you-go basis, like electricity.
E. OUTGROWS PROPRIETORS: A small business outgrows its proprietors (Ben and Jerry have to hire professional managers). In the same way, computing is outgrowing the Wintel (Microsoft-Intel) partnership. Brands (like Intel and Microsoft) are losing their value, and related stock market prices will, too.
WTO AND PATENTS ALLOW MICROSOFT DOMINANCE
Of course the evolution of electronics now involves more than just hardware; so we turn to the software leader, Microsoft.
Basically the problem -- which I and others plan to get corrected -- is that U.S. federal courts did not have the technical (computer and engineering) competence to deal with the rise of consumer software, something still very new in 1981, the birth year of the IBM PC. The same year, in 1981, in the Diamond versus Diehr ruling, the U.S. Supreme Court ruled that software could be patented ALONG WITH a physical-processing device -- a rubber-curing press in the Diehr case -- that uses the software for control (of temperature, in Diehr).
In 1982 the U.S. Congress created the Federal Circuit court to hear patent cases. This court issued rulings that made patents in general (not just software) stronger and harder to challenge. Software algorithms still can't be patented by themselves under U.S. patent law. To circumvent the law, patent attorneys draft software patent applications to bundle an algorithm with a "device with memory," which a computer is, of course. Eventually the Supreme Court might deal with this "clever draftsmanship," as it's been called.
There is some hope for court relief in the U.S., as software patents really are contrary to the statutory prohibition against patenting abstract ideas (which software is, when considered minus any materials-processing machinery like Diehr's). Of course if the Supreme Court ever invalidates software patents, the software industry's lobbyists probably can bribe Congress to pass laws specifically legalizing such patents. There is more hope in Europe, India, and other places where software is not patentable (see http://www.nosoftwarepatents.com/en/m/basics/index.html) and U.S. lobbyists are less dominant. Despite bullying by the U.S. Trade Representative office and its partners like Steve Ballmer of Microsoft, people in these places can easily circumvent or ignore U.S. software patents. http://www.allamericanpatriots.com/m-news+article+storyid-15176.html I suppose it is only fair, if Microsoft is to circumvent the laws and the Constitution to patent software, fair for others circumvent the patents.
One of the main organizations that Microsoft and the U.S. Trade Representative work through is, of course, the World Trade Organization (WTO).
THE RADIO EXAMPLE
Long before iPods, little transistor radios gave a generation of youth a way to take their music, which soon (1955) became rock and roll, with them and away from their parents. Bardeen, Brattain, and Schockley had invented the transistor in 1947, but Haley and Presley finally made it sell (helped by the movie "Blackboard Jungle").
LOOK, MA, NO HANDLES! (FIRST RADIO TO HOLD IN THE PALM OF YOUR HAND)
The first commercial transistor radio was American-made, not a Sony. The Regency TR-1 of 1954 (SEE REGENCY TR-1 PHOTO), had four transistors. (Note the printed-circuit board. Today, inside ICs, the connecting "wires" are still printed, but the transistors are too, fully harnessing the power of printing.) Most of these were not transparent but red, etc. The shirt-pocket size Regency was 5 inches high and weighed 12 ounces. The old tube Atwater Kent model 20 was 26 inches wide and weighed about 18 pounds (a "compact" model 20-C version was 20-inches wide and weighed about 14 pounds). The Regency transistor radio was 24 times lighter and 4 or 5 times smaller than the Atwater Kent tube radio of 30 years earlier.
Of course the transistor radio, with only two controls, was vastly simpler to operate, just as computers will get.
Some stages in the development of portable radios were as follows. Each stage was like a standard size or generation, and many models were made in each. Smaller tubes, followed by transistors, made successive generations of smaller radios possible. http://home.earthlink.net/~moores-law/
One could trace a similar pattern in portable computers, from barely luggable to too small for touch typing. The shirt-pocket radio is about as small as possible for easy use. The videotape-size computer (see $150-200 COMPUTER PHOTO) is about as small as possible for touch typing (rapidly, without watching the keys). With a foldable screen and foldable keyboard, it could be made about the size of a paperback dictionary.
The transistor radio was expensive at first. The pioneering Regency TR-1 cost $50 in 1954, or $364 in current (2006) dollars. The Atwater Kent model 20 cost $100 in 1924, equivalent to $1145 now. So the Regency transistor radio cost less than one-third ($1145/$364 = 0.318) what the Atwater Kent tube set of 30 years earlier had, in constant, inflation-adjusted dollars.
The Regency was of course simple/easy-to-use. Its only two controls were TUNING and VOLUME. It didn't require the hookups of the old AK -- external-speaker, antenna, ground, and battery wires. (Similarly, notebook are doing away with computer cables.)
Six years later (1960) Japanese manufacturers, following the theme of smaller-cheaper, were offering in the U.S. very small sets (4 inches high) selling for $20, or $132 in today's money, or about 12% of the price of the old Atwater Kent. U.S. sets still were higher, about $30 minimum. The cheap Japanese radios became ubiquitous, and cheaper, and the first large-scale Asian electronics industry was launched. Brand names no longer mattered much, and the transistor radio soon became a commodity.
Transistor radios have long since become so cheap as to be disposable. That time will come for computers, too. Fifty-two years after the Regency came out, one can buy a similar radio (ICs have replaced discrete transistors), with FM added, for $4 ($2 more for one with two speakers) versus, in today's dollars, $364 for the Regency and $1145 for the tube AK. A pen-style FM radio (no speaker) with pushbutton digital-scan tuning runs a dollar -- over 1000 times cheaper than the AK and incredibly better. Here we clearly are near the floor of the S-curve for price -- you can sell no cheaper than zero.
The price floor for computers will be higher, but $100-200 computers will revolutionize computing and the related industries.
ECONOMISTS "GOT A GOOD THING GOING ON" WITH MOORE'S LAW
(HEDONICS, OR "BANG FOR THE BUCK")
It is common to see Moore's law used to claim that computer prices halve every 18 months.
Actually, the average unit price of PCs has declined much less. (Refer to the bars on the right in the PRICE-DECLINES BAR CHART.) Average unit prices -- the price-tag prices that people actually pay per computer -- declined at an annual rate of only 4.6 percent for mobile PCs and 8.7 percent for desktop PCs in 1995-99 (per a 2000 report by Berndt and Rappaport). http://www.bea.gov/bea/ARTICLES/BEAWIDE/2000/1200hm.pdf There is one unit price that has fallen significantly, and that is for transistors. Transistors cost $8 apiece in 1953, the year before the first transistor radio. At that rate, the transistors in a Montecito microprocessor would cost almost $14 billion (yes, billion: 1.72 billion x $8). Clearly, there is no reason for today's high prices of computers.
Hedonic Prices: Literally "hedonic" means "relating to pleasure," as in the Billy Paul song "Mrs. Jones" ("We got a good thing going on"). To economists it means utility, usefulness, value, or quality. Hedonic prices are prices based on formulas developed by economists as opposed to unit prices that people pay. The formulas FUDGE prices to compensate for quality increases. Hedonic prices are "bang for the buck" prices that work like this: If you replace a 2-year-old computer that cost $2000 by a new one that also costs $2000, but if the speed and other characteristics (included in the formula) have improved twofold, then it is counted as approximately a $4000 computer in the GDP (gross domestic product).
Conversely, an economist using the same hedonic formula would assume that the price you would pay today for the same utility as in your old computer has declined 50 percent in 2 years. Hedonics is keeping your cost of living down, right?
According to one *hedonic* price study (that of Aizcorbe in 2000), the annual price decline in a similar period to that studied by Berndt was 31 percent for desktops and 26.3 percent for notebooks. (See left bars in PRICE-DECLINES BAR GRAPH.) Now this hedonic pricing is close to Moore's Law. To fall by a factor of 2 (to one half) in 2 years (Moore's Law), the price would have to fall about 29.3 percent per year (0.707 x 0.707 = 0.5 = 1/2).