Monday, June 05, 2006

Data Storage & Retrieval: Does Moore's Law Really Account for Better, Cheaper Technology?

The only thing in my life that consistently gets better, faster and cheaper has been digital technology. Even bread gets more expensive and not any better. Why is that, I wonder? Well, here is some of why. Since the early days of electronics, the industry has been pushing hard to store more stuff in a smaller space, which reduces their costs. They have been competing hard, which passes the cost-savings along to us (hooray! -- this works much better than antitrust settlements!)

Librarian types may already be familiar with Moore's Law. I will quote the nice explanation from Wikipedia's article (link) on the "law:"

Moore's law is the empirical observation that the complexity of integrated circuits, with respect to minimum component cost, doubles every 18 months[1].

It is attributed to Gordon E. Moore[2], a co-founder of Intel. However, Moore had heard Douglas Engelbart's similar observation, possibly in 1965. Engelbart, a co-inventor of today's mechanical computer mouse, believed that the ongoing improvement of integrated circuits would eventually make interactive computing feasible.

Well, that's a simplification, and there are lots of arguments out there. Moore himself argues about what, exactly, he said. The clearest statement in 1965 was in a 4 page article, "Cramming more components onto integrated circuits", Electronics Magazine 19 April 1965, written by Moore, but he never termed it a law, and apparently did not say 18 months, exactly. At any rate, there are two interesting articles, one at Ars Technica and one from First Monday, discussing Moore's Law and its application to our future. Also, you'll want to read about "perpendicular storage" on magnetic hard drives, which is what allows these new high density drives from Toshiba and Seagate.

Here is a teaser from the First Monday piece,
link which is very thought-provoking:

The balancing of supply and demand, however, eventually requires a perspective from which we can say something about the actual benefits of computing.

Over-investment in technology development is not necessarily a bad thing. Speculative bubbles of various sorts have played an important role in restructuring economy and in creating new unforeseen and unintended opportunities. Indeed, today we have much computing and information processing capability. The question is what are we going to do with it.

Moore's Law gave us a compact and a deceptively exact way to express beliefs in technological determinism. Later it became transformed to economic determinism, which argued that people would buy computers because they will be ridiculously cheap. Moore's Law also provided a convincing basis for arguing that the development of economies and societies is at the hands of technical experts. The fact that Moore's Law has so often been misrepresented and used together with contradictory evidence indicates that it has expressed strong and fundamental convictions about the nature of progress. Contrary to what its users have often claimed to say - that the history of semiconductors and computing has followed a well-defined exponential path - the rhetoric point of Moore's Law has been directed towards the future, determined by technological development and understood by the speaker.

I would recommend reading Jon Hannibal Stokes' article "Understanding Moore's Law" at Ars Technica link first. He does an excellent job, not only of laying out in detail the pieces of the law, but also the physics and practicalities of semiconductor manufacture that affect cost and storage density.

For instance, he shows graphically how manufacturers are able to lower costs by making more integrated chips out of a single wafer as they are able to make the components smaller. This means that the same number of inevitable defects on the wafer waste a smaller percentage of chips for each wafer. If the wafer can only make 16 chips and 4 are spoiled, 25% are wasted. If the wafer can be made into 64 chips and 4 are still spoiled, only 4% are wasted. The cost/chip has gone down. Stokes sees four factors driving the number of transisters (and thus cost of chips down) each year:

Ultimately, the number of transistors per chip that makes up the low point of any year's curve is a combination of a few major factors (in order of decreasing impact):

1. The maximum number of transistors per square inch, (or, alternately put, the size of the smallest transistor that our equipment can etch),
2. The size of the wafer
3. The average number of defects per square inch,
4. The costs associated with producing multiple components (i.e. packaging costs, the costs of integrating multiple components onto a PCB, etc.)

Improvements in chipmaking technologies focus on all of these factors in order to bring us higher levels of integration at lower costs. In the next part of the article, we'll talk about the wide variety of things we can do with higher levels of integration.

If you can improve any of the four factors, you can increase speed/memory/reduce costs. So you can add more transistors per inch. By doing this, you can either reduce the size of the chip, or increase the functionality of the chip, keeping the same size. You can increase the size of the original wafer, or increase the useable portion of each wafer. You can reduce the average number of defects on each wafer, the defect density. And/or you can reduce the costs of producing the chips.

Stokes points out that new computers (and I say many new operating systems) drive the need for more computing power. Microsoft Windows and other Windows software, in particular, have reputations as "bloatware," hogging huge amounts of memory that really is not necessary. At any rate, the increased chip power, when concentrated on a single motherboard, creates a great deal of heat. Eventually, high-density silicon chips in dense arrays will create a new need for cooling systems beyond what current computers provide. (The other problem with shrinking silicon chips, as reported here link
earlier, is electrical leakage when the silicon linkages get too small.) Finally, Stokes believes that Moore's Law will be tested or twisted by the rise of mobile computing:

Mobile computing demands that more functionality be added to a single die, but not for the purpose of increasing performance. Rather, an ideal mobile computing IC would combine nonvolatile memory, volatile memory, a CPU, and multiple types of wireless capabilities (Bluetooth, 802.11b, 802.11g, GSM, etc.) on a single die. The types of circuits that implement each of these functions are so different that they're very difficult to integrate on a silicon substrate. Some vendors are facing the challenge head-on by trying to integrate them anyway, while others are looking to new packaging technologies that combine multiple chips with multiple functions into a single module. Which of these two approaches will ultimately win out remains to be seen, but it is certain that the peculiar demands of the mobile computing space will have just as much of an impact on the way that Moore's Law is exploited as the demands of the server and desktop markets have had in the past two decades.

The interesting counterpoint to Stokes is the essay by Ilkka Tuomi, "The Lives and Death of Moore's Law," at First Monday link. Tuomi argues that Moore's Law has been a specious and misleading observation that only appeared to explain the phenomenon of technology growth. He argues that continuing to follow Moore's Law as a guide in planning for technology growth would be disastrous. Tuomi examines the empirical evidence to see how well Moore's Law corresponds to the actual data and finds that it really does not predict very well:

a simple test of Moore's Law would be to see whether the number of transistors actually follow an exponential growth curve. In other words, we can ask whether the number of components on a chip double regularly. A first approximation can be found using data provided by Intel. We can simply use data published on the Intel Web page that describes Moore's Law [22]. According to the provided data, the first microprocessor 4004 was introduced in 1971, with 2,250 transistors. The last available data point at the time of writing was Pentium 4, with 42,000,000 transistors. Without doubt, technology had advanced rapidly.

Using this data we can fit an exponential growth curve and see how good the fit is. Figure 2 shows the percentage differences in number of transistors from their predicted values. According to this simple fit, there were some five million transistors missing from Pentium II in 1997, representing about 70 per cent, and some 7.5 million, or some 18 per cent, too many in the Pentium 4 in 2000. The estimate for Pentium 4 comes relatively close to its actual value as the data, indeed, has been used to fit the exponential, and the final point is heavily weighted. The error would be greater if we were to use previous points to fit the curve and then predict the number of transistors of Pentium 4.

[Graph did not reproduce]

An alternative simple test for Moore's Law would be to see whether the number of transistors actually have been doubling every 18 months or every two years. One would expect that many people who refer to Moore's Law would have done this exercise. Actually, it seems to have been done relatively rarely. Indeed, one of the few exceptions is a comment written by Brenner (2001). It notes that a figure published in an earlier Physics Today article (Birnbaum and Williams, 2000) shows a doubling time of about 26 months, whereas the caption text claims that "Moore's Law is the empirical observation that the number of transistors on a single integrated-circuit chip increases by a factor of four every three years," implying that the data shows doubling time of 18 months.

Leaving aside for a while the question how well the above mentioned graph with 26 months doubling period reflects available evidence, it seems obvious that Moore's Law has become so commonly accepted that it is possible to publish articles that contain both its historically incorrect version and data that contradicts the provided version. The discrepancy between the standard version of Moore's Law and reality, however, should not be particularly difficult to see. Starting from 1971 with 2,250 transistors and using 18 months as a doubling time leads to about 1.4 billion missing transistors in year 2000. This represents about 3,400 per cent error. If one uses the two-year doubling period the error is less dramatic. Only about 10 million transistors are missing from Pentium 4. Pentium II had somewhat more missing transistors in 1997, corresponding to about 150 per cent too few transistors on chip.

Tuomi has many other arguments about why Moore's Law, as extended to processing time, or as used to explain why we are simply able to buy better technology for less money, is being mis-applied and misunderstood. In sum, however, he believes that we must avoid leaning on that formulation to predict future growth and needs in technology fields, and to look more clearly at the history to foretell our future.

And last, news about high-density hard-drives has been sloppily reported in my local paper, but here is a better report from C-Net link
First, Seagate announced higher density hard-drives, and now Toshiba claims the crown for highest density. The trick is perpendicular storage. Where most current hard-drives magnetize the tracks in a horizontal axis, which makes each track a bit wider, these are able to magnetize each track in a vertical axis -- perpendicular -- which takes less space per track.

By the way, you might notice that the hard-drive in the picture (which is a Seagate Momentus 5400.3, a 2.5-inch diameter hard drive link)
has a little arm. This can wiggle back and forth very quickly as the drive whirls at 5,400 revolutions per minute to read the data on the disk, stored perpendicularly. If they ever decide to speed up optical storage devices, like CDs and DVDs, they could do it first by making the laser that plays them move freely. Right now, they move on a screw, sequentially, across the disk, because the assumption is that most of what you'll ever store on these media will be music or video, and you'll want to read it sequentially. How annoying.

No comments: