X
Home & Office

No Requiem for Moore's Law

Will technology keep extending processor capacity and bandwidth? Does it even matter? Yes, yes, and yes.
Written by John Blackford, Contributor
These days, computers run so fast that worries over computational limits pale before concerns about sufficient bandwidth to deliver advanced Web services. Still, chip designers have not been able to guarantee continuous advances much beyond 2010, when integrated circuits will approach fundamental limits.

Three issues will constrain next-generation (and beyond) computing and communications: the CPU and related components, bandwidth to link individual PCs and intelligent devices, and the applications computers can deliver.

It would be premature to speculate about all this—except that advances just announced suggest that processor and bandwidth limits will not constrain progress for at least 20 years. This raises provocative questions about what we'll do with all those MIPS.

For 36 years, Moore's Law has accurately predicted a doubling of transistor density every 12 to 24 months in integrated circuits. Intel's latest processor, the Pentium 4, delivers 1.5GHz with 42 million transistors—enough horsepower to drive streaming video on the Web, real-time video encoding, 3D visualization, speech recognition, and voice over IP. Clearly, today's high-end processors are capable of handling sophisticated computational tasks.

On the cable side, the issue is more about deployment than technology, because fiber-optic cables and routers can move truly stunning amounts of data. The optical advantage stems from the fact that photons are without mass, unlike electrons. With optical cable, there's no set limit to the number of photons you can cram into a particular fiber, while only a certain number of electrons can use a single circuit.

In fact, one bottleneck for optical communications at present is that the fibers pass through standard electronic processors for amplification and switching. Had those circuits been optically based, communications today would face even fewer constraints.

Optical fiber handles vast amounts of data by multiplexing several wavelengths of light into a single fiber. On the backbone side, providers this year expect to install fibers able to deliver 40Gbps per wavelength, or the equivalent of 750 complete channels of streaming video. Each fiber can handle eight or more wavelengths, and cables hold 100 or more fibers per cable. In round numbers, trans-Atlantic-type cables deployed in the near future will be able to transmit the equivalent of nearly 1 million simultaneous television channels. Routers and switches being deployed can handle a similar load.

It's where the backbones step down to individuals that bottlenecks occur. At present, many Web users have no choice but to employ analog phone lines. Over the next few years, DSL, cable, and satellite providers will boost the bandwidth for home users by a factor of 10. In the long term, communications bandwidth will be a function of economics, not technology.

On the CPU side, road maps from chip makers suggest that by 2010 we'll see at least 10GHz CPUs, with circuit widths as low as 0.05 micron. Many designers believe that delivered computer power will be 100 times that of today's 1GHz CPUs—not the mere tenfold increase implied by the clock-speed advances alone—because of greater parallelism and more sophisticated processor architecture.

With such hardware, we'll see widespread use of virtual personalities in all kinds of man-machine interfaces, such as virtual travel agents or training instructors.

However, a recent breakthrough just announced by IBM pushes projections for circuit widths to 0.01 micron (10 nanometers) and below. The advance, known as V-Groove, allows circuits to be etched in a V-shape, reducing electrical crosstalk. Even without the help of optical, molecular, or atomic computing, that takes us beyond common-sense projections and into a realm in which computer power is sufficient to tangle with issues such as which capabilities are uniquely human.

After deploying the full range of multimedia, we'll move to continuous speech recognition and real-time translation, plus Web-linked e-paper and advanced medical diagnostics. These advances will only get us to 2010. Since we can now project beyond the year 2020, we expect by then CPUs of 100GHz or faster, with terabytes per second delivered to the home.

That's when we'll see machine capabilities begin to overlap human ones. By 2020, computer and communications technology will be embedded everywhere, not just in toasters and televisions, but also in floors, sidewalks, and roads. And these entities will be constantly connected to the Web. Computer-aided devices will provide sight to the blind and hearing to the deaf; artificial personalities will participate in nearly all routine activities, such as travel reservations, shopping, and information seeking. Eventually, computer technology may be implanted in humans, for the ultimate "always-on" experience.

In short, the next couple of decades are likely to be the start of a wild ride. Want to come? john_blackford@zdnet.com

Editorial standards