Here's the long and short of coming computer technology: Silicon will be the primary building block of processors and memory for the next 20 years.
Here's the long and short of coming computer technology: Silicon will be the primary building block of processors and memory for the next 20 years. We'll start seeing a few optical components creeping into computers in a few years. But all-optical computers, those that operate at the speed of light, are probably a couple decades away.
Predicting computer technology in the near term is a snap the boxes will be smaller, faster, and almost everywhere. Faster means higher clock speeds. Chip-maker AMD, for instance, has already released a 1,000-MHz microprocessor they've dubbed Althon. Not to be outdone, arch rival Intel has released 933-MHz ICs, and is touting Willamette, the company's room-temperature 1.5-GHz design code.
That computers will continue to get smaller is evident in ever-lighter, ever-more powerful portable designs that measure less than an inch thick. More evidence is in the recent introductions of a raft of palm-sized devices and compact telephones that also link to the Internet and look like Captain Kirk's communicator.
Some computer components will be difficult to shrink, particularly user-inter-face devices keyboards, screens, and mice. These may become less obvious with tiny screens that clip onto glasses, improved voice-recognition software that lets you tell computers what to do, and cursors controlled by sensors that see where you are looking. The Air Force is working on eye-controlled computers that will keep pilots' hands free to fly the plane. It's not hard to imagine these devices making their way into the civilian sector.
A CLOSER LOOK
Smaller hardware may not be feasible in the next decade. It might be easier and more economical to harness parallel schemes that bring more than one processor to bear. Parallel-processing schemes are not new, but they will become more prevalent as the methods are perfected and more engineering software is rewritten for parallel execution.
Software companies already have several parallel processors commercially available. MSC.Software, Los Angeles, for example, has developed a clustering scheme that ties together two or more computers. The network relies on Beowulf, a clustering software package developed by NASA, and will run a version of MSC.Nastran analysis software the company has optimized for parallel operation. Interestingly, the clustering software can run on two and three-year-old computers, effectively resurrecting aging technology into a personal supercomputer.
Another multiprocessor design, this one from SGI, Mountain View, Calif., and Ansys Inc., Pittsburgh., uses the Linux operating system. The two companies plan to unveil a server equipped with several Intel 64-bit processors soon after they are available. Ansys FEA software, which has featured parallel processing for several years, will be the major engineering draw for this device. The overriding idea is to let analysts set up problems on their desktop computer, then pass them to the server to solve. It should provide answers much quicker than a single-processor machine could and free up the desktop PC for other work.
On a processor level, technology is changing fast. For example, many manufacturers are exploring 0.18- technology. (The dimension refers to layer thickness in a processor). Thinner layers mean lower drive voltages and, therefore, less heat to dissipate. Intel reports that it's ready to begin experimenting with 0.13- layers, and thinner layers are certain to follow.
Intel also recently announced several advancements for its 32-bit Pentium 4 processor. It will soon have a so-called hyper pipeline to queue and execute instructions inside the processor at higher rates than those in Pentium IIIs. Meanwhile, extensions to Pentium III internal code will speed encryption and video, and support future Inter-net improvements.
Other processor advances put engineers at the dawn of widespread 64-bit computing with technology that will push silicon processors to their limits. Sixty-four-bit chips have been available from IBM, SUN, SGI, and Compaq, but only in their top-ofthe-line machines. Intel's version, Itanium, along with AMD's Sledgehammer will make such processors more common.
Itanium incorporates explicitly parallel instruction computing or Epic technology, according to Hewlett-Packard, Intel's development partner on the chip. HP says Epic will eventually replace today's Risc and Cisc (reduced and complex instruction set computers) technology. Although Epic is the first of many commercial 64-bit processors, it will also be 32-bit compatible so that it can handle current engineering software as well as that written for 64-bit packages.
Itanium's initial clock speeds are expected to fall in the 500-MHz range, which is below that possible with today's Pentium IIIs, but its throughput should far outdistance existing processors. One industry observer, for example, estimates Itanium will process 3D-graphic tasks in only 5% of the time needed by a Pentium Pro. Intel says Itanium will hit 3-Gflops (3 10 9 floating point operations/sec, or three times the speed of competing 32-bit chips) working at extended precision, or 6-Gflops at single-precision performance. Its faster processing rate comes from chip features which translates sequential code into parallel 128-bit bundles. Essentially, the processor does more than one operation per clock cycle.
Equally important for future engineering tasks will be the 400-MHz system bus on the Pentium 4 and other 32-bit systems. This internal network will send data throughout the computer faster three times the rate of the 133-MHz bus on current Pentium III-based computers. As fast as that may sound, the current bus technology the ribbon of wires that connects a processor to memory, disk, and outputs will have to give way to wider data paths, and eventually to optic fibers which should yield bandwidths well beyond those of current designs.
The first computers with optic components could appear as soon as the next decade. The light-operated devices will eliminate much of the wiring that makes up today's idea of a bus. For instance, 128-bit processors are predicted to be available by 2025. They would need standard busses with 128 wires running from processors to other components. Its not hard to imagine the tangle of wires that might fill a computer if it had to use existing design techniques. Optic components seem a likely alternative.
But there are several problems that must be solved first. Today's computers pass information as bits described by electrons. Optic components use photons little packets of light. Engineers are now looking for a good method and a reliable device that will transform electrons to light, and back.
Frazier and coworker's latest effort has been to find organic (hydrocarbon based) thin films that respond more vigorously to light than existing materials. Inorganics, such as silicon, would serve as a base for organic materials and allow building devices using both photons and electrons in hybrid computer systems, which will eventually lead to all-optical computer systems. Frazier says the lab work of his team could pay off in fast, small, lightweight, and lower-cost computers and optical communication devices.
"We are trying to synthesize organic materials that interact rapidly with light.
This involves engineering thin-film microstructures that react to light in an optimized way," says Frazier. One set of promising materials are called polydiacetylenes. "Certain functional groups can be added or subtracted from the material to improve optical switching. But those might wear out quickly so the material must be robust enough to withstand light signals, and in some cases bear extrusions necessary for manufacturing," says Frazier.
One manufacturing method, developed by Mark S. Paley at Marshall, deposits a polymer (many units of a molecule to form a long chain) film on a substrate by irradiating a solution containing monomer (single units of polymer) through the substrate into the solution with a particular wavelength of light. "We can paint patterns on substrates giving it light-guiding properties and a fast response time," says Frazier. These devices would function where an optical fiber ends and electronic transmission begins.
NASA is also studying application of films in the microgravity of space because it minimizes convection in the films and therefore the amount of embedded aggregates. "In some cases we can enhance polymer chain alignment and produce better electro-optic material. This would have near-term applications," says Frazier. Aggregate clumping caused by convection in the thin films limits the efficiency of the wave guiding devices. For now, space-grown films are "benchmark" materials designed to expose gravitational effects on processing in the absence of gravity induced convection.
These developments are letting researchers think beyond Moores law, the observation by Intel founder Gordon Moore that suggested the number of transistors on a processor (and hence, its speed) would double every 18 months.
THE SHAPE OF THINGS TO COME
Several design companies responded to Intel requests to conjure up the shape of future computers. The accompanying images show outside styling as advanced as the processors they're designed to hold.
The Gozinta from AndersonDesign Inc., Plainville, Conn., for example, makes connections easy by taking advantage of the strides made in USB and Firewire ports over the last several years. Firewire is a protocol for fast transfers of graphic and video files.
"Computers are so fast today, many buyers look for easier ways to add peripheral equipment, such as scanners, printers, and speakers, rather than pure performance" says Bob Marvin, director of research with Anderson Design. "That was a design goal for the Gozinta. It has ten USB ports and two Firewire ports."
Another goal was reducing manufacturing overhead. "The small tower behind the main unit has two card slots with more USB ports," says Marvin. Intel wanted to build the computer as a legacy-free unit, partially by eliminating ISA and PCI slots, which date back to the 1980s. This reduces manufacturing costs because there are fewer upgradable components.
Tom Draper Design