Livermore, IBM to build world's fastest supercomputer

Lawrence Livermore has awarded IBM a $93-million contract to build the world's fastest supercomputer-a machine designed to deliver 3 trillion calculations per second (3 teraflops). The 3-teraflop IBM RS/6000 SP production model is scheduled for demonstration in December 1998.
Most of today's supercomputers are capable of performance in the gigaflop or billion-calculations-per-second range, notes Dave Cooper, Associate Director for Computations at Livermore, who suggests that "If computing at gigaflops were represented by a one-way trip between San Francisco and Los Angeles, then computing at teraflops represents a round trip between San Francisco and the moon in the same amount of time."
The Livermore-IBM collaboration is the latest project in the Department of Energy's 10-year, $1-billion Accelerated Strategic Computing Initiative program. The program, which involves three major DOE facilities (Livermore, Los Alamos, and Sandia national laboratories), calls for five generations of high-performance computers. Development of the first generation ASCI machine-a 1.8-teraflop computer-is under way, teaming Sandia with Intel Corporation. The three other supercomputers that ASCI wants developed will deliver calculation capability in the 10-, 30-, and 100-teraflop range.
In discussing the IBM contract in late July, Lawrence Livermore Director Bruce Tarter said that he expects this project to provide Livermore and the other national laboratories involved in ASCI "with an opportunity to lead the next revolution in high-performance computing in support of important national security objectives."
ASCI supercomputers will be used to construct three-dimensional simulations of stockpiled nuclear warheads, allowing DOE scientists to analyze effects of weapons aging and the implications of potential problems. According to David Nowak, head of the ASCI program at Livermore and a former weapon designer, "Numerical simulation is the glue that holds stockpile stewardship together."
Advanced computations, specifically three-dimensional modeling and simulation capability, are central components of "stockpile stewardship"-the safe and reliable maintenance of the nation's nuclear arsenal in the absence of nuclear testing. Tera-scale computing also will provide a commercial platform for medical simulations, global climate modeling, aerospace and automotive design, and other applications.
The system will be based on a building block approach to high-performance computing in which the system consists of clusters of shared-memory processors. The new supercomputer will be designed to accommodate as many as 512 nodes with eight processors per node to perform numeric- and data-intensive tasks. It will be delivered with 2.5 trillion bytes of memory, as compared with existing high-end systems in which the memory is measured in billions of bytes or with personal computers where the memory is measured in tens of millions of bytes. It can be easily switched between a secure environment for national security applications and an open environment for university collaboration.
According to Nowak, the system naturally scales to the 10-teraflop range by increasing both the processor speed and the number of processors per node.
Contact: Dave Nowak (510) 423-6706 (nowak1@llnl.gov).

Lab receives six awards for R&D innovation

R&D Magazine will honor Laboratory researchers in Philadelphia next month at its annual R&D 100 Awards dinner. Each year, the magazine selects the top 100 commercially viable research and development advances to receive its coveted award, considered by many the R&D community's "Oscar."
This year, the magazine cited six technologies developed at Lawrence Livermore for R&D 100 honors. That brought to 61 the number of awards the magazine has bestowed on Laboratory R&D teams since 1978, when Livermore began participating in the competition.
Two of our award-winning developments this year involve industrial partnering agreements. Five stem from research pursuits in our Laser Programs.
Alan Bennett, director of our office of Industrial Partnerships and Commercialization, described this year's R&D 100 honors to the Laboratory as "a tribute to the ongoing excellent scientific and technical work that routinely takes place here." Said Bennett: "Once again, Laboratory scientists have shown that ideas developed in the course of our mission activities may have tremendous benefits for the U.S. economy." Here is a brief look at the Laboratory's 1996 R&D 100 Award winners. Each will be reported on in detail in the October issue of S&TR.

Latest MIR use: the electronic dipstick
The Lab's Micropower Impulse Radar (MIR) technology has once again been honored by R&D Magazine, this time for MIR's latest application: an "electronic dipstick" to sense the level of fluid or other material stored in tanks, vats, and silos. The dipstick also can be used in automobiles to read levels of a variety of fluids: gasoline, oil, transmission fluid, coolant, and windshield cleaner.
The electronic dipstick is impervious to condensation, corrosion, or grime on the sensor element, a simple metal strip or dipstick-like wire several inches to dozens of feet long, depending upon the application. The system's electronics are based on low-cost components that fit on a small circuit board. Cost is so low that one licensee of our technology plans to retail gas-cap-mounted dipsticks at $6 each.
Since development of MIR, initially invented as a diagnostic sensor for use by our Laser Programs, 30 patents have been applied for, 16 of these have been granted to date, and hundreds of commercial applications have been identified. The Laboratory has been following the dual paths of licensing the technology to qualified manufacturers and developing programs that use the technology in support of our missions.
Like conventional radar, MIR works by sending out a pulse and measuring its return. In MIR, however, each microwave pulse is less than 5 billionths of a second in duration; an MIR unit emits about two million of these pulses per second. Because current is only drawn during this short pulse time, power requirements are extremely low. One type of MIR unit can operate for years on a single AA battery.
Contact: Tom McEwan (510) 422-1621 (mcewan1@llnl.gov).

SixDOF sensor provides manufacturing flexibility
A small, noncontact optical sensor developed by the Laboratory will increase flexibility in manufacturing processes that employ robots. This invention does so by eliminating the time-consuming and expensive process of "teaching" robotic machinery new motions every time manufacturing changes are required.
Mounted on the tool head of a multi-axis robot arm, our six-degrees-of-freedom device (dubbed SixDOF) can sense its position relative to a workpiece, allowing the robot to autonomously follow a predescribed machining or manufacturing path. As the device's name indicates, SixDOF senses its position in all six degrees of freedom (the x, y, and z axes as well as the turning motion around those axes).
SixDOF is 250 times faster, 25 times more accurate, and one-sixth less expensive than its nearest competitor, which can detect only three degrees of freedom. The sensor works by emitting a laser beam and detecting the reflection off reference points mounted on the workpiece. Inside SixDOF, the beam is split and directed onto three photo diodes. The analog signals from the diodes are digitized and fed into a computer that can instruct corrective action or provide position readings.
The sensor could be used to control a six-degrees-of-freedom computer mouse, to assemble large and complex parts automatically, or to perform dangerous tasks remotely.
Contact: Charles Vann (510) 423-8201 (vann1@llnl.gov).

Optical crystal delivers tunable ultraviolet laser
Through an industrial partnership forged with II-VI Corporation of Tarpon Springs, Florida (formerly Lightning Optical Corporation), Lab scientists have developed and commercialized a new optical crystal-Ce:LiSAF-that, for the first time, makes an all-solid-state, directly tunable ultraviolet (UV) laser commercially viable.
The crystal consists of lithium-strontium-aluminum- fluorite (LiSrAlF6) doped with cerium (Ce), a rare-earth metal. Before the availability of this lasing medium, generating tunable UV light required multiple complex and sensitive nonlinear optical conversion steps. Ce:LiSAF eliminates these deficiencies.
Coupled with a compact solid-state laser, Ce:LiSAF yields a practical, robust laser system. Producing UV light directly and efficiently, it greatly expands laser applications; directly generating such light in a wide enough color band to provide straightforward "tunability" extends laser uses even more.
Ce:LiSAF lasers are particularly well suited to remote sensing applications. For example, a Ce:LiSAF laser could be used remotely to detect ozone and sulfur dioxide in the environment. Additionally, Ce:LiSAF lasers could be used to locate biological weapons remotely by detecting the presence of tryptophan, a common component of such weapons. The ultraviolet tunability provided by Ce:LiSAF also could be the basis for development of a UV differential absorption Lidar (laser radar) system. Another potential application is secure wireless communication links between infantry units in the battlefield.
Contact: Christopher Marshall (510) 422-9781 (cmarshall@llnl.gov).

Sensor offers greater performance, reduced cost
Teaming with the Read-Rite Corp. of Fremont, California, Laboratory scientists have developed an advanced magnetic sensor, a critical component in magnetic storage devices such as hard disk drives in computers. A typical disk drive in a computer would use 1 to 10 magnetic sensors.
Called the CPP-GMR (current-perpendicular-to-the-plane, magnetoresistance) sensor, this new invention offers greater sensitivity and 100 times greater storage densities than current commercial products. In fact, CPP-GMR storage density ranges from the current state of the art (about 1 gigabit per square inch) upward to the projected limit of magnetic disk drive technology (about 100 gigabits per square inch).
Built of alternating layers of thin magnetic materials and nonmagnetic materials, this giant GMR sensor uses thin-film technologies previously developed at Lawrence Livermore. Because the manufacturing process devised for the CPP-GMR does not require expensive fabrication tools, mass production costs can be kept low.
Because of the sensor's unique properties (it actually becomes more sensitive at the higher device densities needed for next-generation storage systems) and because of the simple manufacturing process employed, Laboratory scientists expect that magnetic heads using this sensor will enable the information storage industry to continue to develop higher density magnetic storage devices at reduced costs.
Contact: Andrew Hawryluk (510) 422-5885 (hawryluk1@llnl.gov).

Tiny optical amplifier offers big boost to signal speed
A dime-sized optical amplifier developed by the Laboratory has the potential for delivering a big boost to signals whizzing through 21st century data communications systems. This semiconductor optoelectronic device is designed to amplify optical signals at ultrahigh (terabit-per-second) rates. Such signal amplification or regeneration is essential in fiber optic communication systems where signals must be distributed over great distances and to a large number of customers.
The Laboratory's approach yields a component that is 100 times less costly, 1,000 times more compact, and more reliable than competing fiber amplifier technology. When produced in volume using present-day technologies, cost per unit is estimated to be as low as $500. As low-cost manufacturing techniques are developed, the cost is estimated to drop to about $50 per unit.
Essentially an optical analog of the electronic amplifier, which is ubiquitous in the electronic world, the Lab's miniature signal booster uses a tiny built-in laser system to eliminate crosstalk problems that have prevented deployment of more conventional semiconductor optical amplifiers.
Relying upon standard integrated circuit and optoelectronic fabrication technology, this device can be incorporated into many types of photonic integrated circuits. Potential applications include wide-area and local-area information networks, cable TV distribution, and computer interconnects.
Contact: Mark Lowry (510) 423-2924 (lowry3@llnl.gov), Sol DiJaili (510) 424-4584 (dijaili@llnl.gov), or Frank Patterson (510) 423-9688 (patterson6@llnl.gov).

Lithography system proves boon for FEDs
The Laboratory has made a significant advance in the effort to fabricate field emission display (FED) flat panels efficiently and cost effectively by demonstrating large-area laser interference lithography. Laser interference lithography is a way to precisely and uniformly produce regular arrays of extremely small (less than 100 atoms wide) electron-generating field emitter tips that are at the heart of FED flat panel screens.
FEDs represent an advance over conventional flat panel display technology used in a wide range of consumer and military products-from digital watches to portable computers. Field emission display panels consume less power than devices using competing active matrix liquid crystal display technology, a field dominated by Japanese companies. Compared to liquid crystal-based devices, field emission display panels can also be made thinner, brighter, lighter, and larger, and they have a wider field of view.
FEDs, however, have not been much of a player in the $8-billion-a-year international flat panel market (projected to more than double by the end of the decade). Their primary drawback: expensive and complex micromachining technology needed for their fabrication.
The Laboratory's cost-effective fabrication technique, however, has potential to assure successful commercialization of large-area, high-performance FEDs. Potential applications range from more efficient and energy-conserving portable computers to virtual reality headsets and wall-hugging TV screens.
Contact: Michael Perry (510) 423-4915 (perry10@llnl.gov).