Skip to main Content
IBM Think Research Online

Think Research Home

chip technology

data management

deep computing

e-business technologies & services

exploratory research


interfaces & accessibility

network technology

pervasive computing


servers & systems management



TR Print Archive

Research Publications

Research News

Featured Concept
Research News

Patents on the Web

A Grand Slam in Patents

New Chip Exceed Speed Barrier

Showcasing Solutions for Industry

Adventures in Software



Seeing the Light

Automatic Circuit Tuning

Patents on the Web

Chemical reactions involve more than just the chemical identity of the reactants. Some take place only when the reacting molecules are properly aligned. Scientists at IBM's Almaden Research Center and the University of California, Santa Barbara, have recently extended the understanding of geometry's role to a new arena: reactions at surfaces. In a pioneering experiment, they have found that molecules that spin parallel to an approaching surface, like helicopters landing, are more likely to react with the surface than molecules that approach perpendicularly with a cartwheel-like spin.

The work has basic and commercial objectives. "The ultimate goal is to understand chemical reactions at surfaces on an atom-by-atom basis," says Daniel Auerbach, an Almaden participant in the team. "That understanding is important for manufacturing processes in microelectronics and storage, and in general to vacuum processes that involve reactions with surfaces."

The team used a simple, well-characterized reaction, between molecules of deuterium - an isotope of hydrogen - and a copper surface. For technical reasons, they studied the reaction backward, monitoring how atoms of deuterium combine on a copper surface to form molecules that then leave the surface. Those observations showed the existence of a reaction barrier that can prevent arriving molecules from splitting into atoms, which attach to the surface, once they land. Molecules that rotate parallel to the surface as they approach, the team reports in the July 4 issue of Science, are up to three times more successful in reacting with the copper surface than those tumbling end over end.

That finding supports the idea that the barrier to reaction is smallest if both the atoms in the incoming molecule can form chemical bonds with the surface at the same time. In that case, explains Auerbach, the energy required to break the molecular bond between those atoms is immediately offset by the energy released as the atoms form new bonds with copper atoms on the surface.

Now, in the effort to gain deeper understanding of chemical reactions at surfaces, the team is seeking better data and more sensitivity for experiments on alignment effects for rotating molecules. The researchers are also studying the surface reactions of strongly vibrating molecules. "We're very excited about understanding the role of large amounts of vibrational energy in reactions at surfaces," says Auerbach. "We're hoping to find new and exciting chemistry with potential applications."

A Grand Slam in Patents

Last year, for the fourth straight year, IBM received more patents from the U.S. Patent and Trademark Office than any other company. Its total of 1,867 exceeded the second-place score by 329, according to figures released by IFI/Plenum Data Corporation. The mark also surpassed IBM's own one-year record for U.S. patents: the 1,383 that it received in 1995. More than a quarter-century's worth of U.S. patents, from IBM and other companies, are available on the company's new Patent Server site (see above).

Filing patents is more than a numbers game. For while no direct link exists between patents and the marketplace, awards of patents manifest the vitality of IBM's R&D and provide the company with both opportunities to license emerging technologies and freedom of action with respect to other companies' patents. Just as important, explains Manny Schecter, Research's intellectual property counsel, many IBM patents involve technology about to be incorporated into products.

IBM's 1996 patents cover a broad swath of the information technology industry, including software, networking, computer systems, microprocessors, memory chips and storage technology. Roughly 400 of the patents resulted in some way from work by the Research Division. As in previous years, the division produced about 20 percent of the patents directly and played a role in another 5 percent of the total.

Highlights from Research in last year's patents include:

  • A wireless local area network invention that enables a mobile ThinkPad to establish and maintain a radio connection even in interference-laden environments such as crowded offices (U.S. 5,533,025).
  • High-quality, digitized visual images encoded with a watermark to protect ownership (U.S. 5,530,759).
  • < A frame-sampling technique to provide users with either browsing or fast-forward capability while screening a movie downloaded from a server through video-on-demand (U.S. 5,521,630).

More than 350 of IBM's 1996 patents are related to networking. That record number illustrates the company's determination to move its R&D toward network-computing-related technologies. "IBM's accelerating rate of patent innovation is a direct response to the challenge of the marketplace - to develop user-oriented products for a networked world," explains Marshall Phelps Jr., vice president of intellectual property and licensing for IBM.

Research is following that cue. According to Schecter, the division has recently stepped up its efforts to earn licensing fees from its patents, both by itself and in a support role for other IBM divisions.

New Chip Exceeds Speed Barrier

Scientists at the Thomas J. Watson Research Center, in collaboration with designers at IBM's System/390® Division, have created one of the world's fastest complex instruction set computing (CISC®) chips. The new chip will serve as the microprocessor for a new generation of IBM S/390 high-end servers.

The chip's clock speed of 400 megahertz exceeds that of all but one other previous CISC chip. Just as significant is its complementary metal-oxide-semiconductor (CMOS) technology, which replaces the bipolar technology used in previous S/390 microprocessors. According to Carl Anderson, who led the Watson team that worked on the design, the choice of CMOS reflects IBM's commitment to higher performance at lower operating costs for its high-end servers.

The 17-millimeter-square chip, which contains 7.8 million transistors, offers higher efficiency than bipolar chips because it is smaller and cooled by air rather than water. As a result, just 32 of the CMOS chips in a single grouping known as a thermal conductivity module (TCM) can replace 50 TCMs containing 100 bipolar chips apiece. "Instead of filling up a big room, machines with this technology are now the size of a refrigerator," says Anderson.

The design features two major innovations. First, the microprocessor is optimized for the sets of instructions that users encounter most often. And it has a unique, extremely fast error-checking capability. Second, two processing units execute every instruction, while a third checks the results to make sure that the concurrently executed instructions match. The effect, according to Anderson: no more than a single unplanned outage every 16 years.

Showcasing Solutions for Industry

As part of the effort to move emerging technologies rapidly into the marketplace, IBM has set up three Industry Solutions Labs. The Labs, intended to introduce existing and potential customers to IBM's technologically advanced solutions, are located at the Hawthorne site at the Thomas J. Watson Research Center, and at IBM laboratories in Stuttgart, Germany, and Yamato, Japan. The three Industry Solutions Labs, which opened in February, are run by the company's Industry Solutions Units (ISUs) "against the backdrop of Research," explains Roger Pollak, a Watson scientist who is helping to coordinate activities between Watson and its Industry Solutions Lab. The first-of-a-kind solution demonstrations assembled at the Industry Solutions Labs differentiate the Labs from IBM's traditional customer briefing centers.

The new Labs will showcase Research technologies in the context of industries' future needs. To that end, they feature displays of the latest products that have resulted from Research's efforts; prototypes from the company's first-of-a-kind research projects developed for specific customers that are replicable for others; and illustrations of more basic research. Facilities include state-of-the-art presentation facilities, lounges, dining rooms, kitchen, offices and other executive comforts. "The Labs are facilities where a customer can come together with Research and other IBM experts to discuss their business problems and collaborate on solutions," explains Maureen Sorbo, the Labs' global industries solutions investment executive.

Corporate delegations of five, ten or more individuals, invited by ISUs, will generally visit for a full day, to share views on where their industries are heading. IBM's basic aim is to engage visitors interactively in discussions about the future of their industries and to point out new areas of its technology that may be customized as solutions for that future. The Labs will focus particularly on network computing solutions that can help companies transform their businesses.

The Research Division also benefits from the Industry Solutions Labs. "We gain a better understanding of customers' needs," says Pollak. "Also, through interaction with industry decision-makers, Research can better anticipate the emerging requirements of the marketplace."

Adventures in Software

Late in 1995, as part of a year-end review, executives at IBM's Thomas J. Watson Research Center concluded that Research needed a new approach in the systems and software area. "We were too focused and risk-averse," recalls Stephen Lavenberg, a member of the systems and software technical staff at Watson. To respond to those deficiencies, a program called Adventurous Systems and Software Research (ASSR) was set up late last year by Ambuj Goyal, VP Systems and Software at Watson. Designed to break new technical ground and to spark sea changes in the systems and software industry, ASSR will foster long-term, high-risk, high-reward research. Under Lavenberg's leadership, a team from Almaden, Watson and Zurich evaluated over three dozen proposals.

Watson's first four ASSR projects have just been announced:

  • New approaches to dealing with the information economy - the anticipated crowding in cyberspace that could develop when billions of agents operate throughout networks.
  • Highly scalable multiprocessor operating systems for the computers that will soon become available to the public.
  • Software, called pellets, that is launched into the network to dynamically access services distributed across multiple Web sites.
  • Speech verification systems (see page 18) that will identify speakers' voices without training.

Meanwhile, the Almaden Research Center and the Zurich Research Laboratory have recently announced their own ASSR programs. Both centers have put out requests for proposals.

Lavenberg emphasizes that ASSR represents just one approach to adventurous activity within Research. What the new program offers, he says, is the opportunity to work on software projects "in the 'white space' outside our regular strategic framework."

This new, occasional feature will highlight fresh programs and projects on Research's agenda.


Donald D. Chamberlin, research staff member at IBM's Almaden Research Center, is one of 85 new members elected to the National Academy of Engineering. In electing Chamberlin, the Academy recognized his contributions to the SQL(TM) database query language.

Dieter Pohl of the Zurich Research Laboratory has been named co-recipient of the 1997 Rank Prize, for contributions to the science and applications of near-field optics (NFO). Pohl and his colleagues conceived of, built and demonstrated a completely new type of optical microscope, on which IBM holds the basic patent.

The American Ceramic Society has elected Zurich IBM Fellow J. Georg Bednorz an Honorary Member. The election recognizes the discovery of high-temperature superconductivity in a ceramic oxide, for which Bednorz shared the 1987 Nobel prize in physics with Zurich's Alex Müller.

The ACM has elected Watson's Philip S. Yu as a Fellow. The election recognizes Yu's contributions to the theory and practice of design and analysis of database systems, parallel architectures and algorithms.

The Institute of Electrical and Electronics Engineers (IEEE) will present its 1997 Cledo Brunetti Award to Watson researcher George A. Sai-Halasz, for developing sub-0.1 micron MOSFET devices and circuits. Sai-Halasz and his colleagues redefined the limits of field-effect transistor (FET) technology by designing and fabricating devices and circuits down to 0.05 micron channel lengths. This work demonstrated that 0.1 micron and below was feasible, and established silicon FETs as the highperformance technology of the future.

The IEEE has also elected several employees of IBM Research as Fellows. They are: Ram Chillarege of Watson, for contributions to the theory and practice of the design of reliable software; Philip G. Emma of Watson, for innovation in high-performance computer architecture; Ronald Fagin of Almaden, for contributions to finite-model theory and to relational database theory; Roger F. Hoyt, an Almaden alumnus now in IBM's Storage Systems Division, for contributions to magnetic rigid disk storage and interface reliability; Klaas B. Klaassen of Almaden, for contributions to advanced measurement and analog circuit designs for magnetic recording; Eric Kronstadt of Watson, for contributions to processor architectures, compilers and operating systems; Jorma J. Rissanen of Almaden, for developing the principle of minimum description length and stochastic complexity for model selection and data compression; Lubomyr T. Romankiw of Watson, for inventing the magnetic thin-film inductive head and the magnetoresistive induc tive merged head, and for major contributions to the science and technology of electrochemistry; Jane M. Shaw of Watson, for inventing the silylation process and other contributions to microlithographic resist technology.

The American Physical Society has elected the following Watson researchers as Fellows: Philip E. Bateson, for contributions to electron spectroscopy studies of matter; Massimo V. Fischetti, for predictive modeling of submicron semiconductor devices; Jeffrey A. Kash, for application of optical techniques to understanding excitations in semiconductors; Fenton R. McFeely, for applying photoemission techniques to processes that underlie microelectronics technology; and James A. Misewich, for applying innovative laser techniques to fundamental problems in molecular dynamics and interactions.


Video Interactivity

HotVideo is a new hyper-linked video technology invented by the China Research Laboratory in cooperation with Watson. It provides rich interactivity to conventional video from various sources: local files, server files, CD-ROM, and Internet. HotVideo has many potential applications in education, digital library, E-commerce, games and advertising. The technology consists of HvMaker, a user-friendly authoring tool for content creation, and HvPlayer, a Windows 95® application and plug-in for Netscape® Navigator and Microsoft Internet Explorer®. It will soon be made available on IBM's alphaWorks site.

Driving Up Disk Density

Researchers at the Almaden Research Center have demonstrated for the first time the ability of product-level components to write and read data stored on a computer hard disk at a density of more than 5 gigabits (5 billion bits) per square inch. Such a density is almost three times that of IBM's Travelstar(TM) VP, the most advanced disk drive now available.

Polymers in Line

Scientists at Almaden and the University of Massachusetts have devised a simple technique for precisely controlling the interfacial energies and wetting behavior of polymers in contact with solid surfaces. This approach, reported in Science, makes it possible to align specific parts of a polymer and could find use in the synthesis of man-made membranes with definable properties.

Seeing the Light

Physical effects can have obvious implications for technology that are easily ignored unless the scientists who know the effects and the engineers who know the technology can meet on common ground. Such was the case with an observation by Jeff Kash and Jim Tsang, scientists at the Thomas J. Watson Research Center. Their recognition, in November 1995, of a previously neglected aspect of a well-characterized physical phenomenon and subsequent collaboration with colleagues in the IBM Microelectronics Division (IMD) have led to a diagnostic tool with immense significance for IBM and, indirectly, the global semiconductor industry.

Scientists have long known that the field effect transistors (FETs) that are the building blocks of the complementary metal-oxide-semiconductor (CMOS) circuits widely used in integrated circuit technology emit tiny amounts of light when a current is running through them. It was also known that currents pass through normally-operating CMOS circuits only when they switch from one state to another ÷ such as "on" to "off." Kash and Tsang put those two facts together and realized that they could use the light emission to monitor the switching of the individual components of CMOS chips. "It was so simple we were surprised for a long time that nobody else seemed to have realized it," says Kash.

To confirm the effect, the pair used basic photodetectors to monitor the light emission from simple high-speed circuits fabricated at the IMD facility in Burlington, Vermont. Then, recalls Tsang, "we realized we had the equipment - an exotic photomultiplier - to look at every device on a chip simultaneously." They soon proved that ability by monitoring a ring oscillator circuit, in which successive components switch on and off in order. Collaborating with colleagues in IMD, they proved that the technique works with more complex systems such as IBM microprocessors.

Since then the researchers have worked on ways to make the technique practical. Its chief application is in spotting and diagnosing faults in chips at the design and prototyping levels. Late last year, for example, Kash and Tsang used the method to identify a problem in the Alliance chip then under development. "Using this technique," says Kash, "you can figure out how to make your chips run faster." The technique can also be applied to the failure analysis of chips emerging from full-scale fabrications plants.

Automatic Circuit Training

In high-performance custom-designed chips, the sizes of individual transistors are tuned to maximize the performance of the circuit. The key parameters are transistor widths, which control the amount of current that flows through each transistor. In simple terms, wider transistors generally lead to faster circuits but consume more power and area, and increase the load on previous stages of the circuit. Designers tune their circuits by adjusting the widths of transistors to obtain optimum performance.

The traditional method, says Chandu Visweswariah, a researcher at IBM's Thomas J. Watson Research Center, is "a slow, tedious, manual and error-prone process." Now Visweswariah and his Watson colleagues Andrew Conn and Ruud Haring have devised a computer-aided design tool, called JiffyTune, that automates circuit tuning while permitting its use on far larger circuits than previous techniques allowed.

The approach stemmed from the search for a method to break the computational bottleneck of dynamic tuning. Dynamic tuning relies on circuit simulation programs, most notably IBM's proprietary simulator AS/X and Simulation Program with Integrated Circuit Emphasis (SPICE), the industry standard for 25 years. While the numerical algorithms in SPICE-like simulators are accurate to within 1 percent, they are slow. Automatic tuning with SPICE in the inner loop is limited to 10 tunable transistors.

Four years ago, Visweswariah wrote a tool called Simulation Program for Electronic Circuits and Systems (SPECS). Although somewhat less accurate than SPICE, it runs about 70 times as fast. Nevertheless, the fast simulation by itself would render feasible the tuning of only about 30 transistors simultaneously. Extending the process beyond 30 tunable transistors demanded more sophisticated techniques.

Enter the team's JiffyTune engine, a gradient-based optimizer that exploits the unique capability of SPECS to compute circuit gradients efficiently. JiffyTune iteratively feeds transistor sizes to SPECS to obtain values of circuit performances such as delay and power, and their gradient with respect to transistor sizes. The values are supplied to LANCELOT, a general purpose nonlinear optimization package coauthored by Conn, which determines each fresh step in the tuning process as a set of transistor size changes. The iterative process stops when the circuit has reached optimum performance within predetermined constraints, such as area, power, delay or minimum transistor widths.

Key to JiffyTune's acceptance is its designer-friendly interface. Built by Haring, it comes with a point-and-click facility, prompts, and graphical back-annotation of transistor widths onto the circuit schematic. All the tuning criteria are stored as attributes of the circuit schematic, thus facilitating design reuse at the push of a button. "With JiffyTune," says Visweswariah, "you never need to redo work you've already done."

So far, JiffyTune has tuned circuits with about 1,000 tunable transistors. The Watson team is now extending the technology. It has recently filed a patent application on a method that significantly reduces the time required to compute circuit gradients, thus enabling JiffyTune to optimize larger circuits. "It's a state-of-the-art way of carrying out optimization," says Conn.

What is JiffyTune's value? "In addition to making IBM circuit designs better," explains Visweswariah, "it improves our designers' productivity."  |  IBM Research  |  About Think Research  |  Terms of use  |  Privacy