engengeniousenious a publication for alumni and friends of the division of engineering and applied science of the California Institute of Technology

winter 2003 The Division of Engineering and Applied Science consists of thirteen Options working in five broad areas: Mechanics and Aerospace, Information and Communications, Materials and Devices, Environment and Civil, and Biology and Medicine. For more about E&AS visit http://www.eas.caltech.edu.

DIVISION CHAIR OPTIONS

Richard M. Murray (BS ’85) aeronautics (GALCIT) www.galcit.caltech.edu

DIVISION STEERING COMMITTEE applied & computational mathematics www.acm.caltech.edu Jehoshua (Shuki) Bruck Gordon and Betty Moore Professor of applied mechanics Computation and Neural Systems and www.eas.caltech.edu/options/appmech.html Electrical Engineering applied physics Janet G. Hering www.aph.caltech.edu Professor of Environmental Science and Engineering bioengineering www.be.caltech.edu Hans G. Hornung C. L. “Kelly” Johnson Professor of Aeronautics civil engineering www.eas.caltech.edu/options/civileng.html Thomas Yizhao Hou Professor of Applied and Computational computation & neural systems Mathematics www.cns.caltech.edu

Melany L. Hunt computer science Professor of Mechanical Engineering www.cs.caltech.edu

Peter Schröder control & dynamical systems Professor of Computer Science and Applied www.cds.caltech.edu and Computational Mathematics electrical engineering Kerry J. Vahala (BS ’80, MS ’81, PhD ’85) www.ee.caltech.edu Ted and Ginger Jenkins Professor of Information Science and Technology and environmental science and engineering Professor of Applied Physics www.ese.caltech.edu

P.P.Vaidyanathan materials science Professor of Electrical Engineering www.matsci.caltech.edu

mechanical engineering ENGenious www.me.caltech.edu

editors Marionne L. Epalle CALTECH ALUMNI ASSOCIATION Peter Mendenhall The mission of the Caltech Alumni Association is to art direction/designer promote the interests of Caltech as a world standard Jessica Stephens of academic excellence by strengthening the ties of goodwill and communication between the Institute, its alumni, and current students, and by maintaining PHOTO CONTRIBUTIONS programs to serve their needs. For more information on the Association and its activities, visit its website. Linda McManus http://www.its.caltech.edu/~alumni Peter Mendenhall Bob Paz Briana Ticehurst SUPPORT CALTECH

http://giving.caltech.edu STAFF COLLABORATORS

Gillian Andrews COMMENTS? Department Assistant [email protected] Briana Ticehurst Multimedia Specialist

© 2003 California Institute of Technology. All Rights Reserved. engeng eniousenious winter 2003

3 note from the chair

4 snap shots ‘Round About the Institute

6 new faculty

Influx of Talent: Division Grows by Nine Marc Bockrath Hideo Mabuchi Michael Dickinson Michael Roukes Charles Elachi Tapio Schneider Alexei Kitaev Chris Umans Nadia Lapusta 8 progress reports 26 Understanding Material Deformation: Insights into the Inner Workings of Complex Materials by Ersan Üstündag 36 idea flow 12 Distributed Integrated Circuits: Wideband Communications for the 21st Century Sensing and Responding: Mani Chandy’s Biologically by Ali Hajimiri Inspired Approach to Crisis Management

16 emeritus 38 research note Kerry Vahala and His Miniature William Bridges: A Rare Combination of Talents Orbits of Light: Lasing Spheres by Amnon Yariv

40 alumni profiles 18 information science and technology Ivett Leyva: An Experimentalist with International Flair When You Come To A Multidimensional Fork In The Road, Take It: Aeronautics, PhD ’99 Conversations with Members of the IST Faculty Planning Committee 41 Eric Garen: Education at the Fore Electrical Engineering, BS ’68 19 Center for Biological Circuit Design 23 Center for the Physics of Information 28 Social and Information Sciences Laboratory 31 Center for the Mathematics of Information 26 Computing Beyond Silicon Summer School: Or How I Spent My Summer in Pasadena

Cover Image: The montage on the cover symbolizes the melding of biology and circuits—one essential theme of ISTI: the Information Science and Technology Institute. ISTI is dedicated to systematically exploring information science and technology on sev- eral fronts, including the theoretical underpinnings, biological circuit design, the physics of information, and societal structures such as financial markets and social organizations. Starting on page 18 are four conversations with the Caltech faculty charged with laying the groundwork for this new enterprise. Cover components: The organic shape is C. elegans, a nematode from Professor Paul Sternberg’s lab. Professor Jehoshua Bruck, in collaboration with Sternberg, studies the nematode as a proving ground for computational models. The circuit, a low-noise device from Professor Ali Hajimiri’s lab, is a precursor to Hajimiri’s current work in distributed integrated circuits (see page 12).

note from the chair

This is our third issue of engenious and I am coming up on my third anniversary as Division Chair for the Division of Engineering and Applied Science. One of the accomplishments that I’m very proud of is the establishment of a strategic plan for the Division, developed with the help of representatives from each of the options and major centers in the Division. This plan is guiding our educational programs, our faculty hiring, and our fundraising activities, and is being further developed by faculty planning committees in each of the major thrust areas. If you’d like to have a look, it’s available online at http://www.eas.caltech.edu/strategic_plan; we would welcome your comments and feedback. Our feature piece reports on the formation of Caltech’s Information Science and Technology Institute (ISTI), one of the major elements of our strategic plan. The article fills almost one-third of this issue, indicating how important and far-reaching the research and teaching efforts of this new venture will ultimately be. Campus-wide, one-fifth of the faculty and one-third of all stu- dents will be involved in this multi-layered, multidisciplinary exploration of the analysis and design of complex natural and synthetic information systems. With its center of gravity in E&AS, ISTI will give us tremendous reach in bringing the best and brightest to Caltech, creating an approach that is likely to be emulated elsewhere, and changing the way the world thinks about and harnesses information. ISTI is a rare opportunity to pull the future a little closer. As we look around the rest of this issue, you will notice an addition to one of the regular fea- tures. Our Alumni section profiles two alums, a younger one (graduated within the last 5 years) and a more established one. We hope you’ll enjoy (and reminisce) reading about the post-Tech adventures of Ivett Leyva (PhD ’99) and Eric Garen (BS ’68).

Sincerely,

richard m. murray Chair, Division of Engineering and Applied Science

Image at left: Domain patterns in the ferroelectric material PbTiO3 obtained by using polarized light microscopy. Such microscopic patterns form spontaneously in ferroelectric materials giving rise to unique electro-mechanical properties which are useful for applications in micromachines. Nine faculty members from E&AS are investigating how such patterns can be engineered to produce desired properties well beyond those currently avail- able, and at the limits of what is theoretically possible (see http://www.femuri.caltech.edu). This photograph was taken by graduate student Rongjing Zhang in the laboratory of Professor G. Ravichandran. The area imaged is about 2.5 mm x 1.2 mm. 2 3 snap shots

’Round About the Institute

Go, Erik, Go: JPL’s Chief Technologist, Erik Antonsson

Erik Antonsson, professor of and former executive officer for mechanical engineering, has a new calling: chief technologist for the Jet Propulsion Laboratory ( JPL). A national search led by Richard Murray, professor of mechanical engineering and chair of the Division of Engineering and Applied Erik Antonsson Science, followed a long and winding road to Erik’s door. They knocked, and he answered. Charles Elachi, director of JPL, said, “Dr. Murray and his committee interviewed a number of nationally rec- ognized technology leaders and determined that Dr. Antonsson’s expertise and experience are an out- standing match for the position.” Erik began his two-year leave of absence from Caltech last September (though he still comes to campus one day a week to continue his research). Probably best known to the public as the creator and driving force behind the course ME 72 and its wildly seri- ous engineering-design contest, Erik will no doubt have new seriously wild adventures at JPL that he may incorporate into future ME courses.

Find out more about Professor Antonsson at http://www.design.caltech.edu/erik/antonsson_bio.html

Two VPs and a Dean: Margo Marshak, Gary Dicovitsky, and Erica O’Neal Join Caltech

Three new administrative appointments are bringing a wealth of talent to Caltech’s door. Dr. Margo Marshak is Caltech’s new vice president for student affairs. “I’m obvi- ously thrilled to have the opportunity to come to such a great institution,” says Marshak. “I’m enormously impressed From left to right: Margo Marshak, Gary Dicovitsky, and Erica O’Neal. with the quality of the students, faculty, and administra- tion.” Fresh from her role as vice president and dean of students at the University of Chicago, Marshak will be the sen- ior Caltech executive responsible for envisioning, leading, advocating for, and managing student welfare and interests. Gary Dicovitsky is the new vice president for development and alumni relations. Most recently, he served as vice president for development at Pomona College. “Caltech’s superior reputation as a research and teaching institution with such depth and capacity in interdisciplinary projects and programs was fundamental to my interest,” remarked Dicovitsky. “I am honored to be offered the privilege to help build on the institution’s past successes and to interact with such extraordinary faculty, staff, and alumni.” Dr. Erica O’Neal is Caltech’s new associate dean and director of the office for multicultural education and student affairs. She arrived at Caltech from Stanford, where she served as associate director of development in the School of Humanities and Sciences and previously as an assistant dean in the School of Engineering. “I am enthusiastic about joining the Caltech family and making new contributions that serve to increase diversity and build community among the student body,” says O’Neal. A cum laude graduate of Harvard, O’Neal holds an MS and a PhD in higher education from the University of Pennsylvania.

engenious winter 2003 Simon Wilkie Goes to Washington: Simon Wilkie New Chief Economist for the FCC

Simon Wilkie, a senior research associate in economics at Caltech, has assumed the position of chief economist for the Federal Communications Commission (FCC), the agency responsible for regulating the likes of, well, just about everything in our media-mediated world. The FCC, established in 1934, is responsible for regulating interstate and international communications by radio, television, wire, satellite, and cable. As chief economist, it is Wilkie’s responsi- bility to provide independent, nonpartisan advice—from an economic perspective—to the commissioners on various regulatory issues. “We sort through what is oftentimes conflicting advice given to the FCC, then provide guidance to the commis- sioners and the chairman.” He notes that Congress, for example, will mandate that the FCC should do certain things, like develop regula- tions for telephone network access by new market entrants, or develop a fair system for auctioning off high-speed bandwidth. But, they don’t spell out the specifics of how to do it. “Our job is to come up with the right formula that works, one that is fair to all concerned, and in the best interests of the public.”

Stainless Steel, Travertine, and the Biological Sciences: The Broad Center

Caltech has a new building, and it’s a looker. (See the inside back cover for a glimpse.) The Broad Center for the Biological Sciences was dedicated in September 2002, and will use its 120,000 square feet to house laboratories and offices for 13 research teams focusing on magnetic imaging, computational molecular biology, and investigation of the biological nature of consciousness, emotion, and perception. Principal funding for the structure came from Caltech trustee Eli Broad and his wife, Edythe. Speakers at the opening included David Baltimore, president of Caltech; Benjamin Rosen, chairman of the Caltech Board of Trustees; Allen Rudolph, principal with construction company Rudolph and Sletten; Elliot Meyerowitz, chair of the Division of Biology and professor of biology; and Eli Broad. “The Broad Center adds a dis- tinguished architectural achievement to Caltech’s already beautiful campus,” said Baltimore. “It is a testament to the generosity of many friends of Caltech, led by Eli and Edye Broad, and to the genius of James Freed, its design archi- tect. Most importantly, it’s a highly functional building, providing a framework for advances in the biological sciences in the 21st century.”

In Hot, Bubbly Water: New Comfort Zone Added to Braun Athletic Center There’s always room for more at the Sturtevant Spa. Bradford Sturtevant’s family and friends have created for the Caltech com- munity a place of respite after long days of using brains and brawn. The Sturtevant Memorial Spa was dedicated on May 2, 2002, commemorating the late Bradford Sturtevant, Liepmann Professor of Aeronautics. A seemingly tireless swimmer and promoter of swimming, Sturtevant was a strong advocate for athletics at Caltech, and served for many years on the faculty athletics committee. He played a key role in the planning and construction of the Braun Athletic Center, as well as the planning and construction of the Sherman Fairchild Library of Engineering and Applied Science.

Learn more about our athletics facilities at http://www.athletics.caltech.edu

4 5 new faculty

Influx of Talent: Division Grows by Nine

Six new professors have joined the Division and Caltech over the past several months, bringing fresh insights and new research directions our way.

Also new to the Division are three faculty members already part of the Caltech community: (pictured from left to right) Charles Elachi, Director of the Jet Propulsion Laboratory, joins Electrical Engineering; Hideo Mabuchi is now a member of both the Physics faculty and the Control and Dynamical Systems Option; and Michael Roukes, also a member of the Physics faculty, joins both the Applied Physics and Bioengineering Options.

Marc Bockrath: Assistant Professor of Applied Physics

Professor Bockrath’s interests are in nanofabrication, and the electronics and mechanics of systems that have critical dimensions on the nanometer scale, which represents the ultimate limit to miniaturization. These systems include materials such as carbon nanotubes and individual molecules. Currently, he is inter- ested both in investigating the new and interesting transport phenomena that arise in nanostructured materials, and in investigating the properties of nanostructures that have mechanical degrees of freedom. Potential applications include nanoscale switches, logic gates, and sensors. Bockrath received a BS degree in Physics from the Massachusetts Institute of Technology in 1993, and a PhD in Physics from UC Berkeley in 1999. Most recently he was a postdoctoral fellow at Harvard University.

Michael Dickinson: Professor of Bioengineering

Professor Dickinson’s primary research interests concern the physiology and mechanics of flight behav- ior in insects. Specifically, he has focused on the flight-control system of flies—arguably the most aerody- namically sophisticated of all flying animals. His research strategy is to tackle flight behavior using approaches from such disparate disciplines as neurobiology, structural engineering, and aerodynamics. Thus, Professor Dickinson’s lab attempts to study flight-control behavior at several levels of analysis simul- taneously, from the physiological properties of individual neurons and circuits to the skeletal mechanics of wing motion and the production of aerodynamic forces. This multi-level approach is challenging and yet rewarding, as novel insight is often gained by addressing a problem simultaneously from several perspec- tives. Dickinson received his ScB degree from Brown University in 1984 and a PhD in Zoology from the University of Washington in 1989. He comes to Caltech from UC Berkeley, where he was the Williams Professor of Integrative Biology.

Alexei Kitaev: Professor of Theoretical Physics and Computer Science

Professor Kitaev’s research area is quantum computation, which includes quantum algorithms, error correction, and quantum complexity classes. Professor Kitaev has devised a phase-estimation algorithm and topological quantum codes, as well as an efficient classical algorithm for the approximation of unitary operators by products of generators. He has also studied complexity classes BQNP and QIP. His other important idea is error correction at the physical

engenious winter 2003 level, in particular fault-tolerant quantum computation by anyons. He is currently working on physical models that would make this scheme feasible. Kitaev received an MS degree from the Moscow Institute of Physics and Technology in 1986, and a PhD from the L.D. Landau Institute for Theoretical Physics in 1989. He worked at the L.D. Landau Institute until 1998, then spent a year at Caltech as a visiting researcher and lecturer, two years at Microsoft as a researcher, and came back to Caltech as a senior research associate in fall 2001. Professor Kitaev has a joint appointment in the Division of Engineering and Applied Science and the Division of Physics, Mathematics, and Astronomy.

Nadia Lapusta: Assistant Professor of Mechanical Engineering

Professor Lapusta’s research interests are in continuum mechanics, computational modeling, fracture and frictional processes, and the mechanics and physics of earthquakes. Her work is directed towards understanding fracture and frictional phenomena on all scales, from frictional failure in earthquakes and dynamic cracks in solid structural components to tribological processes on micron-sized asperities and complex atomic and molecular interactions at crack tips. A significant effort is devoted to developing efficient computational techniques applicable to such nonlinear, dynamic, and multiscale problems. Her current studies include nucleation and dynamics of frictional instabilities, models of earthquake sequences, dynamic fracture on bimaterial interfaces, and shear heating effects during rapid slips. Lapusta received her Diploma in Mechanics and Applied Mathematics from Kiev State University (Ukraine) in 1994, and both her SM (1996) and PhD (2001) degrees in Engineering Sciences from Harvard University.

Tapio Schneider: Assistant Professor of Environmental Science and Engineering

Professor Schneider’s research interests are in the dynamics of the global circulation of the atmos- phere and in large-scale atmospheric turbulence and turbulent transport. His current research focuses on developing theories concerning the turbulent fluxes of heat, mass, and water vapor that contribute to maintaining such basic climatic features as the pole-to-equator surface-temperature gradient, the ther- mal stratification of the atmosphere, and the distribution of atmospheric water vapor. Schneider received his PhD from Princeton University in 2001, and did his undergraduate work at Freiburg University (Vordiplom, 1993). Professor Schneider has a joint appointment in the Division of Engineering and Applied Science and the Division of Geological and Planetary Sciences.

Chris Umans: Assistant Professor of Computer Science

Professor Uman’s research area is theoretical computer science, in particular complexity theory. He has studied the computational complexity of fundamental optimization problems from application areas such as circuit design and learning theory. His recent work centers on basic questions regarding the power of randomness in computation. Other research interests include explicit combinatorial construc- tions, hardness of approximation, coding theory, and algorithms for problems from graph theory and algebra. Umans received a BA degree in Computer Science and Mathematics from Williams College in 1996, and a PhD in Computer Science from UC Berkeley in 2000. Before joining Caltech, he was a postdoctoral scholar in the Theory Group at Microsoft Research.

Center photos, clockwise from top left: Marc Bockrath, Nadia Lapusta, Tapio Schneider, Chris Umans, Alexei Kitaev, and Michael Dickinson.

6 7 progress reports

Understanding Material Deformation: Insights into the Inner Workings of Complex Materials by Ersan Üstündag

n most engineering calcu- tion, as they provide in-situ I lations, the mechanical information about internal performance of structures strains (and indirectly, stresses), or components is estimat- crystallography (to help identify ed under the assumption that the different phases), and texture material is homogeneous or can be (or preferred grain orientation). represented by a continuum. Diffraction techniques use a Although this assumption is often material’s crystalline lattice as sufficient, it prevents a true under- an “internal gauge,” and are standing of deformation mecha- therefore sensitive to changes nisms, as most structural materials occurring on the atomic scale. are actually composites (comprised In addition, when a large sam- of multiple phases) and/or poly- pling volume is chosen, contri- crystals (composed of many grains). butions from many regions are It turns out that the interactions included in the overall “signa- between phases and grains largely ture” of the material, leading to determine the overall behavior of an effective averaging or bulk the material. These interactions characterization. X-ray and neu- occur over multiple length scales, tron diffraction can be used from nanometers to centimeters. independently or in a comple- Ersan Üstündag Any experimental technique that mentary manner, as the former intends to fully characterize materi- can probe sub-micrometer al deformation must be sensitive to regions while the latter is more models will be valuable to engi- such a scale range. The technique suitable for in-situ bulk studies on neers designing and constructing must also be non-intrusive, as it the scale of millimeters to centime- complicated structures or devices as should not cause damage while ters. varied as jet turbine engines, cars, interrogating the material. Another In our research, we employ buildings, satellites, and electronic important requirement is that the both x-ray and neutron diffraction chips. technique should allow in-situ for a complete, multiscale charac- This report details one impor- studies, that is, monitoring of terization of material deformation. tant aspect of our research, namely material deformation under a vari- Our aim is to develop accurate con- the use of neutron diffraction in ety of conditions, such as applied stitutive laws describing the behav- deformation studies. It also load, temperature, or atmosphere. ior of a composite or a polycrystal. describes our recent efforts to Diffraction is a powerful tech- Accurate description of constitutive design and construct a dedicated nique for material characterization, behavior is crucial for successful engineering neutron spectrometer and easily satisfies these require- modeling of material behavior, called SMARTS. Much more than ments. Especially attractive meth- including prediction of expected a catchy acronym (standing for ods are x-ray and neutron diffrac- lifetime. We anticipate that our Spectrometer for MAterials

engenious winter 2003 Research at Temperature and under vacuum or in a controlled material and scatter in all direc- Stress), SMARTS is currently atmosphere is now routine. This tions. Some of these reach one of unique in the world. It is the first represents a significant increase the two detector banks centered on instrument specifically designed for over previous standards. Some of the horizontal plane at 90° to the engineering stress/strain studies at the exciting capabilities provided by incident beam. Each detector con- a spallation neutron source. Lo- SMARTS include measurements of sists of three panels with a total of cated at the Los Alamos Neutron spatially resolved strain fields; 192 3He gas-filled aluminum tubes. Science Center (LANSCE) in New phase deformation, and load trans- Interactions between the neutrons Mexico, it was commissioned in fer in composites; the evolution of and 3He in the detector tubes pro- 2001. It is funded by the Depart- stress during high-temperature fab- duce 4He plus gamma radiation ment of Energy (Office of Basic rication; and the development of and ionize the gas, creating a cas- Energy Sciences), and was built by strain during reactions or phase cade of electrons with associated a team led by the author. transformations. charges. These charges are digitized SMARTS is expanding the use The layout of SMARTS is and converted electronically to pat- of neutron diffraction to a wider shown in Figure 1. At LANSCE, terns of intensity versus scattering range of engineering problems than neutrons are produced by spalla- angle. Data from the tubes are was previously possible. With its tion, which involves accelerating combined to provide time-of-flight extensive array of in-situ capabili- protons to very high energies neutron-diffraction patterns. Anal- ties for sample environments, it toward a tungsten target, then col- ysis of the diffraction patterns is enables measurements on small (1 lecting the polychromatic neutrons carried out with a least-squares fit- mm3) or large (1 m3) samples. Ease that form. These neutrons pass ting routine called the Rietveld of access to the sample bay is one through a water moderator, which method. Data acquisition is based significant new feature. Com- reduces their energies to a range on virtual memory extension tech- ponents with dimensions up to 1 suitable for diffraction. After pass- nology and uses web-based visuali- meter and mass up to 1,500 kilo- ing through the T0 chopper, a zation and control software. grams can be positioned precisely device which further removes in the path of the neutron beam. fast neutrons and the gamma Permanently mounted alignment flash (to minimize back- xperiments can be con- theodolites provide a simple and ground), the thermal neutrons E trolled remotely from the efficient way to position samples or reach the guide. The guide is user’s laboratory (any- equipment to within 0.01 mm. coated with 58Ni, and, via the where in the world), and Achieving this level of precision is process of near-total reflection, real-time data analysis can be critical for stress-strain measure- keeps most of the neutrons in the accomplished with a unique soft- ments; misalignments of more than beam path. The guide terminates at ware package called Expert System. 0.1 mm can result in significant the inner surface of the cave wall. This software represents a radical pseudo-strain artifacts. Two aperture sets (located between new approach to experiment plan- A furnace and load-frame suite the exit of the guide and the sam- ning and execution in the neutron- allows research on materials under ple) permit the beam cross-section diffraction field. For the first time, extreme loads (60,000 pounds or to be defined continuously in shape the experimenter has a chance to 250 kN) and at extreme tempera- and area between 1 and 100 mm2. optimize an experiment according tures (1,500°C or 2,700°F). In-situ When the neutron beam pene- to his/her needs and predict results uniaxial loading on samples 1 cm in trates a sample, some of the neu- even before starting. Moreover, diameter at stresses over 3 GPa trons interact with atoms in the during the experiment, data are

8 9 Figure 1. Neutrons from the moderator pass through a series of collimating

apertures before entering the neutron guide. A T0 chopper removes fast neu- trons and gamma flash that would otherwise contribute unwanted back- ground. Slow thermal neutrons continue down the guide to the entrance of the SMARTS cave (about 5 x 6 m in size). On exiting the guide, neutrons pass to the center of the cave where some are scattered by the sample to the detectors. Samples or ancillary systems are placed directly on the translator, which can accommodate up to 1,500 kg, move in three orthogonal directions, and rotate about a vertical axis. Theodolites provide precise optical triangula- tion and alignment capability for equipment or samples. Here, the load- frame-furnace suite is shown on top of the translator. In some experiments where a three-dimensional sampling volume is desired, radial collimators are inserted between the detectors and the sample. When used with the incident collimation, selection of an appropriate radial collimator defines a sampling volume for spatially resolved measurements.

analyzed in real time, allowing a mum set of data points required to tion. When this occurs, robust data quick assessment of the results. obtain a desired outcome in the comparison between these facilities Figure 2 describes the interactions shortest possible time. The latter will be achieved for the first time. between the user and the various issue is important since beam time This is also expected to lead to components of Expert System. is very expensive. For this reason, standardization of engineering First, the user is asked to input the current version of Expert detailed material data and the System includes real-time data strain error desired. The software analysis to determine then simulates the expected diffrac- the exact time when tion pattern. This calculation incor- enough data have been porates realistic models of the collected to satisfy the instrument optics so that the simu- user’s specified initial lated pattern is truly representative error value. of the sample. In addition, based on Expert System was the experimental parameters (e.g., mostly programmed by tension vs. compression) and speci- a group of undergradu- men characteristics (e.g., monolith ate students led by vs. composite), Expert System will Richard Karnesky (BS ’02, past soon be able to perform several president of Ricketts House) mechanics calculations that will who is now pursuing a PhD simulate the stress-strain behavior degree at Northwestern of the material. This is necessary to University. Other contributors Figure 2. Schematic illustration of the work- determine the optimum data-col- include Justin Fox (currently a sen- ing principle of the SMARTS Expert System software. lection points so that all critical ior in E&AS) and Dr. Bjorn events during a material’s deforma- Clausen (of Los Alamos). The soft- tion (for instance, its yield point) ware was written in Java, so it can can be captured. Another planned be used on different computing stress/strain measurements using upgrade involves optimization of platforms and run over the diffraction. The outcome will likely experimental conditions using Internet. We expect it to be adopt- be a rapid growth of the field and inverse problem analysis. This will ed by various national facilities, its application to a multitude of yield the mathematically most opti- both for neutron and x-ray diffrac- materials science and engineering

engenious winter 2003 progress reports

problems in both academe and We have also used SMARTS To achieve greater understanding, industry. to study bulk metallic glass (BMG) we have started working on model During the commissioning matrix composites developed at specimens suitable for high-energy phase, we used SMARTS for a Caltech by Professor Bill Johnson’s x-ray diffraction studies. By com- variety of projects. In a study fund- group. These composites retain the bining the neutron-diffraction data ed by NASA, we investigated high- high strength of BMG but improve we have obtained so far with the temperature deformation mecha- it further by providing ductility spatially resolved x-ray diffraction nisms in structural ceramics and and damage tolerance. Our aim was data, we intend to elucidate the ceramic-matrix composites. Some to understand deformation mecha- complete, multiscale deformation of these materials are already in use nisms in these composites and to mechanisms in BMG matrix com- in new jet turbine engines, but identify the best reinforcement posites. before they can be employed fur- material and its morphology. Some ther, it is necessary to understand BMG matrix composites their “creep” behavior. Creep refers require applied stresses over n short, the SMARTS to permanent (i.e., inelastic) defor- 2 GPa to fully observe their I system we have built mation at high temperatures. This deformation. However, since together with the Expert understanding will allow us to con- they include heavy elements System software allow struct advanced models that predict (such as zirconium and tungsten) unprecedented experimental capa- the lifetime of these materials that absorb x-rays, neutron diffrac- bilities that are revolutionizing our under demanding conditions (tem- tion (and SMARTS specifically) is ability to characterize materials in peratures above 1,200°C, highly the only technique available to situ under a variety of environmen- corrosive atmospheres, and so on). study in-situ deformation of the tal conditions close to what materi- Since SMARTS is able to provide reinforcements under high applied als will actually encounter. This is temperatures similar to those found stress. expected to lead to a better under- in a jet turbine, we collected in-situ Due to its amorphous nature, standing of how various materials crystallographic data for the first the BMG matrix cannot be inter- fail, and how we can improve the time for one of the most important rogated directly with diffraction to design of practical systems, such as structural ceramics, Si3N4. The dif- obtain lattice-strain data. However, aircraft, cars, engines, buildings, fraction data (including lattice we were able to use diffraction data and even microdevices, to avoid plane specific strains) were used in to develop new mechanics models such failure. a self-consistent model to calculate (finite-element or self-consistent) the elastic stiffness tensor of this that allowed deduction of the material at this temperature—a cal- behavior of the BMG matrix. We Ersan Üstündag is Assistant Professor culation previously unattainable. In showed that in all composites, the of Materials Science. late 2002, additional Si3N4 tests metallic reinforcements yield first were conducted in the creep and then start transferring load to There is more on Professor Üstündag at regime. The results suggest that the BMG matrix. The matrix later http://www.matsci.caltech.edu/ grain rotation and boundary sliding deforms by initiating multiple shear people/faculty/ustundag_e.html are active creep mechanisms. This bands that make it “plastic,” and more about his project at is the first time that they have been enhancing the overall ductility of http://smarts.caltech.edu observed in situ. The data are now the composite. The full microme- being used to develop a new chanical details of these events are mechanics model. still not fully understood however.

10 11 Distributed Integrated Circuits: Wideband Communications for the 21st Century by Ali Hajimiri

lobal communications mitted per second (i.e., bit rate) G have rendered our world a determines the speed of a digital smaller, yet more interest- communications system. C.E. ing place, making it possi- Shannon, the founder of modern ble to exchange visions, ideas, information theory, proved that the goals, dreams—and PoKéMoN maximum achievable bit rate of a cards—across our small planet. digital communications system Modern communications systems, increases linearly with the available such as the internet and portable range of frequencies (i.e., channel wireless systems, have added new bandwidth) and logarithmically dimensions to an already complex with the signal-to-noise ratio. world. They make us aware of our Thus, three critical parameters, similarities and differences and give namely, bandwidth, signal power, us an opportunity to communicate and noise, are the most important with people we have never met parameters in determining the per- from places we have never been. formance of any given communica-

The fusion of education with com- Ali Hajimiri tions system. munication is already bringing One of the more common about new levels of awareness, methods of increasing the band- accompanied by creative upheavals million) of reliable active (e.g., tran- width, and hence the bit rate, of in all aspects of modern life. sistors) and passive (e.g., intercon- any given system is to migrate to However, the ever-increasing nect) devices. Further, they are rela- higher operating frequencies. The demand for more connectivity tively inexpensive to incorporate maximum speed of operation in inevitably increases the complexity into mass-market products. electrical systems is determined by of such systems. Integrated systems The realization of revolution- the performance of both active and and circuits continue to play a cen- ary ideas in communications passive devices. While in modern tral role in the evolution of compo- depends heavily on the perform- integrated-circuit technologies the nent design. Silicon-based integrat- ance of the integrated electronic single-transitor maximum frequen- ed-circuit technologies (particularly circuits used to implement them. cy of operation can be quite high, complementary metal oxide semi- Let’s consider some well-estab- actual circuits rarely operate any- conductor, or CMOS) are the only lished theoretical background for a where near these frequencies.1 This technologies to date capable of pro- moment. The maximum number of provides further motivation to pur- viding a very large number (over a bits (1s and 0s) that can be trans- sue alternative approaches to allevi-

1 A transistor in a given process technology is usually characterized by its unity-gain frequency shown as fT . This is the frequency at which the cur- rent gain (the ratio of the output current to input current) of a transistor drops to unity. While the unity-gain frequency of a transistor provides an

approximate measure to compare transistors in different technologies, the circuits built using these transistors scarcely operate close to the fT and usually operate at frequencies 4 to 100 times smaller depending on the complexity of their function. There are two main reasons for this behavior. First, many systems rely on closed-loop operation based on negative feedback to perform a given function independent of the parameter variations. An open-loop gain much higher than one is thus required for the negative feedback to be effective. This higher gain can be only achieved by operat-

ing the transistors at a lower frequency than the fT to provide the desired gain. Second, integrated passive devices necessary in most of the high- speed analog circuits have their own frequency limitations due to parasitic components that can become design bottlenecks. The combination of these two effects significantly lowers the maximum frequency of reliable operation in most conventional circuit building blocks

engenious winter 2003 progress reports

ate bandwidth limitations, particu- traveling wave on the input line. larly in silicon-based systems Each transistor adds power in which, despite their reliability, suf- phase to the signal at each tap fer from low transistor speed, poor point on the output line. The for- passive performance, and high ward traveling wave on the gate noise compared with other tech- line and the backward (traveling to nologies. the left) wave on the drain line are The complex and strong inter- Figure 1. A distributed amplifier consisting of absorbed by terminations matched relations between constraints in two transmission lines and multiple transis- to the loaded characteristic imped- tors providing gain through multiple signal modern communications systems ance of the input line, R , and out- paths that amplify the forward traveling in have forced us to reinvestigate our wave. Each transistor adds power in phase to put line, Rout, respectively, to avoid approach to system design. “Divide the signal at each tap point on the output reflections. and conquer” has been the principle line. Each pathway provides some gain and In a distributed amplifier, one therefore the whole amplifier is capable of used to solve many scientific and providing a higher gain-bandwidth product tries to avoid a “weakest-link” situ- engineering problems. Over many than a conventional amplifier. ation by providing multiple, equally years, we have devised systematic strong (or equally weak) parallel ways to divide a design objective paths for the signal. In the absence into a collection of smaller projects gle signal path, distributed inte- of passive loss, additional gain can and tasks defined at multiple levels grated systems and circuits rely on be achieved without a significant of abstraction artificially created to multiple parallel paths operating in reduction in the bandwidth by render the problem more tractable. harmony to achieve an objective. addition of extra transistor seg- While this divide-and-conquer However, this multiple signal-path ments. This is the direct result of process has been rather successful feature often results in strong elec- multiple signal paths in the circuit. in streamlining innovation, it is a tromagnetic couplings between cir- The extended bandwidth of the double-edged sword, as some of the cuit components, which makes it distributed amplifier comes at the most interesting possibilities fall in necessary to perform the analysis price of a larger time delay between the boundary between different and the design of distributed cir- its input and output, as there is a disciplines and thus hide from the cuits across multiple levels, a task trade-off between the bandwidth narrow field of view available at not crucial when using the “divide and delay in an amplifier. each level. Thus, approaching the and conquer” approach. Alternatively, one can think of this problem across multiple levels of This concept can be best seen approach as a method of absorbing abstraction seems to be the most through the distributed amplifier the parasitic capacitances of the promising way to find solutions not (originally suggested by Percival transistors into the transmission easily seen when one confines the and first implemented by Ginzton) lines and making them a part of search space to one level. sketched in Figure 1. This amplifier the passive network. Distributed circuit and system consists of two transmission design is a multi-level approach lines on the input and the out- allowing more integral co-design of put, and multiple transistors t Caltech, one of our most the building blocks at the circuit providing gain through multi- A exciting breakthroughs and device levels. This approach ple signal paths. The forward has been in the area of can be used to greatly alleviate the (to the right in the figure) silicon-based distributed frequency, noise, and energy effi- wave on the input line is amplified circuits for communication systems; ciency limitations of conventional by each transistor. The incident we have achieved unprecedented circuits. Unlike conventional cir- wave on the output line travels for- performance for communication cuits, which often consist of a sin- ward in synchronization with the blocks and systems.

12 13 In particular, we have used the This concept has been successfully have been made in this direction, a concept of distributed systems to demonstrated in the distributed watt-level, truly fully integrated demonstrate an extremely high- voltage-controlled oscillator of CMOS power amplifier has not speed voltage-controlled oscillator Figure 2b where alternative signal been demonstrated using the tradi- using a low-performance CMOS paths have been introduced to tional power-amplifier design tech- technology with small cut-off fre- change the electrical length seen by niques. quencies for the active and passive the traveling wave. components (see Figure 2). This Another component we oscillator uses the delay introduced have devised is the distributed wo main obstacles in the by the distributed amplifier to sus- active transformer (DAT) T design of a fully integrated tain electrical oscillation by contin- power amplifier. The design of power amplifier are the uous amplification of the signal a fully integrated silicon-based low breakdown voltages of around a loop. The oscillation fre- power amplifier with high output transistors and the high losses of quency is determined by the round- power, efficiency, and gain has been passive components. The low trip time delay, i.e., the time it takes one of the unsolved major chal- breakdown voltage limits the volt- the wave to travel through the lenges in today’s pursuit of a single- age swing at the output node, transmission lines and get amplified chip integrated communication sys- which in turn lowers the produced by the transistors. tems. Although several advances output power. The high passive loss Tunability is an essential fea- reduces the amplifier’s power effi- ture for such distributed voltage- ciency by dissipating the generated controlled oscillators (DVCOs), and power in the signal path. These thus it is necessary to devise a problems are exacerbated in most method to control the oscillation commonly used CMOS process frequency. The oscillation frequency technologies, as the MOS transis- is inversely proportional to the total tor’s minimum feature size is con- delay and hence the total length of tinuously scaled down for faster the transmission lines. This proper- operation, resulting in lower sub-

ty leads to a frequency tuning Figure 2a. The racetrack analogy. strate resistivity and smaller break- approach based on changing the down voltages. effective length of the transmission Our DAT power amplifier uses lines. Frequency control can be the distributed approach to perform achieved by introducing shortcuts impedance transformation and in the signal path. This concept can power combining simultaneously to be seen using the racetrack analogy achieve a large output power while of Figure 2a. Here the signals trav- maintaining acceptable power effi- eling on the input and output lines ciency. It overcomes the low break- are analogous to two runners on down voltage of short-channel two tracks running side-by-side to MOS transistors and alleviates the be able to pass a torch at all times. substrate loss problems by provid-

The time it takes them to complete Figure 2b. A die photo of the 10-GHz distrib- ing the power gain through multi- a lap (oscillation period) can be uted voltage-controlled oscillator using 0.35- ple similar stages and signal paths. changed by introducing symmetri- µm CMOS transistors with a tuning range of Figure 3a shows the essential 12% and phase noise of –114dBc/Hz at 1-MHz cal shortcuts for both of them and features of the DAT, which consists offset. In addition to higher frequency of controlling what percentage of the oscillation, the DVCO provides better fre- of multiple distributed push-pull time they go through each one. quency stability. circuits in a polygonal geometry.

engenious winter 2003 progress reports

Each side of the square is a single the substrate power losses while multiple signal paths in a DAT amplifier consisting of a transmis- providing a large overall output necessitates an analysis and design sion line, two transistors, and input power using low-breakdown-volt- approach spanning architecture, matching lines. This particular age MOS transistors. The strong circuits, device physics, and electro- positioning of the push-pull ampli- electromagnetic coupling between magnetics. fiers makes it pos- These examples demonstrate sible to use a wide some of the basic concepts of dis- metal line as the tributed integrated circuit design. drain inductor to pro- The combination of multiple dis- vide natural low-resist- tributed signal paths working in ance paths for the dc and harmony and a design approach ac currents to flow. covering several levels of abstrac- The four transmission tion allow us to achieve higher fre- lines are used as the primary quencies of operation, higher circuit of a magnetically cou- power and efficiency, while creat- pled active transformer. The ing more robust systems. output power of these four push- Bringing this state-of- pull amplifiers is combined in the-art technology into the series and matches their small drain commercial realm, substi- impedance to the load. These four Figure 3a. The essential features of our distrib- tuting easily mass produced silicon- push-pull amplifiers, driven by uted active transformer (DAT). based circuits for the traditional alternating phases, generate a uni- GaAs-based circuits in use today in form circular current at the funda- everything from cell phones to mental frequency around the communications satellites, will fur- square, resulting in a strong mag- ther the revolution in communica- netic flux through it. A one-turn tions systems that defines our mod- metal loop inside the square is used ern era. to harness this alternating magnetic flux and acts as the transformer Ali Hajimiri is Assistant Professor of secondary loop. This is where mul- Electrical Engineering. tiple signal paths converge. Using Figure 3b. A die photo of the 0.18-µm CMOS the DAT, a fully integrated watt- DAT generating 3.6 W of power at 1.9 GHz level power amplifier was demon- into a 50-Ω load using a 1.8-V power supply, There is more on Professor Hajimiri at while achieving a power added efficiency of http://www.chic.caltech.edu strated in a standard CMOS 51%. It is fully self-contained and uses no process technology for the first external components. The distributed nature time, as shown in Figure 3b. The of the DAT structure reduces the sensitivity distributed nature of the DAT of the power amplifier’s efficiency to the sub- strate power losses while providing a large structure reduces the sensitivity of overall output power using low-breakdown- the power amplifier’s efficiency to voltage MOS transistors.

14 15 emeritus

William Bridges: A Rare Combination Of Talents by Amnon Yariv

ease, but who also could fix cars today’s standards, were able to pur- and radios. In addition, he was sue fundamental ideas. The world’s genuinely nice and gentle-man- first laser, the ruby laser, had been nered and unaware of the rare com- invented there by Theodore bination of his talents. Maiman. Some early attempts at Bill graduated in 1956 with a HRL to make He-Ne lasers (newly BS in Electrical Engineering. He invented at Bell Labs) were unsuc- was among the top five in the cessful and Bill was asked to help Berkeley class of 5,000 and also the because of his background in tube top engineering student. Now mar- and vacuum techniques. Before ried, he chose to stay on for gradu- long, Bill found himself immersed ate school in Berkeley and pursue a in the new area of gas lasers, lasers illiam (Bill) Bridges, Carl thesis in microwave tubes (a tech- in which the lasing medium is gas W F Braun Professor of nology still used in microwave present in a mixture of some other Engineering, turned amplifiers in communication satel- gases and excited by an electric dis- emeritus in July 2002, lites). Bill would finish his doctoral charge. thereby closing one chapter in a research under Professor Ned Bill’s biggest claim to his very varied and productive life and Birdsall doing pioneering work on considerable fame occurred at this career, and opening another. I ran instabilities in electron beams in a juncture. While trying systemati- into Bill in my second (his first) vacuum. These instabilities—spon- cally to understand the lasing of year at Berkeley in 1952. We have taneous voltage and current oscilla- Hg, Bill tried mixtures of He-Hg, stayed friends and close profession- tions—were thought to be the Ar-Hg, and other noble gases. In al colleagues to this date, so when cause of much of the performance these experiments he observed a asked to give an overview of his degrading noise in vacuum-tube new and intense blue laser emis- career, I jumped at the opportunity. amplifiers and oscillators. Their sion. After a process of substitution Bill was born in Inglewood, original work has been just redis- and elimination and very careful California on Thanksgiving Day, covered and used recently in Los spectroscopy, which Bill learned on 1934. Bill lost his father at an early Alamos and Russia for extreme the fly, he was able to trace the las- age. The vacuum left was filled by a high-power high-gain oscillators— ing to the argon ion Ar+. This grandfather and great uncle who one man’s instability is another would lead to the discovery of las- introduced Bill to tinkering, build- man’s gain. ing in krypton and xenon as well, ing things—including amateur Bill received his PhD from and to a new class of lasers, the radios. Early on, he acquired that Berkeley in 1962. After considering noble gas lasers. It is difficult to hands-on, “I can build anything” a whole slew of employment offers, work in any physics or chemistry approach that would serve him so Bill chose to join the Hughes laboratory in the world today with- well as a scientist/engineer and as a Research Laboratory (HRL) in out bumping into Bill’s “Argon teacher. Malibu, California. HRL at the Laser.” The invention of this laser I still remember my first time was unique, and probably one was made possible by the unique impression of Bill. I came from a of the most exciting research labo- blend of talents that Bill possesses: background where to be good in ratories anywhere. Run essentially the insistence on understanding at math meant being theoretical and as a non-profit organization by the most basic level why something no good with your hands. Here was Caltech, Berkeley, and Stanford works, the hands-on ability to a kid who handled the tough PhDs, it provided a home to excep- make things work, and the keen Berkeley engineering courses with tional scientists who amazingly, by intellect to combine the two.

engenious winter 2003 The following few years saw also resulted in his becoming, a years, these modulators became one Bill become an internal “guru” at year after his arrival here, the exec- of the key devices in the quickly HRL; he was often asked to work utive officer for EE. Soon after his expanding technology of high- on their most advanced and ven- arrival, Bill’s inability to say no to speed optical-fiber communication. turesome programs. These included worthwhile causes also landed him In an amazing but separate story, gas-dynamic lasers, adaptive optics, on the EE Search Committee, and Uniphase (now JDS Uniphase) and atomic-clock gas masers. But subsequently on the Patent, Health, became the world’s leading manu- he was also being drawn more and Freshman Admissions, and facturer of optical communications more into management and away Undergraduate Academic devices. Bill, who was paid “mostly from the laboratory. Standards and Honors committees. in stock,” became “comfortable,” A Caltech Sherman Fairchild He also became involved with the and was able with his wife, Linda, Scholarship during 1974–1975 pro- Society for Women Engineers as to build their dream home in the vided relief from his management well as the Amateur Radio Club. woods near Nevada City in north- role. Bill spent most of the year In a short span of three years ern California. teaching an optics lab that served Bill had become one of the most Bill’s talents and achievements to remind him how much he involved and effective faculty mem- have, of course, been noted by the enjoyed teaching and interacting bers, whose contributions extended world at large. Besides garnering with students. It also convinced a well beyond his research program, most of the major awards in the group of us in Applied Physics and as well as one of the most sought- optics and laser fields, Bill is Electrical Engineering that Bill after teachers. among a very small number of peo- would make a splendid addi- ple who are elected members of tion to the faculty. An offer both the National Academy of was made and accepted and by n the research side, Bill Science and the National Academy 1977 Bill had joined Caltech. O switched gears at Caltech of Engineering. He also squeezed One of his first projects and started looking into in a presidency of the Optical was to set up and teach a extreme high-speed elec- Society of America. demonstration class in optics, for trooptic modulators. These are What is Bill going to do now? which he built much of the equip- optical waveguides “written” Once the house up north is fin- ment himself. His reputation as a through selective doping through ished, he plans on dividing his time teacher with a hands-on approach masks into electrooptic crystals between Pasadena and the woods, from his Fairchild Scholarship such as LiNbO3. When high-speed traveling more, and finally, getting sojourn caused 70 students to regis- digital voltage pulses are applied to back into the laboratory. I personal- ter for the class—nearly a third of such waveguides they can switch ly was relieved to hear that a good that year’s sophomore class. Bill’s light on and off. The work of Bill fraction of his time will still be hopes of secluding himself in the and his students helped turn this spent at Caltech. The school can- laboratory, however, did not quite modulation scheme to the domi- not afford to lose his splendid materialize (which was partly his nant method of launching bits into counsel and input. fault). He recognized very early optical-fiber systems. that Caltech’s electrical engineering The involvement with LiNbO3 The author, Amnon Yariv, is the students could be better served.The bore some unexpected, and to Bill, Martin and Eileen Summerfield lack of an official EE major in the sweet fruit. Bill had been serving Professor of Applied Physics. curriculum with a set of required since 1986 on the board of a small courses left students confused, and company, Uniphase, which made often resulted in students graduat- small, mundane lasers. When the To learn more about Bill Bridges, visit ing without such basic EE courses CEO started looking at new targets http://www.ee2.caltech.edu/ People/Faculty/bridges.html as electromagnetic theory. Bill’s of opportunity, Bill encouraged the crusade to institute an EE major purchase of the LiNbO3 optical- with well-prescribed requisite modulator business of United courses was highly successful, but it Technologies. In a matter of a few

16 17 What do the complex workings of a cell have in common with the relentless unrest of the New York Stock Exchange? Are these dynamic storehouses of information obeying similar laws in terms of the ebb and flow of information, and can they be modeled, analyzed and understood by a unified mathe- matical framework? Caltech is launching a new Institute-wide intellectual adventure with the cre- ation of the Information Science and Technology Institute (ISTI)—drawing the curtain back, so to speak, on the nature of information itself, and redefining the way we approach, understand, and implement information science and technology.

When You Come to a Multidimensional Fork in the Road, Take It: Information Science and Technology at Caltech

Within the next decade, information at Caltech will be a unify- anism for bringing people and their ideas together from each of ing, core intellectual theme spanning the physical, biological, and the “six corners” of the Institute. social sciences, and engineering. Such a formidable, collective leap forward is the result of two idiosyncrasies: Caltech’s long- Over the last year, in an effort to define and help grow an IST standing and imaginative blending of traditional disciplines community at Caltech, groups of faculty convened, conferred, and the low one or two degrees of separation between disciplines, and converged on a set of unifying principles for four new faculty, and students which allows exceptional people from seem- research centers that together provide the critical mass necessary ingly disparate fields to work together naturally. Put another to launch ISTI. Jehoshua (Shuki) Bruck, Gordon and Betty way: we’re fabulously small, we engage in a lot of scientific gos- Moore Professor of Computation and Neural Systems and sip, and the standard departmental boundaries are all but invis- Electrical Engineering, chaired the IST Faculty Planning ible around here. Committee, which issued its final recommendations in early January. The proposed centers, ultimately to be housed in a new ISTI’s interdisciplinary research, academic, and outreach agenda building, are the Center for Biological Circuit Design (CBCD), is large and will develop roots in each of Caltech’s six divisions, the Center for the Physics of Information (CPI), the Social and with participation of more than 20% of the faculty, and nearly Information Sciences Laboratory (SISL), and the Center for the 35% of all students through curriculum. We aim to create a com- Mathematics of Information (CMI). These four new centers mon language for the study of information, will join the established Lee Center for Advanced Networking one that will stimulate fundamentally new and the NSF Center for Neuromorphic Systems Engineering to thinking about problems facing not only the form the initial core of ISTI. As ISTI matures, research advances usual suspects (computer science, quantum and the natural dissolution of older research initiatives will physics, electrical engineering, applied drive the creation of new centers. physics, and applied mathematics) but also those not normally associated with informa- From these vibrant centers will emerge a unique academic pro- tion science and technology such as experi- gram, the first of its kind in the country. The new undergradu- mental economics, pure mathematics, and ate and graduate programs will combine engineering and science developmental biology. By approaching with a clear focus on information, and direct exposure to the cen- information science and technology from tral issues across the entire intellectual landscape. And finally, to multiple levels of abstraction, we’d also like create the broad societal impact commensurate with the out- to figure out new tricks for atoms, light, mol- standing research and academic components of ISTI, we will ecules, cells, circuits, algorithms, and net- design and conduct a highly visible outreach program. Through works. executive, visitor, and industrial affiliate programs, we hope to supplement and share Caltech’s contributions by collaborating What will be the outputs? Absolutely smash- with members from key academic institutions, government, and Jehoshua Bruck ing scientific and engineering discoveries, industry. Workshops, lectures, and summer schools will round out students who’ll go out into the world and (we the menu for the continuing revolution in information science hope) one-up their thesis advisors, and technological advances and technology. only yet imagined in our wildest dreams… Listen in on the following four conversations Such an ambitious program will involve two phases. The first is among Caltech faculty engaged in thinking about the creation of the research component, the wellspring from what these new centers will bring to Caltech and which the corresponding academic and outreach programs of the society at large, as Caltech embarks on this second phase will flow. And we need look no further than three unparalleled and profound exploration. words—multidisciplinary research center—for the formal mech-

engenious winter 2003 Center for Biological Circuit Design: Soft Circuitry and Liquid Algorithms— A New Bioengineering Frontier Takes Form A Conversation with Niles Pierce, Paul Sternberg, Erik Winfree, and Barbara Wold

Biology computes, that is, living structures store, process, and communicate information in organ- isms and ecosystems. The CBCD is being organized to understand the form and function of these biological circuits and to develop the tools needed to design new and improved circuits.

WOLD: There is a computing rev- tweak things a lot. We break things But what gives a biological system olution going on across the board and see what happens. That’s the its real properties—for instance, its in many areas of biology—from heart of classical genetics. Or, we robustness in the face of various molecular, to cellular, to develop- take things out of the cell and kinds of insults? What are the mental and neurobiology. At an make them work in a test tube— dynamical properties of important obvious level, the revolution is driv- that’s biochemistry’s challenge. So circuits? How is information really en by rapid changes in the kind and at this point, from all of our tweak- encoded or stored by a given amount of data we work with, ing, the biologists have learned a molecular or cellular circuit? beginning with entire genome lot about the molecular compo- Getting at these questions using a DNA sequences and everything design focus is the core mission I that now flows from them. The see for the CBCD: it will take the basic challenge is to turn data into ...you don’t really understand fruits of all the research of past real information, then turn that the properties of something decades, combine it with the cur- information into real understand- until you can sit down and— rent revolution in biological infor- ing. At another level, biologists from scratch—design it, test it, mation processing, and focus on have long been interested in infor- circuit design. This is tremendously mation in living systems—how it is and see if it behaves as you exciting, and central to deep under- encoded, stored, recalled, and trans- predicted: have you got it standing of biological systems. duced from one site to another. right? These are themes that the faculty STERNBERG: Another way to in this Center will be addressing in describe what we want to do is the a very particular way. nents of gene circuits. Similarly, “reverse engineering” of biological After talking to many of our neurobiologists know a lot about systems and circuits. But it’s going faculty and colleagues, in and out the cellular components of neural to be much easier to learn how to of the Biology Division, we hit circuits that ultimately lead to brain do it with, for instance, a Model T upon the idea of focusing on bio- function and behavior. In the mid- rather than a Boeing 777. Organ- logical circuit design. In some dle are the people studying signal isms have been around for a billion sense, you don’t really understand transduction—that is, how signals years, making nature’s designs the properties of something until travel from the outside of the cell incredibly complex and sophisticat- you can sit down and—from to the inside of the cell, or from ed. Even the simplest organisms are scratch—design it, test it, and see if one cell to another. intricate integrated machines. They it behaves as you predicted: have The state of the art is this: we have embedded controls that are you got it right? I’m not an engi- know an enormous amount about really hard to tease out. It’s a lot neer, but I think that’s a major what the circuit components are, easier to build something from engineering process, or at least an and something about how they’re scratch and then learn how to important one. Biologists have hooked together. We know a good model it. been, so far, quite timid about deal about how the inputs work, In my lab, we’ve looked at sig- wholesale design. We go in and and, globally, what the outputs are. nal transduction and we’ve come up

18 19 eng enious atclrfntos Mylabworks functions. particular tion toscreen formolecules with a novel approach for lab hasdeveloped Richard Roberts’ orenhancedfunctions. new lution toobtainmoleculeswith usesdirected evo- bycontrast, lab, Frances Arnold’s novel functions. or proteins withenhancedstability computational methodstodesign Mayo’s Steve labuses components. have beenworking on the toolbuilderswho wehave as Paul said, First, challenging topic. to progress onthisvery directly that contribute are allworking inareas people atCaltechwho of three different types PIERCE nents. make thosecompo- thinking abouthow to and theircolleaguesare Frances Arnold, Mayo, Steve Niles Pierce, Rightnow, thought. and thattakesalotof alent ofthosebricks, want tohavetheequiv- You bricks. using Lego this tocreating asystem You liken can robust. goingtobe really which componentsare have todetermine We and interactions. to definecomponents weneed circuits, logical To buildbio- comes in. approach synthetic That’s where the simple. very wanttostart really Thatmeansyou to makeitwork. selection processes ofevolutionary littledetailhasbeentunedby every because theyfallapart, real cell, Butwhenwegointothe powerful. nicemodelsthatare with very : We have cbcd in-vitro lcws rmtplf:NlsPec,Pu trbr,BraaWl,andErik Winfree. Barbara Wold, Paul Sternberg, NilesPierce, from top left: Clockwise selec- Paul Sternberg, Eric Davidson, Eric Paul Sternberg, Leibler, Stan Elliot Meyerowitz, MelSimon, including circuits, occurring and functionofnaturally whostudythestructure elsewhere] a numberofbiologists[here and wehave Finally, circuits inbacteria. design ofcellularsignal-processing Arnold are collaboratingonthe andFrances Jared Leadbetter, Quake, Steve process information. bedesignedto can systems logical computation andissuesofhow bio- group isinterested inbiological Erik Winfree’s of DNA andRNA. designing molecularmachinesout for on computationalalgorithms biological circuits. biological this questtodesignandunderstand tobrain neuroscience in neering engi- all thewayfrom chemical interests allows ustomakebridges and intheirexpertise Bringing in acell. instance, systems—for ittoothercomplex how toapply out ing thisapproach andfiguring Theyare goodatarticulat- neously. simulta-ple placesinonestructure issuesforcircuit component-level diverse setofpeopleworking on there isa So amongothers. Wold, andBarbara Tom Knight, Weiss, Ron JohnMcCaskill, Collins, Jim Thanos Siapas, Kennedy, Mary winter 2003 lyze acomplexcircuit. lyze ofhowent view toana- they haveadiffer- study, they ties ofthesystems oftheproper- Because ones. to designnew and how onemightlike circuits work,occurring about how naturally gram—who are thinking pro- Neural Systems Computation and munity—the the neural biology com- the neuralbiology researchersincluding in massoftalent, a critical STERNBERG thetic pointofview. circuits fromcal asyn- toapproach biologi- try to now, right positioned, communities are well Allthree structured. tion andhow they’re how thosecircuits func- deep understandingof Thelatterhavea cuits. cir- occurring naturally orstudying ic circuits, creating synthet- design, information from multi- recordGilles Laurent and Thanos Siapas : There’s information science and technology

Caltech is more than ready to nent are important and which are impossible, became possible. The make this very interdisciplinary, just implementation details, not ability to build electronic circuits very ambitious goal happen. And really relevant to the function. This and integrate them with biology again, it can only happen here, leads to new levels of abstraction. brought to the table a new way of because in all the divisions you For instance, we ask: “Is this atom doing science. The possibility of have people who are really good at over here the critical atom, so I building biochemical circuits—for what they are doing, of course, but need to focus my attention down instance, novel genetic regulatory also imaginative enough and inter- here at the molecular level? Is it circuits to hold the concentration ested enough to be able to learn of an enzyme at a constant level, or other approaches. I think what will to trigger a reaction just at the happen in the Center is that at the Caltech is more than ready to right time—will provide an entirely start, everybody will come in with make this very interdiscipli- new approach for understanding his or her own ideas, leading to an nary, very ambitious goal what goes on inside the cell. incredible effervescence. Then we’ll happen. One thing that excites me is condense our focus on a couple of thinking about the relationship projects that seem tractable and between the concepts that comput- seem to be the right way to learn to critical to the function or not?” er scientists have developed and the prove principles that will lead to Researchers try to understand that realities that biologists are observ- new technology. The new technolo- by making mutations, changing a ing. Understanding the kinds of gies will then be applied in many moiety here or there, and so on. algorithms that biology has been directions and spawn new indus- This is another approach for deter- exploiting, and the design space of tries. mining which parts of a system are those algorithms, is fascinating. important, and which are merely Programming with biochemical WOLD: One of the other things accidental. I hope that going reactions rather than with logical the CBCD will spawn will be an through the design process will “and gates” and “or gates” is a entirely new generation of students help elucidate this completely. whole different beast. and post-docs with a worldview Many advances in biology have that is some interesting combina- been driven by instrumentation. ENGENIOUS: What are the prac- tion of all these inputs. Without tical goals of the CBCD? the Center, a few students might make the interesting connections Programming with biochemi- PIERCE: One way to encapsulate a that biological circuit design cal reactions rather than with long-range objective for the CBCD requires. With the Center, and the logical “and gates” and “or is to say that we’re going to try to concomitant “lowering of the ener- gates” is a whole different recreate the remarkable technology gy barrier,” so to speak, the path beast. of the compiler. A compiler takes toward this kind of research train- an algorithm written in a program- ing will be much more easily and ming language and turns it into frequently traversed. So the For instance, once people under- instructions that a computer can impact—through these people— stood that a feedback loop coupled understand. Given a conceptual ultimately goes far beyond Caltech. to an electrode in a cell could lead design for a circuit, we’d like to be to something like a patch clamp, a able to “compile” a set of molecules whole new way to characterize bio- that can be introduced into a test logical circuits was born. It became tube and be observed to function WINFREE: One of the problems possible to measure currents, I-V according to the principles for engineers face is understanding curves, and so forth. An entire which that circuit was designed. which aspects of a given compo- range of experiments, previously This outcome would be tremen-

20 21 eng enious ici nartoa a.Te,per- Then, circuit inarational way. andfititintoa it inyour toolbox, put it, thencharacterize evolution, example—by somekindofdirected for protein, nents—a particular the wayyou mightdesigncompo- and approach, that’s animportant WINFREE the firstplace. you werehow tooptimalin close Thisallows you tosee mization. foropti- rapidevolution to very thensubjectthesystem design first, WOLD tion. topredict howability itwillfunc- basedonyour together asystem where you put tematic design, sys- withrational, one thatworks, abunchofthingsandselect you try where approach, this “irrational” It willbeinteresting tointegrate tools. Theseare important proteins. tobothcircuits and ed evolution Frances Arnoldappliesdirect- ties. sequences withfunctionalproper- in-vitro does RichardRoberts For instance, is already beingdoneatCaltech. inthedesignprocess principles ary WINFREE this work? beusefulin biology evolutionary ENGENIOUS shot atmeetingit. butIthinkwehaveareal dard, Thisgoalsetsahighstan- cuits. produce working moleculesandcir- robustdesign framework enoughto a array ofcomponentstodevelop lenge ofworking withacomplex butalsoforthesheerchal- cations, appli- andmedical biotechnological forits excitingnotonly dously : selection todesignprotein A hybrid approach isto A hybrid : : bouey Ithink Absolutely. cExploiting evolution- bcd : Will principles of principles Will sector. canradiatethey outtothetech technologies toalevel where ting fundamental ideasand CaltechWhat doesbestisget- the compiler. the remarkable of technology ...we’re torecreate going totry motn i f“vlto”from bitof important “evolution” WOLD changed. andradically derfully research approach mightbewon- each person’s that willtakeplace, Butgiventhecollaborations them. and learnhow tobuildthingswith these known robust components We with couldstart timing device. acts asalittlemolecularswitchor It has beenobsessedwithforyears. moleculethatMelSimon a tein, wehavetheGpro- molecular level, Atthe ofcomponent. are onetype Neurons andelsewhere. colleagues, byJohnAllmanandhis Caltech, thing thatwasdiscovered at That’s some- makes usdifferent. It’s thecircuit designthat ent ways. but they’re wired togetherindiffer- neurons, as manyothercreatures’ neurons are thesame Our cessful. suc- are very forinstance, Neurons, maintained theircentralcharacter. buthave in manycircumstances, which componentshavebeenused toseewhat’s worked—at evolution STERNBERG thatcircuit. to optimize ofevolution doanotherlevel haps, : That maybethemost : Then you can look Then you can We know how toprogram and inyour digitalcamera. in your car, They are inyour microwave oven, becoming ubiquitousinourlives. whichare quickly microprocessors, Mostprograms existin cally. timehistori- short avery 50 years, inthelast andtechnology neering, engi- affectedscience, profoundly things andtoautomatetaskshas toprogram Theability realm. cal orbiologi- tothemedical restricted here are implications not nological WINFREE the techsector. where radiateoutto level theycan mental ideasandtechnologiestoa Caltech doesbestisgettingfunda- What spillover. biotechnological work willhaveasignificant ofthis theimplications Certainly that’s output. notapractical bymanyotherpeople, as viewed But would behavingthecompiler. result apractical tous, the day, Attheendof principles. underlying greatest passionisforthedeep Our outputs. potential practical whattodo.and know exactly they willbeattractedbythetheme WOLD but theydon’t know ityet… even leagues wethinkwillbeinvolved, There are alotofourcol- people. to changethedirection ofmany focus ondesigningcircuits isgoing andcommittingtoa articulating STERNBERG inaction. evolution else’s pointofview—intellectual ferent wayandtousesomebody other tolookataproblem inadif- intellectualpressureexert on each We our immediatepointofview. One lastthingconcerning : winter 2003 But wehavefaiththat : And thepossibletech- : And that’s why information science and technology

exert embedded control over tion and algorithms are fundamen- WOLD: Actually, at the end of the macroscopic electromechanical sys- tal elements of the chemical day, that’s the point. We don’t usu- tems, and this has revolutionized process. Nature has polymers, like ally start with that. What’s the goal technology. DNA, which contain information. of your Center? To have fun. But Nature, through biological The cell interprets that information we know it will be… processes, has transformed the as a program for directing its earth by exploiting algorithms and behavior. Evolution changes the embedded control at the chemical program to carry out an incredibly Niles A. Pierce is Assistant Professor level to fabricate cells, bodies, and wide range of functions. This is a of Applied & Computational ecosystems; to build forests from technology that isn’t just biological: Mathematics. Paul W. Sternberg is Professor of Biology and Investigator, light and chemical nutrients, for biology is only one possible result Howard Hughes Medical Institute. example. Intellectually, we don’t of programming biochemistry. Erik Winfree (PhD ’98) is Assistant really understand how these things Working with atoms and molecules Professor of Computer Science and exert an influence over chemistry in systems will turn out to encom- Computation and Neural Systems. and organize it into meaningful pass a wide world, and is going to Barbara J. Wold (PhD ’78) is the constructs. Biochemistry is where be very fun. Bren Professor of Molecular Biology we see most clearly that informa- and Director of Beckman Institute.

Center for the Physics of Information: The Impending Overthrow of the Silicon Monopoly: Revolutionary Substrates Unite! A Conversation with André DeHon, John Preskill, and David Rutledge

Silicon is a superb computational substrate...but sooner or later it will run out of room. The CPI is devoted to inventing the new computational substrates, architectures, and algorithms for the com- puting devices of the future.

PRESKILL: We’re kind of an odd mix of people, you know. I’m a the- oretical physicist, Dave’s an electri- cal engineer, André is a computer scientist. But I think we have some things in common. In those areas of overlap there’s a potential for some really exciting scientific and technological developments. We know that the advance of our infor- mation technology, which has been dazzling for so long, is confronting limitations that come from physics and, in particular, from the size of atoms. And we don’t know beyond say, a decade, what we are going to From left to right: André DeHon, John Preskill, and David Rutledge.

22 23 eng enious into practical devices. into practical neat scientificideastransformed scientists andengineerstoget really mean strong collaboration between up through thepractical—will integration—from thetheoretical Thisvertical integrated circuit. of theideastowork onasilicon getting some forexample, require, Thatwould products. engineering workside toseeiftheyreally in physics onthescientific, developed to takesomeoftheideasbeing tage. That’s Caltech’s really advan- area. whoare connectedtothis campus acrossdifferent the departments in many outstandingjuniorfaculty Caltechhashired Butrecently tion. There’s agreat tradi- Mead. Carver and JohnHopfield, Feynman, come tomindare Richard Three peoplethat information. ofsmallthingsand to thephysics making fundamentalcontributions of goodhistory Caltech hasavery sotospeak. Center isitsancestry, think isinteresting aboutthe RUTLEDGE advanced quantumcomputing. along theroad thatwillleadusto nite ideaabouthow toprogress don’tbut wereally haveanydefi- ideas aboutquantumcomputing, We havethesebeautifultheoretical It’s exciting. really computation. amazing advanceinthespeedof seean we’ll If itcomestofruition, interested inquantumcomputing. I’m Inparticular, and practice. between someofthoseconcepts but there’s atremendous gulf ofmanipulatinginformation, ways There are alotofideasaboutexotic thinking aboutinthisCenter. And that’s whatwe’re goingtobe know how we’re goingtogetthere. WeWe don’t don’t know what. going torequire ideas. new really It’s gottenaccustomedto. we’ve ofprogress do tocontinuethetype We here havetheopportunity

: cpi One thingthatI my way.” areold abstractions getting in these “Okay, the onessaying, ...really good engineerswillbe come from physics... confronting limitations that is been dazzlingforsolong, which has tion technology, ourinforma- ...the advanceof DEHON algorithms mightnotsitsohighin algorithms ENGENIOUS and soon. programming language, the what you useforcomputation, to re-evaluate allofourmodels: that’s goingtoforce So us atoms. few) ing withindividual(orvery substrates where wewillbework- ofcontrol with have thattype we’ll whether Butit’s notclear place. about amillionatomssittinginone we’rethings work because talking Silicon’s reliable; beenvery pose. have inthepastwilldefeatpur- andabstractionswe same interfaces here isthatusingsomeofthe clear very What’s getting inmyway.” theseoldabstractionsare “Okay, engineers willbetheonessaying, good andreally the costschange, change, whentherules other hand, Onthe nice setofdefinedlayers. There isa ontopofthat.” rithms andthenthere are algo- program; here iswhere webuildthe of that; build upsomearchitectures ontop thisiswhere we into thegatelevel; know “this iswhere wecollapse Andwe ontopofsilicon. software tions fordesigningcomputersand setofabstrac- well-developed very gota we’ve Presently, many layers. berethinking abstractionsat we’ll alsomeans tion onahigherlevel : I think vertical integra- I thinkvertical : So forinstance, So and the eventual implementation. and theeventual most efficientarchitecture possible, substrate wasinseparablefrom the ofthe physics that theunderlying more understand Ibegantoreally the outwith, VLSI work Istarted DEHON applications. ofphysical types won’t forcertain beapossibility quantumcomputingjust So regard. of goodfeatures mayfailinthat technologieswithlots Some errors. resistant to bevery potentially would system ofphysical what type wehadtorethink For instance, area isoneofmymajorinterests. this Inquantumcomputing, ple. rection thebestexam- isprobably Error cor- through from thestart. tothinkthings rithm—and andalgo- architecture, substrate, wellinthepast— usvery served to letblurthoselayerswhichhad We should bewilling than others. applications ter suitedforcertain maybebet- systems physical Some less centralthanitwasinthepast. willbe device a general-purpose PRESKILL attention oftheengineer. tothe needtobring things we’ll There are different and change. effects are goingtoshift probably Ithinkthedominant to optimize. up thedominanteffectsyou need detailandbring hide unnecessary abstractionis good engineering todoin One ofthethingsyou try things thatmaybelesshidden. RUTLEDGE: DEHON tal way… to dealwitherrors inafundamen- you have really numbers ofatoms, DEHON some way? fundamental substrateitselfin might bemore embeddedinthe the hierarchy anymore? They winter 2003 : : : The deeperIgotintothe …because there are…because some that. I wouldbelieve : Maybe theconceptof Maybe lo with smaller Also, information science and technology

And as VLSI got smaller, the land- And training students so they have Look at this from a student’s scape changed. Wires got more the broad background that’s neces- perspective. I maintain that our expensive, for instance. Ultimately, sary to get the big picture. current and future students will go our computations do depend on the out into the world and have the physics we use and the structure of ENGENIOUS: How will the struc- same impact the Caltech VLSI stu- the physical world. After looking ture of the Center facilitate break- dents are having now—maybe even down at VLSI for so long, it’s good throughs? more so if we can get students from to just look up and realize scientists every area to interact with each are working with some amazing RUTLEDGE: We’re interested in other. For example, a student comes new phenomena: carbon nanotubes, creating an environment conducive here to study molecular electronics, experiments trapping a single atom. but this area doesn’t exactly pan So you say to yourself, “How can ...with smaller numbers of out. However, the real benefit will we harness these things?” have come from interacting closely For me, a central issue is atoms, you really have to deal with other people doing perhaps understanding computational cost with errors in a fundamental biomolecular and quantum work, structure. When the cost structure way… and from being taught how to changes things radically, the nature think broadly about these areas. I of the solutions changes as well. think our students will certainly be The general-purpose processor that to professor and student interac- in a position to found, transform, made a lot of sense in VLSI just tion. And we’re anticipating that and lead the industry. doesn’t make sense for these new there will be a new Information things. We are off in a completely Science and Technology building as PRESKILL: The students are really new playground, which is very a result of the fundraising cam- the key. Caltech should be the exciting for an architect. Caltech is paign. University professors are place, the number one place, that a a place that allows me to think prone to being trapped in an area; student thinks of if he or she is sometimes at the circuit level, this is a good way to force them interested in the future of informa- sometimes at the manufacturing out into new things. tion technology in the long-term. level, sometimes at a mathemati- Actually André and Erik [Winfree] cal/statistical yield level—all over DEHON: People like Bill Dally did a great thing this summer— the map. And for something new [PhD ’86; now Professor of they were involved in the like this, where no established dis- Electrical Engineering and Computing Beyond Silicon cipline exists, it’s important to Computer Science at Stanford Summer School, which attracted gather people from various areas University] and others came to people from all over. who can think broadly about the Caltech in the early ’80s because it issues. This is what the CPI will was the place for VLSI. And that’s DEHON: We had 45 students for accomplish. really what we want—for Caltech four weeks and 12 guest lecturers— the top people—coming from dif- PRESKILL: We’re searching for It’s so important to have the ferent institutions and intellectual new paradigms, something that freedom to be daring... areas. It was really something. Caltech does especially well. Maybe we won’t be the place that actually ENGENIOUS: How did the stu- builds the next revolutionary gener- to be the place for the next revolu- dents deal with this new conceptual ation of devices, but I think what tion in novel computing. There’s a framework? we should aspire to is becoming the great deal of uncertainty about world’s leading institution for lay- what’s going to happen in this area, DEHON: It was interesting ing the scientific foundations which and yet that’s what makes it excit- because it’s not a “done thing,” will be the basis for information ing. What’s going to happen at the there is no orthodoxy. The students technologies of the future—we will chemical level? At the biomolecular [continued on page 27] be generating absolutely new ideas. level? At the quantum level?

24 25 Computing Beyond Silicon Summer School Or How I Spent My Summer in Pasadena

There are revolutions at hand in the way we understand an implement computation, driven by an awareness of impending barriers to VLSI scaling and new understand- ings of the physical world. This fundamental shift in per- spective allows us to contemplate engineering computa- tional substrates at the molecular and atomic levels. To develop and exploit these new substrates will require an intimate understanding of both the physical substrates and the nature of computation, as well as the relation between them. Research and researchers whose competencies span across the disciplines will be necessary to drive progress in this area of novel computational substrates…

…Thus the read the opening paragraph of the announcement for the Computing Beyond Silicon Summer School (CBSSS). Coordinated by André DeHon, assistant professor of computer science, and Erik Winfree (PhD ’98), assistant professor of computer science and com- putation and neural systems, the program brought together leading research faculty and 45 outstanding undergraduate and graduate stu- dents from many disciplines and institutions across the country (including Caltech). Part boot camp, part pleasure cruise, CBSSS served as an intensive four-week introduction to the emerging fields of molecular, biomolecular, and quantum computing. Lectures, reading assignments, and a paper and presentation project kept the students active. In between all this, students seized the opportunity to hang out with the guest lecturers, Caltech faculty, and each other. They came, they learned, they met future collabora- tors—and they had fun. A potent combination. And of course, ditto 3 for the faculty and guest lecturers… As a prototype of ISTI’s outreach program of summer schools, CBSSS’s unique collection of people and ideas in one place at one time points to the future of Caltech as a hotbed of research in novel computational substrates. i

For more information on who was there and what they did, go to http://www.cs.caltech.edu/cbsss

The CBS3 students gracefully posed for “mug shots” for posterity. To engage the students beyond the lectures, the CBS3 faculty asked them to self-organize into small project

p teams to expand on issues related to or motivated by the subject matter presented in lectures. The students had roughly three weeks to focus in on a topic and put together a brief report. See http://www.cs.caltech.edu/cbsss/report1.html for the resulting collec-

tion of student reports. Almost none of the students were “experts” in the issues they cbs studied when they entered the program. Nonetheless, these reports show that the mul- tidisciplinary teams assembled were able to dig deeply into a number of interesting

c problems and point out some promising directions for further inquiry.

engenious winter 2003 information science and technology

[continued from page 25] definitely went through a little big place—with something happen- in the middle of established proj- mind expansion. There were EE ing in the Media Lab, and then ects. Just to get out of the kind students who thought [the EE there are people over in the AI of environment where you’re told framework] was the only way the Lab, far from folks in Mechanical what to do every two months is world works...and in some cases Engineering. So you know, maybe liberating. biology students who didn’t at the it’s a little bit harder to get coher- outset realize that maybe computa- ence between the groups. PRESKILL: Absolutely. It’s so tional complexity meant something important to have the freedom to to them. All of them were chal- We are off in a completely new be daring, not to have to defend lenged and out of their comfort the project on the basis of some zones. I think many of them had playground... short-term goal, some milestone the experience of “Wow, the world event. is bigger than I thought it was.” There is an opportunity to do ENGENIOUS: What is the one RUTLEDGE: I want to mention interesting work at, for instance, thing that excites you most about that junior faculty will be instru- the intersection of computer sci- the Center? mental; they have already con- ence and biology. tributed in fundamental ways to PRESKILL: Well, from my own getting things started. People like PRESKILL: And in some ways, it’s parochial point of view, I’m excited Erik Winfree [PhD ’98], Ali easier for students than it is for us, about making quantum computers Hajimiri, Hideo Mabuchi [PhD you know. For me, the work I do at a reality. It’s just one of the emerg- ’98], André of course, and a hand- the interface of physics and infor- ing frontiers. If something like the ful of others. mation science seems kind of “out Center for the Physics of Infor- there,” novel and daring. But to my mation can make that possible, I PRESKILL: Yes, I think that’s students, it seems very natural. think that’s very exciting. pretty good evidence that we’re on Those are the things they’re inter- the right track. Looking around ested in. Combining computer sci- DEHON: The Center will really campus and seeing so many young ence and physics is second nature allow us the opportunity to build faculty involved in exciting projects for them. critical intellectual mass. My stu- at the interface of physical science dents and I can sit there and ask and information science tells me ENGENIOUS: Caltech seems to each other questions, but having that we are in a good position to have both a deep intellectual reser- the ability to work with people live up to the legacy of Feynman, voir and a smallness of size that from other areas thinking about the Mead, and Hopfield. allows us to attack these problems same problems will be powerful. much differently than anybody else. The new solutions will create new Are there other universities that abstraction hierarchies and new André DeHon is Assistant Professor of can do what you anticipate doing? ways of decomposing problems. Computer Science. John P. Preskill is the John D. MacArthur Professor of Things will not be the same as they Theoretical Physics. David B. RUTLEDGE : Smallness is a part were. Let’s think out of that Rutledge is the Kiyo and Eiko of it. Caltech feels the same size as proverbial box and come up with Tomiyasu Professor of Electrical the entire EE department at some wild ideas. Engineering. Berkeley. There, someone “far away” from you intellectually meant RUTLEDGE: I see two things. someone that was making super- One is the opportunity to work conducting detectors in the elec- with people across a wide range of tronics department. However, there disciplines in a serious way. And are a lot of good places out there, the second is consistent support. and a lot of competition. I’ve run government centers, and it’s astonishing how much of your DEHON: Certainly MIT has the life gets taken up by requirements breadth. On the other hand, it’s a and crazy things that change right

26 27 Social and Information Sciences Laboratory: Markets and Other Noisy Human Artifacts—Can Computation Bring Them Out of the Bronze Age? A Conversation with Yaser S. Abu-Mostafa, K. Mani Chandy, and John O. Ledyard

Social systems such as financial markets, political processes, and organizations aggregate and dis- seminate immense amounts of noisy information—but can this be done more efficiently? And can new, innovative structures be invented with the assistance of more sophisticated information tech- nology? SISL will be exploring these and related issues.

ABU-MOSTAFA: There is an something about markets already, ence and technology to design new abundance of data and an abun- that 200 years of study have pro- markets. Economists have generally dance of computational resources in duced remarkable insights about attacked these problems assuming the world, yet our ability to manage them. What’s of importance in this computation was free and easy— these resources, to be able to look Center, however, is the role of tech- which it’s not. Bringing the reality at data and efficiently extract the nology in the way markets operate. of information processing into mar- correct information, is limited. ket design is really important. The Highly distributed, data-rich, and The question is whether we role of SISL is to bring the exper- generally unstructured, the world’s tise of engineers and information financial markets seem to work can leverage new advances in scientists together with the exper- well—remarkably well given the information science and tech- tise of economists—each has some- loose structures and lack of super- nology to design new markets. thing the other doesn’t. Working vision—but they can be improved. together, something really special The players in the markets are will emerge. individuals, institutions, sometimes There are barter markets, which simply computer programs. They have been around for thousands of ENGENIOUS: Will you be invent- are looking at pieces of information years, which are not very efficient. ing new computational tools to deal that may be different from one The information technology under- with these problems? source to another. They’re all inter- lying the New York Stock preting information differently. Exchange is still primitive in that CHANDY: At this point, I don’t They have their own ideas and humans are crucial at many points think we really know. That’s why preferences regarding risk, value, in the process. Many aspects of SISL is so interesting. From my volume, etc. Eventually, all of this is markets work wonderfully. If I’m point of view, the research of this aggregated in global quantities like fixing my house and I need a nail, I center will bring “power to the peo- price, volatility—things of that sort. know I can go to the hardware ple.” Economic power has two So a basic understanding of how store, and the nail is sitting there parts: resources and information. such a general system results in waiting for me. How did they know Information technology today is at efficient information aggregation is I would need a nail that day? It’s a place where one half of the eco- very important for two fields: eco- not centrally planned. It’s not man- nomic power equation—informa- nomics and engineering. On the aged the way engineers like to tion—is widely available. And this economics side, we would like to manage things. It’s dispersed, dis- represents a significant dispersal of better understand markets and organized, decentralized, but it power from the few to the masses. eventually be able to design mar- does compute some pretty incredi- I’ll give you three examples of how kets. Once we do that, we can ble things. this is going to change your life. design markets in different arenas There are other pieces that When the defense department where there are no markets now. don’t work very well: supply chains, wants to buy planes, it puts out a From the engineering perspective, for instance, and public good kinds request for proposals, companies we’re interested in learning from of problems. Markets don’t work respond, and they finally choose a the principles of how markets work very well in these cases, partly plane. DOD can afford to do that how to generally manage distrib- because there aren’t very many par- because DOD budgets billions of uted information and be able to ticipants. They’re very specialized dollars for a plane. If you want to aggregate it in a meaningful way. and may not have much volume, so buy a car, you don’t have the same you can’t rely on immediacy. The flexibility. You don’t request pro- LEDYARD: Economists would question is whether we can leverage posals for cars that fit your specifi- suggest that perhaps they know new advances in information sci- cations. Nor, if you want to travel,

engenious winter 2003 information science and technology

do you put out a request for pro- futures contracts on carpenters, posals for tours with certain specifi- masons, roofers, and locks in a cations. You can’t do that because schedule. This is going to require the cost of the transaction is high. some interesting theoretical work But apply computational resources in terms of how you capture what to this scenario, and things will are essentially “metaphorical” change dramatically. ideas—the idea that I want a house The second example is futures overlooking a lake, with three sto- markets. We are familiar with the ries, etc. futures market on things like The classic example of where wheat, oranges, pork bellies, and so this gets mishandled is the on. But what if there were a futures California electricity market. That market on services like carpentry, was a designed new market. plumbing, and electrical work when Somebody said “Let there be mar- you add on to your house? kets,” and voilà! They did that in The third example is the cre- Russia and it was a disaster because ation of financial derivatives. Today, they forgot they needed banks and large financial services companies property rights and various other create financial derivatives tailor- things. In California, they forgot to made for companies doing ship- integrate engineering, electricity, building in Poland, for example. and the laws of physics with the Financial services companies create market. They also made some bad custom-made derivatives and sell assumptions about how people them for lots of money. But with behave. There’s been research, a lot the kind of technology we will of it at Caltech over the last 30 develop, companies will want to sell years, which could have prevented you derivative products for yourself this problem from occurring. based on your personal situation. Simon Wilkie had a very nice arti- So these are a few examples of cle in Engineering and Science how the Center’s research will help [Economic Policy in the economic power devolve to “the Information Age, E&S, Vol. LXIV, people.” No. 1, 2001, page 28] on just this problem. Engineers like to control LEDYARD: Here’s a sort of com- mon theme in the story: let’s say Designers of distributed you want to build or buy some- thing, a car or house or vacation. systems can control the rules Today, you have to go to somebody of the game, but they cannot who’s packaged everything up control the players. without your particular needs or desires in mind. You can have peo- ple specially build your cars for you, everything. Economists hate to specially build your house, but it’s control anything. Integrating these expensive. With computational two kinds of approaches is going to capability, you can allow people to be interesting, but it’s required for a express what they really want to successful energy market. Give buy in a marketplace. So, rather SISL up to ten years, and we’ll pull than hiring a project manager to it off. build your house, the computer From top to bottom: Yaser Abu-Mostafa, With experimental economics, organizes schedules, locks in the Mani Chandy, and John Ledyard. we have a way of demonstrating to

28 29 people howl these things really mechanism design. It’s also very so the work of that Center—creat- work. We can actually bring science much an economics problem, where ing efficient representation choic- to bear on it. The combined ener- we recognize the incentives people es—will be useful to the work of gies of those working in this Center have to follow the rules or not. SISL. will create an intellectual core that

anybodys working in these fields ENGENIOUS: What other ENGENIOUS: This work, taken as simply won’t be able to ignore. Caltech faculty do you anticipate a whole, sounds like it could be an being involved? entirely new intellectual discipline.

ABU-MOSTAFAi : I’d like to add something to the idea of exotic CHANDY: In computer science, derivatives: one of the biggest there are two relevant areas: apply- If we knew what would hap- advantages of having the computa- ing economic principles to distrib- pen two years from now, it tional technology to price these uted systems, and applying technol- wouldn’t be research.

things iss being able to communi- ogy to economic principles. For the cate the derivative to so many play- first part, we have Steven Low’s ers, thus creating a commodity. It work on the internet and algo- LEDYARD: It has the potential. becomes a real market—a place of rithms, and also John Doyle’s theo- exchange between buyers and sell- ries on control and robustness ABU-MOSTAFA: When you ers—because of the number of applied to non-traditional applica- design a research enterprise like players and because of their ability tions like markets. this, you have to have a gut feeling to come to an agreement on price about it being special. But then and to communicate instantly. LEDYARD: We have been using these things create a life of their markets as examples, because many own. If we knew what would hap- CHANDY: John said that engi- people have contact with markets. pen two years from now, it wouldn’t neers like to control things… but a But the same conceptual structures be research. Once the collabora- true distributed system is one in and questions arise in issues of vot- tions begin, who knows what can which you don’t know the partici- ing and elections, committees, and happen? We’ve been discussing pants, or even how many there are. organizing large organizations. In markets because they are tangible, Designers of distributed systems my Division, we have Tom Palfrey and have real and immediate can control the rules of the game, working on political processes. impact on people, but there is a but they can’t control the players. Peter Bossaerts studies the dynam- wide range of applications for this So there are two parts to a distrib- ics of financial markets and the research, including the organization uted system: the visible hand, or process of price discovery. Charles of corporations, the health-care Plott studies information aggrega- system, etc. Engineers like to control tion processes. Matthew Jackson does fundamental research on net- CHANDY: I really believe that everything. Economists hate to works. All of them will be involved, SISL will have a direct impact on control anything. as well as others. society, on ordinary people in addi- tion to large institutions. This con- CHANDY: We will also work with fluence of economics and informa- the rules by which all the partici- people from the Center for the tion technology will impact every- pants play, and then the invisible Mathematics of Information. We body. hand—how many participants, and share an interest in the growth of how participants operate provided data, the extent of data. Essentially, they play by the rules. Markets are data come in three forms. There are Yaser S. Abu-Mostafa (PhD ’83) is beautiful examples of this, and we structured data, like the price of a Professor of Electrical Engineering and Computer Science. K. Mani need to understand better how we car. There are totally unstructured Chandy is Simon Ramo Professor and get global behavior from these poli- data, like news about an explosion Professor of Computer Science. John cies. This is very much an engi- in Azerbaijan near an oil well. And O. Ledyard is Allen and Lenabelle neering problem. then there are semi-structured data, Davis Professor of Economics and for instance, auction information Social Sciences. LEDYARD: The process Mani is like you would find on E-Bay. All describing is what economists call three kinds are increasing everyday,

engenious winter 2003 information science and technology

Center for the Mathematics of Information: Information Theory Revisited: Mathematicians and Friends Tackle the Whole Enchilada A Conversation with Emmanuel Candes, Michelle Effros, and Pietro Perona

Mathematics has provided the foundation for virtually every major technological advance of human society. And now, there is a fundamental need to rethink the meaning and scope of computation, information gathering, and extraction. CMI will provide a home to the dedicated community of mathematicians, engineers, and scientists concentrating on developing the key mathematical ideas necessary to take information science forward.

EFFROS: There is a lot of excitement in research at the boundaries between traditional areas. The thrusts of ISTI reflect that excite- ment. One way to cross traditional boundaries is to focus on the appli- cations of informa- tion science. Another way is to work on the topics From left to right: Emmanuel Candes, Michelle Effros, and Pietro Perona. that different infor- mation science applications share—which will be will bring together mathematical throughs in mathematics. The job our approach. The CMI is focused tools from communications, statis- in the information sciences is by no on understanding the essential tics, signal processing, and comput- means done. Roughly, communica- nature of information itself, the er science with those developed tions looks at bandwidth, controls common properties shared by across a wide variety of applications looks at feedback, and computer information in all of its physical and build a shared foundation for science looks at computation. What forms and applications. We hope to studying information science. is needed for today’s more complex learn how to collect, quantify, com- Over the past 50 years, practi- systems, whether natural or municate, and manipulate informa- cal problems in communications, designed by people, is some way of tion efficiently. In studying the controls, and electronics have bene- capturing these things together and mathematics of information, we fited enormously from break- understanding how they interact.

30 31 eng enious out how toautomatethis process of We needtofind there intheworld. out amazing amountofclutter There isan are notbuiltthisway. Butmachines and makesenseofit. information toorganize try neously built inawaythattheysponta- are Humans andsoon. logical, techno- economic, sions—medical, deci- make allkindsofimportant thebetterwecan processes are, Themore efficientthese become. themore powerful we information, transform datainto thatis, it, to digintodataandmakesenseof PERONA foradvancement. tleneck bot- extraction process isacritical oftheinformation a centralpart mous andtheneedforpeopletobe quantities ofinformationare enor- the anddisease, ed withheredity ing tounderstandpatternsassociat- mation gathered byresearchers try- orgeneticinfor- impending storms, towarnpeopleabout Service bytheNational tracked Weather orweatherpatterns epidemics early, Control inanattempttoidentify lected bytheCenterforDisease it’s patientstatisticscol- Whether extracting meaningfrom data. in Humansare stillcritical process. automatedthe none haveentirely butsofar signal toinformation, question ofhow togofrom raw fieldshavelookedatthe Many ing. extract from that somecore mean- andyou wantto orrawsignal, data, Imagine thatyou haverawbitsof example ofanarea toinvestigate. Representation choice : The more weare able cmi is one tion. andextrac- mation gathering, infor- computation, scope of rethinking themeaningand of There is...afundamental need ol,andonthethirdside ofthe world, knowledgewe haveprior aboutthe onanotherside we havethedata, Ononeside tasks. toward certain theyare finalized self-contained, PERONA trivial. a setofspecifictaskscompletely representation—the onethatmakes you wanttofindthecorrect really For agivenproblem, tific thinking. toscien- critical sentation isreally Thisconceptofrepre- division easy. and multiplication, subtraction, and 1s—whichmakesaddition, 0s system—only that useabinary Now wehavedigitalcomputers moretasks. perform complicated it’s because handierto was adopted, why theArabicnumerationsystem That’s it’s acompletemess. system, to addtwonumbersintheRoman Ifyou try efficient forcalculating. itwasnotreally give itupbecause buttheyhadto numeration system, upwitha The Romanscame expressing numbers. ofsimply tory atthehis- Look tions allthetime. CANDES to ignore. ones—andwhich are theimportant understanding whichfeatures easily : : Humans userepresenta- Representations are not fiin a ooeae When efficient waytooperate. itisthemost because to thebrain, ments ofthesensationstransmitted frag- isms payattentiontoonly another related problem—organ- for theCenter. dinated—this isanotherbigtheme the samedatathatneedtobecoor- representationsmay beseveral of made more complexinthatthere Theproblem is either. cartesian sure thattheserepresentations are We don’t know for of theworld. representationssions ofgeometric makes atleasttwodifferent ver- Thebrain hands andmyeyes. finditbothwithmy that Ican coordinatesso resented inworld objecthastoberep- the an object, IfImove myhandtorub motion. withrespect tothat invariant want myrepresentation tobe move withrespect totheheadbutI myeyeshaveto because my body, object isinrespect tomyheador Iwanttoknow where the eyes, IfImove my sentations are useful. Allofthesedifferent repre- nates. andtheninbodycoordi- dinates, objects are expressed inheadcoor- Next the retinal coordinates. objects intheimageare assigned andthen bytheretina, captured create animagethatis Photons spacearound aperson. the physical brain’s different representations of here atCaltechare the studying mycolleagues For instance, Center. one ofthebigthemesfor This is be usedforagivenproblem. mine whichrepresentation should All three deter- isthetask. triangle Attention andawareness is winter 2003 information science and technology

confronted with practically infinite do feel that existing representations computations each one of them data, how do we know what to pay are somehow limited. There’s a needs to do separately. I care about attention to? How do we shift our whole world out there of new rep- how much communication between awareness? Several researchers in resentations that we would like to them is necessary to make the sys- the Computational and Neural explore systematically. Any major tem work. I care about how best to Systems option are dealing with the advances that we make will be use- use feedback. I care about represen- engineering issues behind aware- ful to other key players in the other tation choice. And I care about all ness and will play a big role in the ISTI centers. of these things simultaneously.We CMI. are now at a point where it is, I believe, critical to figure out how to EFFROS: Many people on campus put all of these pieces together. So are focused on representation EFFROS: What is the smallest in information theory, the tradi- choice. Some are concerned with amount of computation I can use to tional view has been to look at how vision, some with attention and perform a particular task? My own many bits it takes to communicate awareness questions, and we have field of communications or infor- or store information, but the com- computer science people thinking mation theory focuses primarily on putation resource has been consid- about representation choice for the the quantity of information, ered to be unlimited. You can have purpose of being able to do certain whether you measure that as band- as much computation and delay as kinds of computations. The CMI width or just as the number of bits you want, but feedback is going to will bring all these electrical engi- that you need to represent some be a problem. These other resources neers, computer scientists, and particular piece of information. were allowed to be unlimited so we applied mathematicians together to Controls researchers focus on feed- could see where the critical points tackle the foundations, the funda- back. To think about how these dif- were in the one resource on which mentals of representation choice ferent resources interact or trade off we focused. If you look at these independent of the realm of appli- other fields, they’ve done the same cation. There are people all over cam- kinds of things. However, pus who are thinking deeply researchers in each of these fields CANDES: I’d like to emphasize about the mathematics of are now realizing that we really the timeliness of the Center. It’s information. The goal in need to take all of the resources clear that scientists and engineers many senses is to bring them into account. are engaged in acquiring massive all together. Taking advantage of Caltech’s data sets—in many areas of biology, small size and cross-disciplinary bioengineering, and finance, many nature, we think that we can make people are involved in massive data is fascinating to me. If I’m working real progress in putting these things collection. It’s clear that any kind on a control system, say a distrib- together. In trying to understand, of progress we make in the area of uted control system where I have a for example: is there a dynamics of data representation will have a huge bunch of different devices all trying information? What would the impact across many sciences. And to work together to perform a par- dynamics of information look like? though we’re not the first ones to ticular task, I care about many Is there a conservation of informa- think about data representation, we things. I care about how many tion? What are the properties of

32 33 eng enious however, happen only ifweinvest a happenonly however, damental paradigmchangecan, fun- Such information processing. tation toredefine ourapproach to scientific research comesthesolici- From manyendpointsof extraction. and informationgathering, tation, the meaningandscopeofcompu- a fundamentalneedofrethinking infact, There is, manner. peripheral ina beanswered only cannot thatthesechallenges itisclear cific, advances are goingtobefield-spe- manyaspectsofthese While tion containedinmassivedatasets. novel toolstoprocess theinforma- research upontodevelop iscalled modest mission. would liketoformulate amore CANDES inasense? mation theory, reinvent thefundamentalsofinfor- ENGENIOUS the otherresources thatare critical. byabstractingawaymanyof beauty and resource withincredible clarity one Hecaptured information. ortostore through anoisychannel, eithertocommunicate takes,” lookathow many bitsit “Let’s said, Hejust strain theamountofdelay. computation andhedidnotcon- did notconstraintheamountof He madeassumptions. Shannon But Shannon’s work for50years. sciences? we haveinthephysical that Do theyparalleltheproperties resourcethis new ofinformation? Every singlefieldofscientific Every We havebeenbuildingon : fyualwm,I If you allow me, cmi : Could thisCenter need toexist... the conceptthatdisciplines disciplinesandeven between destroy alltheboundaries where...this istheplace we things can happen. things can where these a homeifyou will, Center willcreate anenvironment, our Inshort, around information. thinkingcenteredtheoretical considerable amountofresources in o-onve n oiehw in top-down andnoticehow, view PERONA unified whole. ture thatputsitalltogetherinto a We’re missingthebigpic- realms. give usfocusedpictures incertain piecesofthepuzzlethatonly a few Butit’s asifwehave that are there. withthepieces something wrong EFFROS area. emerging interdisciplinary generation ofscientistsinthis the Centerwillhelptrainanew Andthird, information technology. scienceandkeyareasmatical in betweenmathe- create bridges new strengthen existinginteractionsand theCenterwill Second, actions. toolsviatheseinter- mathematical andnew theories mathematical andtocreate new mathematics; challengesinto new to import ininformationtechnology; rithms andalgo- theories, ideas, matical todeploy mathe- the opportunity is,theCenterwillcreate First, : : It’s notthatthere’s You couldtakeamore eino ope,robust systems design ofcomplex, the So will stop. I thinktheworld the internetgoesdown foraweek, If tolive. a tree hangingwithfruit and the waterwellandchicken needing far pastthepointofsimply We are depends onthesesystems. Humanity to keepworking. Theyhave andhuman errors. sorts, hardware problems ofall ware bugs, soft- withtheinevitable gracefully andstillbeable to deal cations, communi- include control, include computation, thatinclude systems distributed complex, about large, ideas thatwillallow ustothink coming upwithkeymathematical accordingtoplan. and willperform glitches orthattheywillbestable andsoftware be robust toviruses guarantee thattheywill we cannot to designtheseintegratedsystems; wedon’t know how Unfortunately, was designed30yearsago. that withsoftware stocked pletely they’re com- However, complex. areThese systems increasingly ning ofubiquitousnetworking. And thisisjustthebegin- tions. telecommunica- biles thatinclude andautomo- PDAs andcomputers; have telephonesetsthatbecome integrated andconnectedsoyou are things being Nowadays, lives. Theyhavechangedour the goods. things are welldesignedanddeliver Allofthese airplanes. automobiles, personalcomputers, telephones, Think effective atdoingonething. thatwereered extremely systems deliv- technology the pastcentury, A bigthemeinthisCenteris winter 2003 information science and technology

will be another important research continuous, multi-scale, dynamic, bursting out of the seams of exist- area for the CMI. To do this, scien- and complex. ing fields. And this Center is going tists from different disciplines will to capture them. We hope to attract have to come together, transcend PERONA: We hope the Center the best talent in the country, both their respective disciplines, and will bring the pure mathematicians at the level of graduate students broaden the scope of their research. at Caltech in contact with the tech- and at the level of young faculty. nologists. We will be working very They will want to come to Caltech CANDES: Absolutely, and at the closely with the theorists in the because this is the place where we same time we want to rethink com- physics center [CPI] as well. destroy all the boundaries between putation, particularly large-scale disciplines and even the concept computation. A trivial answer to EFFROS: Making that connection that the disciplines need to exist— the large-scale problems is: give me between pure mathematics and we’re focusing on the real problems more flops. Here is an area where applied mathematics is critical. You of today. mathematics could play a role by would be amazed how broadly our providing a more efficient data theme sweeps. There are people in structure through more efficient economics, humanities, and social Emmanuel Candes is Assistant Professor of Applied and Computa- representations of operators for cal- sciences who are worrying about tional Mathematics. Michelle Effros culation. the mathematics of information. is Associate Professor of Electrical There’s another very interest- There are people all over campus Engineering. Pietro Perona is Professor of Electrical Engineering ing avenue that we will explore— who are thinking deeply about the and the Director of the National while the world we live in is con- mathematics of information. The Science Foundation Center for tinuous, and we have the laws of goal in many senses is to bring Neuromorphic Systems Engineering physics formulated in a continuous them all together. at Caltech. way, computers are only able to handle equations and sets of data CANDES: I’d also like to empha- that are discrete and digitized. So if size that the CMI will provide a you’re looking at numerical real link to and between the other schemes, or if you digitize an equa- ISTI centers. ISTI will bring the tion, you have violated a lot of divisions of Caltech together in physical conservation laws that profound ways, and this particular nature prefers to be preserved. How Center will be the glue for ISTI. can you think really discrete all the way through without violating PERONA: At the beginning, creat- physical laws in your end results? ing this Center felt like a construc- That’s a topic people will gravitate tion. But now it feels like an around, and that scientists at inevitable fact. It seems impossible Caltech have already started attack- not to have thought about it a little ing. Squarely addressing this chal- bit earlier and it seems impossible lenge will be critical for moving that it will not exist. I see signs, all beyond this limited, digitized com- over the country, that the best, putational view, to one that takes young creative people in every area into account that the real world is that deals with information are just

34 35 idea flow

Sensing and Responding: Mani Chandy’s Biologically Inspired Approach to Crisis Management

rganisms receive streams responding to the crises. This idea O of information from their was the seed that led to iSphere’s senses, yet they manage eventual products: decision-making to avoid information systems for financial and trading overload and system breakdown by institutions, manufacturing con- instantaneously aggregating infor- cerns, and corporations requiring a mation to identify threats and “sense-and-response” approach to opportunities, and responding tremendous amounts of unpre- appropriately. Organisms are dictable data in all sorts of formats “sense-and-respond” systems that and configurations. The software detect and respond to important had not only to duplicate the events in their environments. “sense-and-response” functions that Abstracting this relationship of humans bring to decision-making, organisms to their environment and but also to amplify human capabili- turning it into a computer-based ties, in both the amount and types information system was the chal- of data sensed, and the time it takes lenge that Mani Chandy, Simon to respond. Ramo Professor of Computer Chandy began working on this Science, set for himself almost a biologically inspired, inverse data- decade ago. base problem in 1992/93, but in What he came up with could 1998 had the key breakthrough of be thought of as the inverse of a “sense and response” as the organ- traditional database: rather than Organisms that respond to non- izing principle behind his concep- repeatedly accessing well-defined events waste energy. Organisms tions. As given his nature, Chandy static data structures with different that don’t detect threats and oppor- put all his results on his web site, queries, imagine data that constant- tunities in the environment don’t including downloadable software. ly change, have no well-defined survive. One day, Caltech’s Office of structure, but questions about the In 1998, Chandy took an aca- Technology Transfer called him data that are more or less constant. demic leave from Caltech to start and suggested he take everything You can liken this to an organism’s iSpheres, a company devoted to down, as they were getting inquires day-to-day stance to its environ- creating information systems for from companies who wanted to ment—organisms get continuous crises management. In the after- license the ideas! This phone call streams of data, with varying struc- math of the TWA 800 disaster, led to the germination of iSpheres. tures and formats from their senses, Chandy was inspired to develop a iSpheres faced several chal- but the conditions that define system so that all kinds of informa- lenges. Sense-and-respond systems threats and opportunities change tion (weather reports, emails, engi- have to manage large volumes of slowly. Occasionally there are neering data, police and fire depart- heterogeneous data ranging from “events”—sudden, dramatic, some- ment responses, and so on) could stock ticks at 10,000 per second to times catastrophic happenings that be accessible and useful for rescue news stories and email. These dif- require immediate response. workers and later, investigators, ferent streams of data have to be

engenious winter 2003 integrated with information in only by other academics. It was also spaces, giving away systems freely is databases and business-intelligence a revelation to pitch ideas to more effective than building commer- warehouses to detect threats and investors—the process of getting cial products. He’s come full circle, opportunities such as arbitrage. your ideas funded in the venture- but in the process created and dis- Then the system must respond capital marketplace was quite dif- seminated ideas far and wide. Other appropriately to events, and appro- ferent than getting your ideas fund- application areas he is working on priate response often requires ed by granting agencies, and more include systems that would assimilate orchestration of services within and satisfying in many ways. the vast amounts of heterogeneous outside the organization. The fun- Chandy learned how industry biological information available to damental problem is to build sense- works, how essential effective lead- detect significant opportunities, as and-respond platforms, layered on ership is, and that a company is not well as systems that would allow one top of information technologies simply a technology. The culture of to study new types of financial mar- such as databases and application the company is crucial, and he kets (see article on Information servers, which can be configured to strove to make iSpheres a nurturing Science and Technology, page 18). deal with different applications. environment for employees, an He is eager to become engaged again Though the applications are usually environment where they could grow with the deeper mathematical issues complex, the interfaces for config- and evolve. However, the discipline of the work. If he were younger, he uring the platform are relatively of the marketplace was the constant hinted that perhaps working in start- simple: after a few weeks of train- and final arbiter of what the com- ups periodically, or even on a fulltime ing, clients can configure the soft- pany could (or could not) do. This basis, would be a tempting direction; ware themselves, for their particular fiscal discipline was brutal, especial- but now, with his start-up company applications. ly concerning hiring and firing, and out of its infancy and on its way to An exciting part of starting a this was perhaps the most difficult childhood, he is very satisfied to company, Chandy explained, was part of starting a company. delve back into the challenging theo- talking to users in different indus- retical concerns at Caltech. tries ranging from energy and stock trading to supply chain he original idea of crisis- management, and solving their T management systems that There is more on Professor Chandy at concrete problems directly. would assist rescue work- http://www.infospheres.caltech.edu Creating a product that affects ers and investigators people’s lives was exhilarating and evolved at the company into deci- demanding because the product is sion-making systems for financial used in mission-critical applica- institutions and manufacturing con- tions. In the academy, the “metrics cerns (the investors realized that for success are fuzzier,” the prob- the core product had to be revenue lems solved are removed from con- generating). So, Chandy is back in crete requirements in different his Caltech lab figuring out ways to industries, the pressure to make sys- improve the software for crisis- tems work 24/7 is less intense, and management applications. He feels sometimes solutions are appreciated that for some (but not all) problem

36 37 research note

Orbits of Light: Kerry Vahala and His Miniature Lasing Spheres

onsider a tiny glass sphere ible by the addition of a tracer ele- face tension. It therefore exhibits a C or spherical droplet. ment within the sphere that enables degree of surface perfection very What would happen if up-conversion to the visible green difficult to match by other means. light could be launched at band. Surface blemishes and roughness near-grazing incidence along its Optical whispering galleries tend to randomly scatter light from interior wall? This would yield the can be made in several geometries. whispering-gallery orbits and there- optical equivalent of by degrade light the familiar acoustical storage time. whispering gallery. The lifetime of In optical whis- the mode as given pering galleries, solu- by its quality factor, tion of Maxwell’s or Q value, is an equations shows that easy way to measure light can be guided the superior per- along trajectories that formance of spheri- are tightly confined cal micro-cavities in near the surface of the comparison to sphere. Because of semiconductor- spherical symmetry, processed micro- these whispering- cavities. Q values gallery solutions (also for silica micro- known as modes) cor- spheres formed as respond with solutions molten droplets can to the classic hydrogen exceed 1 billion, system of atomic while the record for physics. If the sphere a lithographically is large—compared to processed structure the scale of the wave- is nearly 5 orders of length of light—we magnitude lower. can expect to be able This difference has to interpret the whis- made droplets and pering-gallery motion as an orbit in In addition to spheres or droplets, spheres an object of interest for the sense of an approximate ray- it is possible to fabricate disks, some time in a number of different optics picture. rings, and racetrack geometries fields. Light trapped within a glass using combinations of lithography In a ray-optics picture, a Q sphere as a ring orbit is shown on and etching processes similar to value this high means that light the opposite page. Although the those used in the semiconductor inside a sphere about 30 microns in sphere shown here has a diameter industry. diameter will trace out orbits up to less than the thickness of this page, However, what makes droplets a million times before leaving the it is nonetheless large on the scale (or spheres formed first as droplets) cavity. Returning to the acoustic of the wavelength of light, and the special is the near atomic perfection analogy, a true whispering gallery mode in this case clearly meets our of their surface finish. Unlike litho- with an equivalent Q value of 100 notion of an orbit. One comment is graphed or etched whispering gal- million could resonate or “ring” for in order: the “light” that has been leries, which are considerably over an hour. trapped in this sphere is actually in rougher, a sphere’s shape is deter- Introducing or “coupling” light the infrared, but has been made vis- mined in the molten state by sur- into the high-Q modes of a spheri-

engenious winter 2003 cal whispering gallery is non-trivial. greater the energy stored. For in which glass actually amplifies Efficient coupling requires first that spheres with diameters in the range certain wavelengths, rather than the orbiting wave’s phase velocity of 30–40 microns, optical power becoming weakly lossy. With this be matched with the input wave, circulates within a ring volume of optical amplification in a cavity, the something not possible from free about 10-15 meter3. glass whispering gallery emits new space. Then, through a process Putting all of this together, a laser frequencies back into the same called directional coupling, it is fiber taper providing incident power taper used to couple the “pump” possible to excite whispering- in the range of 1 milli-watt, cou- wave. Other startling effects are gallery orbits using waveguides of pled to a sphere with Q of 100 mil- observed associated with low-fre- similar (but not necessarily identi- lion, will induce an intensity quency phonons of the glass bead, cal) cross section and refractive buildup within the ring orbit in as well as nonlinear mixing. index. excess of 1 giga-watt/cm2. At these Vahala’s group continues to Remarkably, these waveguides intensity levels, normal glass—one research properties of this and other can be fashioned from optical fiber whispering-gallery-based devices. filaments in the form of a narrow The Raman laser described above, taper. This is of considerable practi- in addition to providing a window cal importance. Optical power, ini- on nonlinear cavity physics, may be tially guided within the interior of a of practical importance as a com- fiber cable, can be converted by the pact, ultra-efficient wavelength tapers into waves guided along source. Vahala and Applied Physics micron-wide filaments, and then graduate students Sean Spillance back again. and Tobias Kippenberg reported on Some time ago the Vahala this device in Nature, February 7, group demonstrated that such 2002. The device set a record for tapers could be prepared in such a threshold power (the power neces- way that coupling both to and from Micrograph showing a 40-micron diameter sary to induce laser oscillation) of orbital modes is exceedingly effi- silica microsphere that is doped with the rare only 60 micro-watts. With further earth erbium. Upon incorporation into silica, cient. This process “links” the erbium ionizes to the 3+ state and exhibits improvements in Q underway, it spherical whispering-gallery system dipole transitions in the green and near should be possible to lower this to the technologically important infrared. In the micrograph one of these tran- value to mere nano-watts. world of fiber-optic communica- sitions has been excited by optical pumping tions. through a fiber taper. The taper can be seen in the micrograph as the slightly out-of-focus horizontal line. The green ring emission from Professor Kerry Vahala (BS ’80, MS the sphere corresponds to a fundamental ’81, PhD ’85) is the first occupant of number of new devices whispering-gallery mode of the sphere. This the Ted and Ginger Jenkins A have demonstrated the particular sphere is also lasing in the 1.5 Professorship in Information micron band (the important telecom band). capabilities of these sys- Science and Technology. Ted Jenkins The lasing emission is efficiently coupled tems: for example, a laser onto the same fiber taper used for optical (MS ’66) and his wife, Ginger, estab- that uses only the glass itself as the pumping. lished the professorship in early 2002. lasing medium. To understand how this device functions, note that of the most linear of optical ultra-high Q in a tiny package can media—exhibits properties out of More on Professor Vahala’s work at concentrate optical energy in a tiny the range of our normal experience. http://www.aph.caltech.edu/peo ple/vahala_k.html volume. Imagine a ring orbit excit- Optical propagation can no longer ed by an optical fiber taper. be understood using linear optics, Consider the buildup of power as the molecular motion of the glass within the ring volume. When the becomes highly distorted and gives taper and the sphere are coupled rise to new optical frequencies and appropriately, energy will store with behavior. a time constant given by the cavity One manifestation of this tran- Q such that the higher the Q , the sition is Raman emission, a process

38 39 alumni profiles

Ivett Leyva: An Experimentalist with International Flair Aeronautics, PhD ’99

With this issue we are beginning our practice of offering two alumni profiles—one of a “neophyte” (an alum who has recently graduated), complemented by a second profile of an established Caltech “ex-pat.”

vett Leyva gradu- publish and present PDE I ated from Caltech work at several conferences. in 1999 after And in 2002, she had six spending, she patents filed. declares, “seven great years” Leyva is also involved in in residence, first as an design and testing of next- undergraduate (transferring generation domestic gas from Whitman College), burners. “I am the liaison then as a graduate student between the manufacturing working with Professor facility in Mexico (where I Hans Hornung. Her PhD can practice my native lan- was in Aeronautics, her the- guage) and our research sis on the shock detachment facility here. What I like process on cones in hyper- most about this project is velocity flows. my exposure to this very Upon graduation, she short business cycle, very left southern California for different from that of air- upstate New York, joining craft engines. It is also grati- the General Electric Global fying to see the very funda- Research Center as a mechanical cal-reaction zone). “I am involved mental research we do get applied engineer. She has been involved in in the conception of ideas, transfor- to such familiar products as domes- a wide spectrum of technologies in mation of ideas to manufacturing tic gas burners.” her first three years at GE, includ- drawings—including the minutiae “One of the things Caltech ing cycle analysis of microturbines, involved with making an idea easily best prepared me to do is be a very experimental testing of fuel cells, manufacturable—and testing careful planner of my experiments,” and currently, design of domestic resulting prototypes.” The final step Leyva observes. “From my advisor gas burners and pulse detonation is analyzing results, then presenting [Hornung] I also learned the power engines (PDEs). and discussing them with program of back-of-the-envelope calcula- “Working on PDEs is managers and VPs. “I have been tions and the great value of doing absolutely fascinating,” Leyva very fortunate to travel twice to CFD [computational fluid dynam- explains. “They promise to be a Russia and work very closely with ics] and lab experiments hand-in- crucial step in the ever-harder fight Russian researchers. I have created hand to strengthen and best use the for higher cycle efficiency for air- joint programs with them, negoti- results of both. Professor Paul craft engines.” In a PDE, energy ated the scope and schedule of Dimotakis taught me that a good from the fuel/air mixture is released projects, and made sure that the experimenter really knows all the through a detonation (a supersonic schedule of deliverables was met.” ins and outs of her experiment, and shock wave coupled with a chemi- Leyva has had opportunities to I try to abide by that philosophy.”

engenious winter 2003 She also fondly remembers her had had more experience with my life. I’m grateful to the GAL- friends on campus, “who made me a while at Caltech is more exposure CIT community who made me very happy student.” to the practical considerations of feel like a family member. I hope “At GE I have learned to merge manufacturing, such as making suc- that through my work and citizen- analytical and academic knowledge I cessful and safe aircraft engines. I ship I make them proud.” gained at Caltech with more practi- have had to learn many of these cal and experience-based knowledge things as I go.” gained through my first few years Leyva feels the years she spent here. Perhaps the only thing I wish I at Caltech “are some of the best in

Eric Garen: Education at the Fore Electrical Engineering, BS ’68

What are the pivotal experiences that shape a person’s life, that lead him or her down one path rather than another? We spoke to Eric Garen in his Los Angeles home about these experiences, about his Caltech education, about the formation of his company, Learning Tree International, and about his current projects. What emerged is a picture of someone who has successfully applied a rigorous, analytical approach to problem solving, whether it be of complex business problems, or of social problems that plague inner-city youth trying to make their way to college.

We begin in the early 1970s, on the eve of the advent of the personal computer. Intel was manufac- turing their early microprocessors (the 4004 and the 8008), and engineers were struggling with how to use these new devices. Eric Garen was one of those engineers.

fter graduating from Caltech that would train other engineers like A in 1968, I went to work at ourselves on new technology. In 1974 Technology Service we formed Learning Tree International. Corporation, a small think We went into business in Dave’s tank in Santa Monica that was an off- spare bedroom. We used his garage to shoot of Rand. After a few years, I store our course materials. We were an began to incorporate minicomputers upscale start-up—we had a bedroom in and then microprocessors in the real- addition to the traditional garage! We time radar simulators we were design- put 20,000 or so flyers describing our ing and building. But learning how to first microprocessor course into the use the early minicomputers and Intel’s mail and sure enough, people started first microprocessors was basically a sending us enrollment forms and trial-and-error process. You made a lot checks. Initially I was the course devel- of mistakes and did things the wrong oper and instructor, and Dave was the way. It became clear that that wasn’t operations department and marketing the best way to learn. So I joined with department. We packed boxes with our fellow engineer and Stanford graduate course materials (and a few stray Dr. David Collins to form a company autumn leaves) out in the driveway, and

40 41 sent them off to course sites. A ized, despite the typical Techer’s ing procedures in place that ensure month after running our first desire in the mid-’60s, mine consistent results. And then having course in Los Angeles, we were included, to appear sort of “laid a “meta-procedure” for reviewing running courses on the East Coast back.” I left Caltech with the abili- and improving our procedures on and a month after that, in London ty to apply an analytical, quantita- an on-going basis, so that over time and Paris. tive approach to problems and to the procedures, and the results, get make data-based decisions. Because better and better. Learning Tree International both Dave [Collins] and I are ana- It’s really exciting to figure out offered courses on a global lytical, it’s not surprising that our how you take a seemingly amor- basis right from the start, and company is highly data driven. We phous field like teaching advanced as a result, half of their busi- have built systems throughout our technology, and turn it into a logi- ness now is in the U.S. and organization for collecting and ana- cal, coherent, structured process half is outside, primarily in lyzing data. Early on, we realized that ensures that results are consis- Europe and Canada. They that we had to start “procedurizing” tently achieved. Every course par- also have offices in Tokyo and things, “systematizing” things, if we ticipant evaluates our courses and Hong Kong. were to grow the company to any our instructors, and each year our size. Most important were the pro- average instructor GPA gets just a Our business concept was to offer courses in new technologies as they were being introduced. Microprocessors formed the first technological wave that propelled our business forward throughout the 1970s. In the 1980s, the net- working wave moved us forward, and created needs for training in distributed computing, UNIX, C and data communications. Then in the 1990s, the client-server wave propelled us beyond the engineer- ing departments we were serving and into our customers’ informa- tion systems groups. And now the Internet wave is pushing us forward again. Today we offer about 150 different courses and have trained over one million IT professionals around the world.

The impact of his Caltech education on his subsequent cedures we developed to ensure the little bit higher. Today, it’s running endeavors was pervasive, but quality of our training, because just over 3.82. We still have some not in the traditional sense of ultimately that’s what drives our room before reaching 4.0, but we’re applying the specifics of his growth. After taking our courses, edging ever closer. electrical engineering back- our participants return to work and ground to his work. succeed in their projects because In 1956, when Garen’s father, they gain the skills they need. So a chemical engineer, took a My Caltech education provid- how do we ensure every attendee at job in the new rocket industry ed me with good organizational every course succeeds when we’re (at Aerojet General), the fami- skills and taught me how to learn. running 8,000 courses a year in 30 ly relocated from Greenbelt, You can’t get through Caltech countries around the world? The Maryland to Sacramento, without being reasonably organ- only way we can do that is by hav- California. A few years later,

engenious winter 2003 alumni profiles

the young Garen found him- column of the printout and discov- reunion each year with our families. self attending Folsom High ering that a 1 in the first column It’s amazing how much satisfaction School, just down the road served as a control character that we get by sharing some wonderful from the Folsom Prison that caused the page to eject. Our print- experiences together each year. Johnny Cash made famous. In out was about a foot and half high, those years, the education with one row of data per page! In 2000, Garen and his wife there was rather fundamen- Eventually we got our program established two scholarships tal… working. I learned a lot of FOR- at Caltech specifically for stu- TRAN in the process, and found dents from very low-income When I got to Caltech, I expe- myself hooked on computer tech- families. In the same vein, rienced a rude awakening because I nology. That experience was piv- they have turned their ener- had no calculus or advanced science otal. I declared an EE major—and gies recently to two communi- classes in high school. Dr. [Rochus] realized that trial and error is really ty programs: One Voice and Vogt did a terrific job teaching a terrible way to introduce people Bright Prospect. frosh physics that year using the to computers. Feynman books. His first lec- Several years ago, we began to ture with air troughs just blew provide support for One Voice, a my mind. It was exciting, but y second pivotal experi- grass-roots community service the pace of the lecture, the M ence occurred in my sen- organization in Santa Monica, course, and my entire freshman ior year. One of my California. One of their programs year were staggering. Dabney House friends, identifies high-performing high- I hadn’t made up my mind Charles Zeller [BS ’68], had mar- school juniors from financially dis- whether to go into biology or engi- ried the year before. His wife advantaged living situations—gen- neering. But in freshman physics attended Pasadena City College erally inner-city kids who have we were given a problem and told and was taking a modern dance proven they are capable of succeed- to solve it by writing a computer class with a young woman named ing at top-ranked colleges, but who program for the largest computer Nancy Graeber. The Zellers intro- are unlikely to get there without on campus, an IBM 7090—most of duced us in the fall of 1967, a week counseling, support, and complete us had never seen a computer later we went to a Grateful Dead financial aid. These kids come from before, much less used one. They concert, and we’ve been together high-risk environments, but they gave us a thin FORTRAN manual since. So you can see I came out of are not at-risk youths. These are and said “Go.” So there we were, Caltech with much more than just young people who have overcome trying to figure out how to com- an engineering education! huge obstacles and done very well pute the trajectory of a rocket trav- I think that for me, the great- in high school through their own eling from the earth to the moon, est thing about Caltech is that it’s a talent and determination. and not getting our program, or our concentrated environment where One Voice counsels them, pre- rocket, off the ground. My partner you establish lifelong bonds with pares them for their SAT tests, and I had to teach ourselves FOR- people who have similar interests guides them on their college essays, TRAN, making one mistake after and similar analytical capabilities. gets recruiters from top universities another. It was an experience that It’s a phenomenal environment and to interview them, helps them prepared me for my similar attracts phenomenal people. decide on a list of schools, and encounter with the Intel 4004 My third pivotal Caltech expe- structures their application process. microprocessor a few years later. rience was meeting the guys—yes, As a result, every student in the After first fighting our way through it was still all guys in those years— program gets admitted to top seemingly endless syntax errors, we who have become my friends for schools, and receives full tuition encountered our first programming life. In fact, for the past 11 years, and room-and-board packages from mistake—putting data into the first seven or eight of us have had a them. One Voice then provides

42 43 supplemental funds for airfare, clo- process that’s successful, you need Bright Prospect, he also thing, books, and living expenses. to document it, replicate it, and touched briefly on family life This whole area has troubled only slowly make incremental as our interview drew to a me for a long time: we have in our changes to improve it. close. He and Nancy have two very affluent society a significant Last spring we identified our children, a daughter, Nicole, fraction of our population that is first group of 12 kids. This past and a son, Steven. He noted economically disenfranchised. And fall, recruiters from 30 top colleges (with a smile) that taking his that gap, if anything, seems to be visited Bright Prospect to meet our daughter to college was just widening. That cannot be a stable students, and soon we’ll know not the same as when he situation and we need to do some- where these students will be going went off to Caltech. thing about it. So this seemed to be to college. a small step in the right direc- Taking our daughter to college tion of helping to create a path was a whole different experience out of that environment for ther programs that have than I remember from arriving at kids who at least have the O attempted to help stu- Caltech. Nicole entered gumption to go that path. dents from similar situa- Washington University in St. Louis When these kids succeed, they tions often experience a last August as a double major in inspire more and more kids to fol- 50% or greater drop-out rate. This pre-med and fine arts. Helping her low. And who knows, at some is not because the students don’t move in, I felt like a rock band point, it may actually start to steer have the academic or intellectual roadie. We practically needed a bus the direction of the boat differently capability. It’s simply a culture and four semis to get everything to than it’s going now. shock. They’re being plopped down her dorm room... well, I suppose The One Voice program has in an environment that’s alien and that’s a bit of an exaggeration, but been extraordinarily successful— that they don’t feel a part of. So the Nicole has a little refrigerator. A they have had over 120 students in support organization must stay microwave. Computers, printers… their program and only one has with the students: by e-mail, by I just showed up in Pasadena with dropped out. And of the kids who telephone, by personal visits, by one suitcase and a manual type- have graduated, close to 40% have intervening, by calling up the dean writer. gone on to graduate school. They if necessary. Our son, Steven, is a junior at have students doing graduate work Bright Prospect’s goal is to Harvard Westlake High School and at MIT. They have students in replicate the One Voice program like his sister, he’s good at both art medical school at Stanford. They’ve successfully, and to raise sufficient and academics. Steven plays guitar graduated their first lawyer who funding to make the program self- in a band—they’re quite good and passed his bar exam on the first sustaining. That will free up our play at the Roxy and the Whiskey shot. seed capital and allow us to go out on Sunset Boulevard. But they Nancy and I sat down with the and replicate the program in other change their name so often I am directors from One Voice a few locations, either by opening more not sure what they are called this years ago and said, we’re helping Bright Prospect offices or by find- week. Maybe my roadie experience 20, 25 kids a year. But is it ing other community service organ- will come in handy again one day scaleable? Could this be 200 kids or izations that want to add this pro- when they go on tour. 2,000 kids a year? That encounter gram to their activities. That’s the A terrific wife who I met at led us to incorporate a new non- vision. In ten years, we would like Caltech. Great kids. Lifelong profit organization located in to have at least 1,000 students in Caltech friends. Applying what I Pomona [California] called Bright our program, 200 students at each learned at Caltech to make a differ- Prospect Scholar Support Program grade level. That’s a modest goal, ence in peoples’ lives. What more whose mission is to replicate the but one we are determined to could one hope for from a Caltech One Voice program, and then achieve. And we hope to make it a education? spread it to other communities. lot bigger than that. Bright Prospect has implemented exactly the processes that One While Garen spoke in detail Voice uses, because in business I’ve about the formation of his learned that when you find a company and the creation of

engenious winter 2003 The mission of the California Institute of Technology is to expand human knowledge and benefit society through research integrated with education.We investigate the most chal- lenging, fundamental problems in science and technology in a singularly collegial, inter- disciplinary atmosphere, while educating outstanding students to become creative mem- bers of society. division of engineering and applied science California Institute of Technology 1200 E. California Boulevard, Mail Code 104-44, Pasadena, CA 91125 www.eas.caltech.edu