COMING OF AGE WITH THE
TRANSISTOR

BY SANDIP TIWARI

The transistor and its usage, semiconductors in general following in its wake, have been one of the greatest technological achievements of the human age. They are a social force through their daily use in computing, communications, transactions and in health. In this 75th anniversary celebration participation, I share here some of the lessons learned while coming of age with the transistor. The new and young entrants to the technological spectrum might find these vignettes amusing and perhaps useful.

Much has been written about the history and the future of the transistor. By the simple means of converting static energy into signal energy, it becomes a physical tool for logical transformations and communication, and even when not amplifying, it is an essential feedback tool for stability. It morphs into many forms and when it is together with other structures, it expands to new functions. With floating charge storage within it, it is a quasi-non-volatile memory, with a capacitor, it forms a fast and dense dynamic memory, and when paired with other transistors, it becomes a very fast static memory. The static random access memory even has the unusual fundamental characteristic of being self-aware. It stays in its state and a similar cross-coupled element is needed to change it or probe its state. The modern age is wrought by the techno-social changes from the invention, its evolution, and the new directions it has spun off: in pursuits related to intelligence/learning or physical well being that are unfolding, and many more that we cannot yet see.

Many have written about these transformations.

Speculation, imagining, and interactions of ideas are critical food of human evolution. Having been born in the decade following the Bell Laboratories demonstration and having grown with it, with a tinny-sounding seven-transistor radio made at the ripe age of 15 without access to a spectrum analyzer or an oscilloscope as the introduction, an education when learning transistors and their circuits were the rights of passage, and following the completion of the formal education, in IBM’s research, and in academe, the transistor age has paralleled mine.

Instead of speculating specifics, it is useful to share some of the lessons of the journey drawn on this sampling of the different worlds. This article is a collection of musings of the physical and social learning from this transistor journey, what it speaks to for the future, and for the judgments and choices that a young person must make in undertaking a journey in this technical world.

Choose a direction while it is nascent: Engineers are creators of what does not exist. Engines of all types, water wheel, steam, electric, or combustion including those of the jets for transformation to mechanical form, all have had a fair period of research and early development of exponential growth and then a maturing linear long-life period before being substituted by a new invention. These Kondratiev cycles have invention-dependent time lengths. Electronics or more broadly optoelectronics as outgrowth of information energy inventions (the transistor, the solar cell, the semiconductor laser, and others) are no exception. For intellectual excitement over a period, the initial part is most interesting. The later part is interesting for the practically inclined, it is diffusive, and new unusual paths are its surprise pleasures. Electric vehicles are bringing a new period of vitality to old technologies of electric engines and batteries. Cars are now computing, sensing and communication engines on wheels. Autonomous motion is the semiconductor technology diffusing and being used in new avenues through the opening up of artificial intelligence, fast communications and multi-modal sensing platforms by the growth of integration.

Let curiosity drive the agenda: When one has solved a specific problem, it often ends with suggestions of other problems. Not always, and then one is at a loss for direction. The former becomes tiresome after a few years. While researching, many questions pop up, and one sets them aside. Letting them simmer and then froth up—a curiosity-driven pursuit—works wonders to setting up new satisfying directional agendas. While starting with compound semiconductor transistors, and their analog and digital circuits, it was not long before what happens in off-equilibrium conditions, or at nanoscale, to quantum and single electron and optoelectronic structures, and then transforming to issues of large scale: three-dimensional, adaptive, probabilistic, Bayesian, to neural became very fruitful venues. So, it is important to not let imaginary boundaries corral one. Developing the tools, carefully choosing available tools, and learning to shift keeps the development of the cocktail of skills alive and life interesting.

Research in the real world needs to be surreptitious; timing and cleverness matters: The industrial and academic research environment is fluid in time. Bell Laboratories or IBM Research do not exist in the fervent form that they had for the forty plus years following the invention of the transistor. The idea of using semiconductors to replace vacuum tubes predates World War II, but the subject was kept in abeyance till the war effort ended. Industrial research is for the benefit of the company, but the research itself benefits from the sharing of learning, so there is an implicit tension of publication, research, and commercial pursuit. Timing matters, sometimes the industrial inventions are early and take a long time before use. Semiconductor lasers’ pervasive use is one example. With transistors, we have CMOS. It appeared in Casio watches in the early 1970s making watches a few dollars a pop with the liquid crystal display. CMOS did not enter the mainstream of computing till the mid-80s. Cleverness matters. During that era, the IBM fabrication technology was three generations ahead of DEC (Digital Equipment Corporation), but the end products, where they competed, were similar in performance because of the higher-level design. DEC’s pdp and alpha computers became the platforms for creation of UNIX and made it into university laboratories for training and usage by the next generation of technologists. When a company becomes large, it is reluctant to undercut a precious source of profit that is slowly being nibbled away by a new technology. To state that the industrial world is real is to imply that the academic world is not. Far from it. The same competition and pursuit of ideas, which requires funding, which supports the students, exists. If anything, there exists much close-to-the-chest-protectiveness, symbolism and eccentricity—some quite contrived—with numerous antinomies.

Growth sneaks up: As a subject area develops momentum, whether in the industry or in the academy, it is adiabatic enough for disconnections to develop even if an approach was working before. In industry it usually arises from do-not-rock-the-cash-cow syndrome or because the leadership has lost an understanding of the needs. In academia it is usually because there is a natural bias towards the smallest unit that one feels one belongs to. The new directions that a subject can be leading towards can be lost. The interesting problems are always at the boundaries. That is where interesting diffusion takes place from the developments of a subject area. Openness to ideas and directions, where one can explore new creations, is vital once one is past the basic research period. This requires making sure that the culture has plasticity and is evolving in resonance with the participants and the subject. This type of diversity came to the transistor world through the competition and the proliferation of start-up companies throughout history. Success of foundries, another example of growth sneaking up, has been incredibly useful to this later period’s success since it has helped undercut the monoculture and given a lower cost possibility for implementing ideas in what is a capital-intensive semiconductor world.

With this broader context, some important themes related to the transistor and the coming period specifically.

Predictions are dangerous: As a graduate student in the late 1970s, a required reading were the technical papers by Hoeneisen and Mead [1] with 0.4 μm as the limit. There have been many other predictions of limits, by Meindl [2] of 0.20 μm, and even one by me [3] of about 10 nm. All these limits have been passed. They were not fundamental. They appeared since one constraint above all—a Mathew’s principle fallacy—was given credence. The nature of research and development is that one has not accounted for the unknowns. An engineer looks at the limits as a challenge.

Learning from false forays is useful: Challenges exist to be overcome. Choices are made in the midst of incompleteness. A specific property looks appealing, yet devices and the systems are an assemblage of properties that need to come out just right. Examples of such forays have included attempts at using compound semiconductors for large-scale digital usage in early 1980s, which gave me my early career start. The difficulties of surfaces, the absence of good contacting approaches, voltage limits, absence of good circuits because of the limitations of devices, etc., all showed up in limiting what heterostructure field-effect or bipolar transistor could do digitally. The devices are still enormously useful for their high frequency and high speed, but only at limited integration scale and with difficulties of reproducibility. Condensed matter physics regularly springs intriguing and unusual properties in new or rediscovered systems. Inducing superconductivity or selective spin in semiconductors, nanotubes, graphene, ferroelectricity, metal-insulator transitions, exploiting single electron effects in logic and for memory have appeared in this world line. Each taught something. Superconductivity the issue of contacting interfaces and limiting voltages; spin the importance of selectivity, relaxation and signal-mode conversion; monolayers and nanotubes the problem of surfaces, contacts and the complete absence of reproducibility; ferroelectricity the issue of imprinting and reliability in the absence of a single-crystal medium; metal-insulator transitions of interfaces and contacts; and single-electron effects the impact of statistics of low numbers. Some of these constraints were clear at the very beginning, nevertheless understanding the behavior was important. The learning led to limited but powerful uses elsewhere exploiting a specific property, and was a learning to be kept in the brain’s archive for later reuse. They taught something new. The SiGe heterobipolar transistor [4] was a result of such a learning. Strain became useful in improving desired properties. Single-electron effects provide one of the most sensitive ways to make precision measurements of fields and charges, and the charge storage in defects rather than nanocrystals is the basis for the enormously useful non-volatile archival memories that have become ubiquitous. While the direction of using reconfigurability around defective elements in large-scale circuits did not work, the learning of the distributions became the source for probabilistic approach of computation in the inexact and Bayesian direction.

Wrong may be right, Right may be wrong: This right-wrong apparition conundrum is quite pervasive in technology. New technology developments in time can overcome practical limitations. A classic example is chemical-polishing for planarization. With the older feel for the very messy grit-based polishing and surface preparation of semiconductor wafers, the prognosis for planarizing multi-material surfaces looked poor. The incredible device structures with buried structures such as trench capacitors, three-dimensional integrated systems, silicon-on-insulator, the multi-decade layered field-programmable gate arrays, and the 3d NAND flash are all examples of chemical-polishing technology’s success. In the 1980s, the phrase “silicon succeeds because of SiO²” was popular. Low interface state density, stability, reliability, scaling, process technology, polysilicon gates, reproducible threshold voltages, were directly or indirectly linked to the oxide. Today, the most advanced device structures have high permittivity materials with ionicity and softness built in with the SiO² interface layer keeping mobility reasonable, metal silicides gates, and other changes abound. SiO² is only in the background.

Statics and dynamics are different: As the scaling era ended, it was inevitable that numerous new ideas floated. Some examples are in the false-forays discussion. But, one specific one connected to understanding ideas—not just of the transistors—is the difference between static and dynamic behavior. The dynamic behavior, because of the digital circuit usage at speeds well below fundamental limits, is often mistaken for quasi-static behavior. The fallacy of the consequences glares when static analysis is used to project dynamics. There are quite a few examples from the transistor age. Low barrier Schottky contacts to replace doped contacts of transistors is one. As a diode, one has access to the electrochemical potential to both sides of the interface through the contacts. Carriers can flood in and out of the semiconductor so dielectric relaxation defines this flow of a Schottky junction. In a transistor, there is an intervening channel of opposite polarity and insulator region. A change from off to on state means that an insulator and a low carrier and charge density region exists in between. Fields are largely terminating at the electrodes in the starting of the dynamic change from the off state. The injection of carriers is now rate limited in a very different way, and the dynamic resistance of (k_B T/e)/I is subject to the actual current I, which is vanishing, and only starts becoming significant once electrons jump across and through the barrier to the channel to change its electrochemical potential. A large time-constant exists. Static and dynamic behavior is different. A different version of this same problem is when impact ionization is to be used to improve subthreshold slope. Impact ionization takes time to build up and needs space for the carriers to acquire the energy. This is the phenomenon of dead time and space known from the days of Impatt diodes, an early microwave generating two-terminal device. Here again, static and dynamic behavior are very different. A very different version of this static-dynamic issue is the use of ferroelectricity in the floating region of the gate. Hysteresis curves are static curves, where relaxation processes are allowed to complete during a change. By constraining—a clamping—one can unfold the negative slope behavior. This is similar to the voltage- or current-driven unveiling of tunnel diode characteristics. One shows hysteresis, the other a negative slope. In the case of a device, with no clamping, and sudden change in bias conditions, one now has the dynamic effect of the propagation of the domains whose end result the hysteresis curve describes. The dynamic behavior is entirely different from the static behavior.

Poisson and Gaussian are different: The transistor usage shows the consequences of the law of small and large numbers and central-limit theorem in a variety of ways with consequential effects that show up both in logic usage and in memories. As transistor dimension decreases, with low averages of dopants, the threshold voltage distributions become skewed Gaussian as the Poisson effect of dopants has a position-levered consequence. This same consequence is important for the two-dimensional effects near the junctions. Poisson consequence of dopants is particularly acute in random-walk memories such as dynamic random access or floating storage of flash. For the dynamic case, the appearance of a potential perturbation by a site that can trap an electron in the close vicinity of the transistor leads to a large change in the sub-attoampere current flowing through the transistor. This leads to serious variable retention. Dynamic memories have to be refreshed far more often to compensate for this leakage in the very few cells subject to this Poisson effect. In the case of floating storage structures, which are meant to be non-volatile, this implies one has to take out the structure’s entire row and column. Such Poisson effects are inevitably going to increase with small devices.

Thermodynamics rules; Energy, speed and power: The mix of small dimensions and numbers and large densities and numbers and a healthy curated assembly of these properties together makes the semiconductor world very thermodynamics centric [5]. Noise and fluctuations—thermodynamics of what is programmed into the substrate, the signal, and the surrounding world—set the lower limits of operation in determinisitic approaches. The probabilistic distribution functions arising thermodynamically do the same in probabilistic approaches. In all these cases the entropy determines the rate at which faults, errors and variabilities will appear both at the transistor scale as well as the integrated system scale. So, Rent’s rule, fault tolerance, power, energy, time, hierarchy of the design are all connected.

It is here that the difficulties with the emphasis on one property above all shows up. Herbert Simon [6] has an interesting parable to illustrate the importance of hierarchy. Hora and Tempus are two watchmakers. Both made watches with a thousand little individual parts. Tempus made his watches by putting all the parts together in one go, but if interrupted, for example, by the phone, he had to reassemble it from all these one thousand parts. The more the customers liked the watch, the more they called, the more Tempus fell behind. Hora’s watches were also just as good. But he made them using a hierarchy of subassemblies. The first group of subassemblies used ten parts each. Then ten such subassemblies were used to build a bigger subassembly, and so on. Proud competitors in the beginning, Tempus ended up working for Hora. If there is a one-in-a-hundred chance of being interrupted during the assembling process, Tempus, on average, had to spend four thousand times more time than Hora to assemble a watch. He had to start all over again from the beginning. Hora had to do this only part of the way. Hierarchy made Hora’s success at this rudimentary complex system possible. The hierarchy of this node-connection assemblage leads to the Janus effect of Arthur Koestler [7]. Nodes of the hierarchy are like the Roman god, whose one face is toward the dependent part and the other toward the apex. This link and proliferation of these links with their unusual properties are crucial to the emergent properties of the whole.

Thermodynamics places constraints via the confluence of energy, entropy, and errors. The complex system consists of a large number of hierarchical subassemblies. Reduction in errors requires large reduction in entropy. This demands energy. But, such a reduction process requires the use of more energy. Heat arises. To be sustainable—avoiding overheating, even as energy flow keeps the system working—requires high-energy efficiency, limits to the amount of energy transformed, and keeping a lid on errors. Having to deploy so much energy at each step makes the system enormously energy hungry. The giant turbines that convert energies to electric form—the mechanical motion of the blades to the flow of current across voltages—need to be incredibly efficient so that only a few percent—if that—of that energy is lost to the volume. And a turbine is not really that complex a system. Computation and communication have not yet learned this lesson and it shows up powerfully in the neural-network kingdom.

Nature has found its own clever way around this problem. Consider the complex biological machine that is the human. The building or replacement processes occur individually in very small volumes—nanoscale—and are happening in parallel at Avogadro number scale. All this transcription, messaging, protein and cell generation, and so on, requires energy and nature works around the error problem by introducing mechanisms for self-repair. Very low energy (10–100 k_B T, the k_B T being a good measure of the order of energy in a thermal motion) breaks and forms these bonds. Errors scale exponentially with these pre-factors. At 10 k_B T energy, one in 100,000 building steps, for example, for each unzipping and copying step, will have an error, so errors must be detected, the building machine must step back to the previous known good state and rebuild that bond, a bit like that phone call to Tempus and Hora, but causing Hora to restart only from an intermediate state. Nature manages to do this pretty well. The human body recycles a body weight of ATP (adenosine triphosphate)—the molecule for energy transformation—every day so that chemical synthesis, nerve impulse propagation, and muscle contraction can happen. Checking and repairing made this complexity under energy constraint possible. So, all the work on reconfigurability and using error-prone parts, which von Neumann [8] started with, may yet come back. Although, so far in neural network implementations, we have largely demonstrated unreliability from reliable components.

Naming matters: There exists a culture in engineering where the naming of inventions or new techniques is an art in itself. In compound semiconductors, MODFET or SDHT didn’t work out, HFET and HEMT did. DELTA FET [9] as a name is lost in history and FINFET rules, with tri-gate transistor struggling behind. I failed with inexact computing [10] as a nom de guerre. Approximate computing was how that subject track followed the early work. Not many recognize an interesting high current effect in double heterostructure bipolar transistors since the paper just titled it as a new effect [11]. But, nanocrystal memory [12] stuck. A good name attracts attention, and in turn, progress in that subject area. Sometimes the naming is a success by being a means to obfuscation as is the case with the node naming. They have nothing to do with the actual channel length that rate limits the transport, but a means to halo building.

People matter, hard work matters: Good ideas come as a spark in an environment where many discussions and thoughts are being exchanged with critical technical assessment based on the current state of learning, which is by itself incomplete. This requires a variety of technical viewpoints and perspectives, interesting personalities, personalities often with attitudes, different cultures and evolving cultures. Being surrounded by brilliant people provides one an opportunity to learn and in turn create. Quite often some of the people may be brilliant, albeit very self-centered, and one needs to figure out ways to live with such folks too. But people who shuffle things from here to there and from there to here should be avoided at all costs. The hard work should be towards something learned and created, and not just heat of the first law of thermodynamics.

Biography

Sandip Tiwari, a native of India, was educated starting in Physics before moving to Electrical Engineering, attending IIT Kanpur, RPI, and Cornell, and after working at IBM Research, joined Cornell in 1999. He has been a visiting faculty at Michigan, Columbia, Harvard, Stanford, TUM, University of Orsay, and ETH, the founding editor-in-chief of Transactions on Nanotechnology and authored the popular Electroscience textbook series of Oxford University Press. He is the Charles N. Mellowes Professor in Engineering Emeritus and Visiting Distinguished Professor at IIT Kanpur. He was director of USA’s National Nanotechnology Infrastructure Network until 2012.

His research has spanned the engineering and science of semiconductor electronics and optics, and has been honored with the Cledo Brunetti Award of the Institution of Electronic and Electrical Engineers (IEEE), the Distinguished Alumnus Award from IIT Kanpur, the Young Scientist Award from Institute of Physics, and the Fellowships of American Physical Society and IEEE. Particularly joyful to him is discovering scientific explanations, uncovering new phenomena, inventing new devices and technologies, and moving in directions that are of broader societal use. His current research interests are in the challenging questions that arise when connecting large scales, such as those of massively integrated electronic systems, a complex system, to small scales, such as those of small devices and structures that come about from the use of nanoscale, bringing together knowledge from engineering and physical and computing sciences. His personal passion is satyagraha, in bringing critical and broad thinking skills and of openness to the young through education, and as a global citizen in the phenomenology that pervades the world.

References

[1] Hoeneisen, B. and Mead, C., “Fundamental Limitations in Microelectronics-I. MOS Technology, Solid-State Electronics,” vol. 15, pp. 819–829 (1972) for field-effect transistor and Hoeneisen, B. and Mead, C., “Fundamental Limitations in Microelectronics-II. Bipolar Technology”, Solid-State Electronics, vol. 15, 891–897 (1972) for bipolar transistors.

[2] J. D. Meindl, “Theoretical, practical and analogical limits in ULSI,” Technical digest of the International Electron Devices Meeting, 8–13 (1983)

[3] S. Tiwari, J.J. Welser and P.M. Solomon, “Straddle-Gate Transistor: Changing MOSFET Channel Length Between Off- and On-State Towards Achieving Tunneling-Defined Limit of Field-Effect,” Technical Digest of IEEE International Electron Devices Meeting, San Francisco, Dec., 737–740 (1998)

[4] S.S. Iyer, G.L. Patton, S.L. Delage, S. Tiwari and J.M.C. Stork, “Silicon-Germanium Base Heterojunction Bipolar Transistors by Molecular Beam Epitaxy,” Technical Digest of IEEE International Electron Devices Meeting, Dec., 874(1987)

[5] S. Tiwari, “Implications of scales in processing of information,” Proc. of IEEE, V.103, 1250-1273 (2015)

[6] H. A. Simon, “The architecture of complexity,” Proceedings of the American Philosophical Society, 106(6), 467–482 (1962)

[7] A. Koestler, The Ghost in the Machine, Hutchinson & Co., London, (1967)

[8] J. von Neumann, “Probabilistic logics and the synthesis of reliable organ-isms from unreliable components,” Automata Studies, 43–98 (1956).

[9] D. Hisamoto, T. Kaga, Y. Kawamoto and E. Takeda, “A fully depleted lean-channel transistor (DELTA)” Tech. Digest of the International Electron Devices Meeting, 833–836 (1989)

[10] J. Y. Kim and S. Tiwari, “Inexact Computing for Ultra Low-Power Nanometer Digital Circuit Design,” Tech. Dig. of IEEE/ACM International Symposium on Nanoscale Architectures, 24–31 (2011)

[11] S. Tiwari, “A New Effect at High Currents in Heterostructure Bipolar Transistors,” IEEE Electron Device Letters, EDL-9, No. 3, p. 142 (1988)

[12] S. Tiwari, F. Rana, H. Hanafi, A. Hartstein, E. Crabbe and K. Chan, “A Silicon Nano-Crystals Based Memory,” Applied Physics Letters, 68, 1377–1379 (1996)

FROM INVENTION OF THE TRANSISTOR TO VLSI
TO UBIQUITOUS COMPUTING

YUAN TAUR

Transistors as Amplifiers

The word “transistor” came from “transfer resistor”. A transistor is a 3-terminal device with the 3rd terminal (the transfer electrode) modulating the current between terminals 1 and 2 (the resistor). It was developed as a solid-state, i.e., semiconductor, replacement of the bulky and power-consuming vacuum tubes for amplifying small input signals in telecommunication. While 1947 is generally regarded as the year the transistor was invented, the concept of the transistor was conceived nearly a hundred years ago in a 1926 patent filing on FET (Field-Effect Transistor). What was discovered in 1947 was actually a bipolar transistor, which operates through the injection of minority carriers in a forward-biased p-n junction. As such, there is a steady-state current flow in all bipolar transistors in the on state. An FET, on the other hand, is a two-dimensional device with the current flow perpendicular to the field applied from the transfer electrode. In that respect, there can be negligible current flow from the transfer electrode in the on state. In particular, a MOSFET (Metal-Oxide-Semiconductor FET) is turned on by producing a field in the semiconductor (between terminals 1 and 2) through an insulator (or vacuum), hence has no steady-state current in the transfer terminal. However, a MOSFET is more challenging to build because the presence of surface states would hinder the effect of the transfer electrode. Only until 1960, more than 30 years after its invention, was the first MOSFET realized. Today, MOSFETs have grown to be the main constituent of the VLSI (Very Large Scale Integration) microchips through its application in digital computation.

Digital Computation

During the same time period of the transistor invention, digital computation has become the mainstream approach to computing systems. The first general-purpose electronic digital computer was constructed in 1946 using vacuum tubes as the active logic elements. In the late 1950s, semiconductor transistors replaced vacuum tubes in digital systems because of their smaller size and lower power consumption. The late 1960s and ’70s witnessed further dramatic advances in computer hardware, fueled by several key breakthrough inventions. The first was the fabrication of the integrated circuit (IC), a tiny silicon chip containing hundreds of transistors, diodes, and resistors. This microcircuit made possible the production of electronic systems with higher operating speeds, capacity, and reliability at a significantly lower cost. Another key milestone in the VLSI evolution is the invention of the one-transistor DRAM (Dynamic Random Access Memory) cell. It utilizes circuit techniques to refresh and maintain the charge stored on a reverse-biased p-n junction. A third key milestone is the invention of the CMOS (Complementary-MOS) circuit. It consists of n-channel and p-channel MOSFETs connected in parallel at the input and in series at the output such that there is no steady-state conducting path between the power supply terminals. Although CMOS occupies a larger area and has a longer delay time as the input signal has to turn on/off two MOSFETs, it becomes the key constituent of logic circuits because of its negligible standby power consumption. Note that while a bipolar transistor makes a better amplifier due to its high voltage gains, a complementary bipolar circuit does not have the same low power feature as CMOS because there is always power dissipation in the turned-on bipolar transistors. CMOS had a humble beginning, with its initial application in battery-operated wrist watches and hand-held calculators. Both are low-speed devices.

LSI and VLSI

The development of large-scale integration (LSI) enabled hardware manufacturers to pack thousands of transistors and other related components on a single silicon chip about the size of a baby’s fingernail. Such microcircuitry yielded two devices that revolutionized computer technology. The first of these was the microprocessor unit (MPU), which is an integrated circuit that contains all the arithmetic, logic, and control circuitry of a central processing unit. Its production resulted in the development of microcomputers, systems no larger than portable television sets yet with substantial computing power. The other important device to emerge from LSI circuitry was the semiconductor memory (DRAM). This compact storage device is well suited for use as the main memory in minicomputers and microcomputers, particularly those designed for high-speed applications. Such compact electronics led in the late 1970s to the development of the personal computer, a digital computer that was small and inexpensive enough to be used by ordinary consumers. By the beginning of the 1980s, integrated circuitry had advanced to very-large-scale integration (VLSI). This design and manufacturing technology significantly increased the circuit density of microprocessor, memory, and support chips—i.e., those that serve to interface microprocessors with input-output devices. By the 1990s, some VLSI circuits contained more than 3 million transistors on a silicon chip less than 0.3 square inches (2 square cm) in area. The use of personal computers grew through the 1980s and ’90s. When MOSFETs are made smaller, the density increases and they switch faster with lower power dissipation. By the late 1990s, all mainframe computers were made in CMOS technology. Today’s microchips contain 10 billion to 1 trillion transistors, with an MPU clock frequency reaching 5 GHz. In the meantime, floating gate MOSFETs, produced since the 1980s as nonvolatile memory arrays, have become inexpensive enough to replace magnetic tapes and laser disks as the portable storage units.

Standard bulk MOSFETs can be scaled to a gate length of 20 nm or so, which requires a body doping reaching its limits. To go beyond that, the VLSI industry turned to double-gate MOSFETs with no doping requirements. The concept of double-gate MOSFETs was introduced in a patent filing as early as 1935. However, they cannot be fabricated using the conventional planar IC technology, especially with the source-drain regions self-aligned to both gates. A version of the double gate MOSFET was fabricated in 1989, which was later given the name FinFET. The manufacturing of FinFET into VLSI products was accomplished in 2012 by applying a sub-lithographic patterning technique patented in 1980. This extended the gate length from 20 nm to potentially 10 nm. It will be extremely difficult to further push the gate length to 5 nm, as that would require the fin thickness to approach atomic dimensions below 2 nm. Because of the limitation to gate length, the clock frequency of MPU has largely saturated at 5 GHz. Instead of pushing the performance of uniprocessors, the industry turned to parallelism with multiple CPUs on a chip to enhance the processor throughput. Another limiting factor is the chip power which becomes a serious issue since the thermal voltage, kT/q, is fixed. The consequence is that the operating voltage of VLSI chips cannot be reduced much below 1 V. One future direction of the IC industry is 3D integration, with which the circuit density increases linearly with the number of active layers.

Evolution of semiconductor devices and circuits and of the circuit performances

Ubiquitous Computing

The spread of the World Wide Web in the 1990s brought millions of users onto the Internet, the worldwide computer network, and by 2019 about 4.5 billion people, more than half the world’s population, had Internet access. Computers became smaller and faster and were ubiquitous in the early 21st century in smartphones and later tablet computers. Other examples include smart cars, CMOS cameras, GPS road guidance systems, etc. The invention of the transistor 75 years ago has brought unprecedented growth to the semiconductor industry, with an enormous impact on how people work and live.

Biography

Yuan Taur is a Distinguished Professor at the Department of Electrical and Computer Engineering, University of California, San Diego. Prior to UCSD, he has been with IBM T. J. Watson Research Center from 1981 to 2001. He has served as the Editor-in-Chief of IEEE Electron Device Letters from 1999 to 2011. He received IEEE Electron Devices Society’s J. J. Ebers Award in 2012 “for contributions to the advancement of several generations of CMOS process technologies.” He coauthored a book with Tak H. Ning, “Fundamentals of Modern VLSI Devices,” published by Cambridge University Press, 1st ed. 1998, 2nd ed. 2009, 3rd ed. 2022.