One of IBM’s most important MOS Technology, 1963-1974: A Dozen Crucial Years contributions to MOS research came from the Components Division, which was responsible for developing by Ross Knox Bassett and manufacturing bipolar for its large computer systems and had very little interest in MOS transistors line can be drawn from the as such. As part of its work on Frosch’s and Derick’s work on bipolar transistors, Donald Kerr and silicon dioxide to the MOS (metal- a group of engineers had discovered A that depositing small amounts of oxide-) ’s domi- nance of semiconductor technology, phosphorous on the silicon-dioxide but it is neither short nor straight. That surface and forming a layer of line has several discernable segments, phosphosilicate glass (PSG) could first from Frosch and Derick’s work, limit the amount of leakage in bipolar until 1963. In this interval, by and transistors and play an important role large, no one thought seriously about a in enhancing the stability of MOS metal-oxide-semiconductor as a viable transistors. Jerome Eldridge and Pieter technology in its own right. The second Balk from IBM Research implemented segment runs from 1963, when the this work by using thin layers of combination of integrated circuits and PSG to make stable MOS devices. the planar manufacturing process had Other important work on the physics Fig. 2. Drawing of Atalla and Kahng’s “silicon-silicon dioxide surface device,” now known as and chemistry of MOS devices done led people to see MOS transistors as a the MOS transistor, from a 1961 Bell Labs technical memorandum by Kahng. (Courtesy AT&T potentially promising semiconductor Archives and History Center) by IBM included an examination of technology, until the mid-1970s, at the important role of annealing in Between 1964 and 1967 Deal, Grove which point the MOS transistor had aluminum structures. He then tried enhancing the surface conduction of and Snow, along with their manager C. been established as a commercially applying aluminum with an electron- MOS devices, and a demonstration of T. Sah, published over two dozen papers successful and sustainable technology. beam evaporator and discovered the advantages of using silicon with a related to the silicon-silicon dioxide This article will detail that second that the key to stability was not the crystal orientation (100) for making interface, which to all appearances 7 segment, concentrating on work done by particular metal used, but the method MOS devices. put the MOS structure on a firm Corporation of evaporating it. He then purposely In spite of all the work IBM scientific footing. The achievements and IBM, and will show that three contaminated devices with sodium, Research did on MOS technology, it of the trio were most clearly seen in types of work were crucial during this lithium, magnesium, or calcium and faced an inherent difficulty in getting a semiconductor physics course that period: first, research on the chemistry ascertained that sodium led to the the rest of the company interested in they put together with other Fairchild and physics of MOS structures; second, highest drifts. (Researchers later found the technology. This lack of interest researchers. Its centerpiece was the trio’s product design and development to that the tungsten filament wire used occurred because MOS transistors were research. Grove published the course create integrated circuits that had some in the evaporation process had been inherently much slower than bipolar Fig. 1. First two pages of Frank Wanlass’s U.S. patent on complementary MOS transistor circuitry. notes as a book, Physics and Technology advantages over bipolar technologies; extruded through a die lubricated transistors in the 1960s, given the gate of Semiconductor Devices, which became and third, organizational change with sodium, and was therefore lengths then achievable. IBM, which the authoritative text in the field for a to create environments where MOS simultaneously evaporating sodium made most of its money from large his CMOS circuitry and the low MOS transistors useless as a product. If generation.5 technology could thrive. and aluminum.) Snow’s work led to computer systems, wanted transistors power consumption it allows has been these problems could be solved, MOS an effort to search for and minimize But in spite of the work of this group, with speeds as fast as practically one of the technical foundations of transistors would be technically viable. sodium in any materials used in the Fairchild was able neither to consistently possible. Skepticism about the value Early Research at Fairchild MOS’s dominance over the last three The first member of the group was MOS production process.3 make stable MOS devices throughout of MOS technology reached the point decades. Wanlass, who often worked Bruce Deal, who came to Fairchild in Another focus of the group was the organization nor to develop a that in October 1965, IBM’s Corporate In its earliest years, Fairchild had put at the very edge of what was possible, March 1963 with a PhD in chemistry surface states. In the late 1940s, John successful MOS product line. Problems of Technical Committee directed a lot of time and effort into studying seldom published and bounced around from Iowa State University and many Bardeen had developed a theory technology transfer hampered Fairchild’s IBM Research to halt work on this the surfaces of bipolar transistors, among marginally successful MOS years researching oxidation processes. of surface states to account for the ability to make stable devices in both its technology. But IBM Research refused but the first sustained work on the companies, and is one of the vastly Andrew Grove joined later that spring failure of experiments suggested by development groups and its Mountain to do so, and the Corporate Technical MOS transistor as a potential product under-recognized figures in the history after receiving his PhD in chemical William Shockley and executed by View manufacturing facility. This would 1 Committee was unable to enforce its came from Frank Wanlass, who of MOS technology. engineering from UC Berkeley, where Walter Brattain and others in hopes have been a serious problem if Fairchild decree within the corporation.8 joined Fairchild in August 1962 after After Wanlass’s MOS transistor work he had worked on fluid mechanics. of building a solid-state amplifying had developed any MOS products that The doubts within IBM about MOS earning a PhD in physics from the in early 1963, Gordon Moore, then Ed Snow came later that year from the device. Bardeen proposed that the people really wanted to buy, but as of technology, verging on downright University of Utah. In a period of director of research and development University of Utah, where he had earned cause of the failure was surface states, 1968 the company had developed MOS hostility in some quarters, led IBM remarkable creativity stretching until at Fairchild, and C. T. Sah, the manager a PhD in solid-state physics based on memory chips, arithmetic-unit chips, 2 charge carriers trapped at the surface Research to redouble it search for his departure for the start-up General of the solid-state physics department, the migration of ions in quartz. of the semiconductor that prevented calculator chips, and customizable- areas where the technology could Microelectronics in December 1963, began putting together a team to In October 1963 this Fairchild group electric fields from penetrating beneath logic chips without finding significant prove relevant to IBM, resulting in Wanlass explored the chemistry and understand the MOS structure and made a breakthrough discovery. Snow the surface. When Fairchild started customers for any of them. 1966 to the program’s reorientation physics of MOS structures, built MOS the silicon-silicon dioxide system in a began a project assuming that different its work, previous work on surface toward semiconductor memory. At integrated circuits, and considered systematic way. Moore’s main reason metals applied as a gate electrode over states usually included caveats about MOS Technology at IBM that time the typical division of how various MOS phenomena could for starting this team was to produce the silicon-dioxide layer might show the difficulties of reproducibility and labor in computers was for logic to be commercially exploited. Wanlass’s better bipolar transistors—Fairchild’s different levels of stability. Researchers that results were valid only for specific International Business Machines be implemented using some kind greatest technological contribution was main area of business. But this work typically evaporated aluminum onto cases. Deal developed a number of was another major contributor to of semiconductor technology and the invention of CMOS (complementary would also be expected to address the the silicon dioxide using a tungsten techniques to eliminate these surface MOS technology. Its research group for memory to employ some type MOS), which led to transistor circuits stability problems of MOS transistors. filament, but because of the extremely states at the interface between silicon at Yorktown Heights, New York, was of magnetic technology. Computer that consumed almost no power in Up to this time, the problems of high melting points of platinum and the oxide layer. He later detailed drawn to this technology as a way to memory was one area in which speed standby operation (see Figs. 1 and 2). MOS stability were so great—an MOS and tantalum, Snow instead used an various characteristics of the surface- begin work in planar silicon technology, was not necessarily at a premium; While the complexities of building transistor’s characteristics might vary electron-beam evaporator to apply these state charge—specifically showing the which previously had been the exclusive cost was also a large factor. Here is CMOS circuits were so great in the by over a hundred volts over time metals. As he examined the stability of role that dry-oxygen heat treatments domain of IBM’s Components Division where MOS technology found interest 1960s that most firms concentrated or with changes in temperature and these platinum and tantalum structures, played in reducing the surface-state in East Fishkill, New York. IBM’s research within IBM; in fact, memory proved on making p-channel MOS circuits, operating conditions—that they made Snow found they were more stable than density. The group published a paper group had previously focused on more to be the most receptive home to MOS claiming that the surface-state density esoteric electronics technologies, such technology at IBM for almost twenty 4 This article is adapted from Ross Knox Bassett, To the Digital Age: Research Labs, Start-Up Companies and the Rise of MOS Technology was highly reproducible. as gallium arsenide or superconducting years. (Baltimore: Johns Hopkins University Press, 2002). electronics.6

46 The Electrochemical Society Interface • Fall 2007 The Electrochemical Society Interface • Fall 2007 47 One of IBM’s most important MOS Technology, 1963-1974: A Dozen Crucial Years contributions to MOS research came from the Components Division, which was responsible for developing by Ross Knox Bassett and manufacturing bipolar transistors for its large computer systems and had very little interest in MOS transistors line can be drawn from the as such. As part of its work on Frosch’s and Derick’s work on bipolar transistors, Donald Kerr and silicon dioxide to the MOS (metal- a group of engineers had discovered A that depositing small amounts of oxide-semiconductor) transistor’s domi- nance of semiconductor technology, phosphorous on the silicon-dioxide but it is neither short nor straight. That surface and forming a layer of line has several discernable segments, phosphosilicate glass (PSG) could first from Frosch and Derick’s work, limit the amount of leakage in bipolar until 1963. In this interval, by and transistors and play an important role large, no one thought seriously about a in enhancing the stability of MOS metal-oxide-semiconductor as a viable transistors. Jerome Eldridge and Pieter technology in its own right. The second Balk from IBM Research implemented segment runs from 1963, when the this work by using thin layers of combination of integrated circuits and PSG to make stable MOS devices. the planar manufacturing process had Other important work on the physics Fig. 2. Drawing of Atalla and Kahng’s “silicon-silicon dioxide surface device,” now known as and chemistry of MOS devices done led people to see MOS transistors as a the MOS transistor, from a 1961 Bell Labs technical memorandum by Kahng. (Courtesy AT&T potentially promising semiconductor Archives and History Center) by IBM included an examination of technology, until the mid-1970s, at the important role of annealing in Between 1964 and 1967 Deal, Grove which point the MOS transistor had aluminum structures. He then tried enhancing the surface conduction of and Snow, along with their manager C. been established as a commercially applying aluminum with an electron- MOS devices, and a demonstration of T. Sah, published over two dozen papers successful and sustainable technology. beam evaporator and discovered the advantages of using silicon with a related to the silicon-silicon dioxide This article will detail that second that the key to stability was not the crystal orientation (100) for making interface, which to all appearances 7 segment, concentrating on work done by particular metal used, but the method MOS devices. put the MOS structure on a firm Fairchild Semiconductor Corporation of evaporating it. He then purposely In spite of all the work IBM scientific footing. The achievements and IBM, and will show that three contaminated devices with sodium, Research did on MOS technology, it of the trio were most clearly seen in types of work were crucial during this lithium, magnesium, or calcium and faced an inherent difficulty in getting a semiconductor physics course that period: first, research on the chemistry ascertained that sodium led to the the rest of the company interested in they put together with other Fairchild and physics of MOS structures; second, highest drifts. (Researchers later found the technology. This lack of interest researchers. Its centerpiece was the trio’s product design and development to that the tungsten filament wire used occurred because MOS transistors were research. Grove published the course create integrated circuits that had some in the evaporation process had been inherently much slower than bipolar Fig. 1. First two pages of Frank Wanlass’s U.S. patent on complementary MOS transistor circuitry. notes as a book, Physics and Technology advantages over bipolar technologies; extruded through a die lubricated transistors in the 1960s, given the gate of Semiconductor Devices, which became and third, organizational change with sodium, and was therefore lengths then achievable. IBM, which the authoritative text in the field for a to create environments where MOS simultaneously evaporating sodium made most of its money from large his CMOS circuitry and the low MOS transistors useless as a product. If generation.5 technology could thrive. and aluminum.) Snow’s work led to computer systems, wanted transistors power consumption it allows has been these problems could be solved, MOS an effort to search for and minimize But in spite of the work of this group, with speeds as fast as practically one of the technical foundations of transistors would be technically viable. sodium in any materials used in the Fairchild was able neither to consistently possible. Skepticism about the value Early Research at Fairchild MOS’s dominance over the last three The first member of the group was MOS production process.3 make stable MOS devices throughout of MOS technology reached the point decades. Wanlass, who often worked Bruce Deal, who came to Fairchild in Another focus of the group was the organization nor to develop a that in October 1965, IBM’s Corporate In its earliest years, Fairchild had put at the very edge of what was possible, March 1963 with a PhD in chemistry surface states. In the late 1940s, John successful MOS product line. Problems of Technical Committee directed a lot of time and effort into studying seldom published and bounced around from Iowa State University and many Bardeen had developed a theory technology transfer hampered Fairchild’s IBM Research to halt work on this the surfaces of bipolar transistors, among marginally successful MOS years researching oxidation processes. of surface states to account for the ability to make stable devices in both its technology. But IBM Research refused but the first sustained work on the companies, and is one of the vastly Andrew Grove joined later that spring failure of experiments suggested by development groups and its Mountain to do so, and the Corporate Technical MOS transistor as a potential product under-recognized figures in the history after receiving his PhD in chemical William Shockley and executed by View manufacturing facility. This would 1 Committee was unable to enforce its came from Frank Wanlass, who of MOS technology. engineering from UC Berkeley, where Walter Brattain and others in hopes have been a serious problem if Fairchild decree within the corporation.8 joined Fairchild in August 1962 after After Wanlass’s MOS transistor work he had worked on fluid mechanics. of building a solid-state amplifying had developed any MOS products that The doubts within IBM about MOS earning a PhD in physics from the in early 1963, Gordon Moore, then Ed Snow came later that year from the device. Bardeen proposed that the people really wanted to buy, but as of technology, verging on downright University of Utah. In a period of director of research and development University of Utah, where he had earned cause of the failure was surface states, 1968 the company had developed MOS hostility in some quarters, led IBM remarkable creativity stretching until at Fairchild, and C. T. Sah, the manager a PhD in solid-state physics based on memory chips, arithmetic-unit chips, 2 charge carriers trapped at the surface Research to redouble it search for his departure for the start-up General of the solid-state physics department, the migration of ions in quartz. of the semiconductor that prevented calculator chips, and customizable- areas where the technology could Microelectronics in December 1963, began putting together a team to In October 1963 this Fairchild group electric fields from penetrating beneath logic chips without finding significant prove relevant to IBM, resulting in Wanlass explored the chemistry and understand the MOS structure and made a breakthrough discovery. Snow the surface. When Fairchild started customers for any of them. 1966 to the program’s reorientation physics of MOS structures, built MOS the silicon-silicon dioxide system in a began a project assuming that different its work, previous work on surface toward semiconductor memory. At integrated circuits, and considered systematic way. Moore’s main reason metals applied as a gate electrode over states usually included caveats about MOS Technology at IBM that time the typical division of how various MOS phenomena could for starting this team was to produce the silicon-dioxide layer might show the difficulties of reproducibility and labor in computers was for logic to be commercially exploited. Wanlass’s better bipolar transistors—Fairchild’s different levels of stability. Researchers that results were valid only for specific International Business Machines be implemented using some kind greatest technological contribution was main area of business. But this work typically evaporated aluminum onto cases. Deal developed a number of was another major contributor to of semiconductor technology and the invention of CMOS (complementary would also be expected to address the the silicon dioxide using a tungsten techniques to eliminate these surface MOS technology. Its research group for memory to employ some type MOS), which led to transistor circuits stability problems of MOS transistors. filament, but because of the extremely states at the interface between silicon at Yorktown Heights, New York, was of magnetic technology. Computer that consumed almost no power in Up to this time, the problems of high melting points of platinum and the oxide layer. He later detailed drawn to this technology as a way to memory was one area in which speed standby operation (see Figs. 1 and 2). MOS stability were so great—an MOS and tantalum, Snow instead used an various characteristics of the surface- begin work in planar silicon technology, was not necessarily at a premium; While the complexities of building transistor’s characteristics might vary electron-beam evaporator to apply these state charge—specifically showing the which previously had been the exclusive cost was also a large factor. Here is CMOS circuits were so great in the by over a hundred volts over time metals. As he examined the stability of role that dry-oxygen heat treatments domain of IBM’s Components Division where MOS technology found interest 1960s that most firms concentrated or with changes in temperature and these platinum and tantalum structures, played in reducing the surface-state in East Fishkill, New York. IBM’s research within IBM; in fact, memory proved on making p-channel MOS circuits, operating conditions—that they made Snow found they were more stable than density. The group published a paper group had previously focused on more to be the most receptive home to MOS claiming that the surface-state density esoteric electronics technologies, such technology at IBM for almost twenty 4 This article is adapted from Ross Knox Bassett, To the Digital Age: Research Labs, Start-Up Companies and the Rise of MOS Technology was highly reproducible. as gallium arsenide or superconducting years. (Baltimore: Johns Hopkins University Press, 2002). electronics.6

46 The Electrochemical Society Interface • Fall 2007 The Electrochemical Society Interface • Fall 2007 47 Bassett reached technological and scientific By 1968 Fairchild’s MOS program Conclusion About the Author (continued from previous page) maturity.”11 was like a computer that had locked Information exchanges took place up. Fairchild had too many MOS and By 1974 MOS technology was firmly Ross Bassett, an associate professor bipolar programs that were contending established as a viable commercial In late 1966 Robert Dennard, a in a myriad of other ways besides of history at North Carolina State for the same commercial territory and technology. Intel had introduced its 4 researcher at IBM, came up with a conferences. Researchers called up their University (Raleigh), earned a bachelor’s thus squeezing each other out. When kilobit memory chip and its second- fundamental breakthrough in MOS friends who worked for other firms. degree in electrical engineering at Robert Noyce and Gordon Moore left generation , the popular memory technology (see Figs. 3 and 4). Companies bought devices from other the University of Pennsylvania. After to found Intel that year, they reset 8080. But more important than the The basic unit in a semiconductor companies and either put them on working at IBM for eight years, he left the system. They would not have so achievements made in MOS was its memory is the cell, a configuration that test or reverse engineered them to get to pursue a PhD in history at Princeton, many competing processes running at future potential, as clearly described in stores one bit of information. Up until a sense of their competitive standing. which he received in 1998. He is author once. They had the ability to choose a a 1974 paper by Robert Dennard and his then, cells had been built using four Researchers carried information with of To the Digital Age: Research Labs, technology or two that they considered IBM colleagues. In it they detailed the or six transistors. Dennard discovered them as they moved from one company Start-Up Companies and the Rise of MOS most promising and concentrate on it. principles of device scaling, showing a way to build a cell using only a to another. In Silicon Valley, of course, Technology (Johns Hopkins University They could also choose the people and that as one reduced the size of an MOS single transistor. After attending an Fig. 3. The MOS developed by workers might stop after work for drinks Press, 2002). He is currently working organizational structure they wanted. transistor by a given factor, the delay IBM Research conference and being Wanlass at Fairchild in 1963. (F. M. Wanlass, at the Wagon Wheel Bar near Fairchild’s on the history of the Indian Institutes With so much research work having of the transistor decreased by the same impressed with the simplicity of the “Metal-Oxide Semiconductor Field Effect Transistor production facility in Mountain View. of Technology and the history of Indian and Microcircuitry,” Wescon 1963 Technical been done on MOS technology, Moore factor, while the power consumption various magnetic memories proposed, In these early years, when the future of faculty and students at MIT. He may be Papers, Session 13.2 Figure 6. © 1963 IEEE.) and Noyce decided that they would not per circuit decreased as the square of he began exploring analogs to these MOS technology was uncertain, these reached at [email protected]. need a research organization at Intel. that factor. Dennard’s group reported technologies in MOS technology. information exchanges helped advance Those who came to Intel from Fairchild its work on one-micron MOS devices, Ultimately Dennard focused on the The MOS Community the cause of MOS technology throughout References Research, like Grove, arrived with a and there was good reason to think simplest possible cell structure, a the semiconductor industry. new function: they would no longer be that MOS technology had a long run capacitor that stored a charge (thus Although the fate of the MOS researchers. ahead of it.16 1. Frank Wanlass, interview by author, producing a voltage) and an MOS IBM, Intel, and the Establishment transistor ultimately hinged on its For their core technology, Moore The 1960s research on the chemistry 18 October 1994. For other accounts transistor that would connect or success or failure at specific companies, of MOS Technology and Noyce decided to concentrate on and physics of MOS structures had of his MOS work at Fairchild, not connect the cell to the sensing its development was a cooperative effort, silicon-gate MOS technology, which been necessary, but not sufficient, see Michael J. Riezenman, IEEE circuitry. The one-device cell, as it came sometimes intentionally, sometimes By 1968 there were companies had been described by engineers and for the success of MOS technology. Spectrum, p. 44 (May 1991); and to be known, would occupy much less not, between companies who were such as American Microsystems or the scientists from Bell Labs. Silicon-gate It also required the development of Clifford Barney, Electronics, p. 64 (8 area than other cell configurations, nominally competitors. This industry- semiconductor operations of General technology had a self-aligning feature new products such as memories and October 1984). His CMOS patent is and therefore allow a single chip to wide effort benefited all who worked on Instrument that were focused on MOS that gave it density advantages over , where its technical Frank M. Wanlass, “Low Stand-by hold many more memory bits. The MOS technology, through the transfer technology, but it was far from clear other MOS processes, but it had yet to advantages could come into play. And Power Complementary Field Effect one-device cell was not widely used of information and the creation of that these firms—or the technology be manufactured in large quantities. MOS technology also needed new Circuitry,” U.S. Patent No. 3,356,858, commercially until the 1970s, but since a supportive atmosphere for the new itself—had the strength to endure. It Before Noyce and Moore left Fairchild, organizational structures in which to filed 18 June 1963. then it has been the dominant memory technology. Information transfers was one large established firm, IBM, and Federico Faggin at R&D had done some flourish without being overwhelmed 2. Bruce Deal, interview by Henry cell for main computer memories. In occurred through conferences, inter- one small new firm, Intel, that would preliminary investigations of silicon- by the more mature bipolar technology. Lowood, 9 June 1988, 12 July 1988, 1988 C. T. Sah called the one-device firm meetings, confidential exchange prove to be critical in establishing MOS gate transistors. Intel took this new Fred Seitz called William Shockley the Stanford Oral History Project, cell “the most abundant man-made agreements, acquisition of artifacts, technology. IBM’s role was important process, which had been described on Moses of Silicon Valley for having a Department of Special Collections, object on this planet earth.”9 and the movement of personnel. Major because as the world’s largest computer paper and been used to make a few vision but being unable to get to the Stanford University Libraries, research labs received information from company it set standards that other devices, and through a series of subtle promised land of a broader commercial Stanford, California; Joshua Cooper 10 new start-ups, and vice-versa. companies were likely to follow. Intel’s steps made it capable of producing large success. In much the same way, Fairchild Ramo, Time, p. 54 (29 December The most important conference for role was to develop an assortment of numbers of consistently stable MOS Semiconductor Corporation became the 1997-5 January 1998); Linda the description of MOS work in the innovative MOS products that opened devices. Moses of MOS technology. Geppert, IEEE Spectrum, p. 34 (June first half of the 1960s was the Solid up new markets for the technology. But a stable process would be of little 2000); George Gilder, Microcosm: The State Device Research Conference. This In 1968 Edward Davis, an IBM manager benefit without a product that customers Quantum Revolution in Economics and invitation-only conference was held who was responsible for developing new wanted to buy in quantity. Moore and Technology, p. 83, Simon & Schuster, every summer, typically on a university computer memories, made the decision Noyce had focused the company on New York (1989); Ed Snow, interview campus, and would attract around 500 that—because of the greater densities semiconductor memory as an area of by author, 9 January 1996. One part attendees. The conference produced possible in semiconductor memories and components that customers would buy of Snow’s dissertation research was no formal publications, as intended what was then regarded as saturation in in large volumes and would not require published as E. H. Snow and P. Gibbs, to promote the more open exchange the performance of magnetic memories— an extensive design effort. After an J. App. Physics, 35, 2368 (1964). of information. The Solid State Device all of IBM’s computer systems would use unsuccessful 256 bit MOS memory chip, 3. “Progress Reports—Solid State Research Conference would typically semiconductor devices for their main a team at Intel led by Robert Abbott Physics Section,” 1 June 1963, 23, be dominated by a single topic that was computer memories. He decided further and John Reed developed a 1 kilobit Box 8, Binder, Fairchild Camera of particular interest to the community that because MOS technology offered memory chip for Honeywell that struck and Instrument Technical Papers of researchers; in the years 1964 and a threefold advantage in density over paydirt. The 1103 chip (see Fig. 5), and Progress Reports, Collection 1965, that topic was the MOS transistor. bipolar technology, IBM computers as it was called, became the standard M1055, Department of Special Researchers gave papers, cornered one would employ MOS memories after an semiconductor memory chip for non- Collections, Stanford University, another in the hallways, and engaged initial phase-in period using bipolar IBM computer manufacturers and Stanford, California (hereafter in spirited rump sessions. technology. In 1972, after a successful established Intel as a viable concern.14 FCI); “Progress Reports—Solid Many of these conference program transferring MOS technology Intel also developed a range of other State Physics Department,” 1 presentations did result in journal articles from IBM Research into development products based on MOS technology, most October 1963, 18, Box 9, Binder, at later dates. A 1967 bibliography of and manufacturing, which involved famously the microprocessor but also FCI; “Progress Reports—Solid State work in “Metal-Insulator Semiconductor thousands of people at five sites, IBM the erasable, programmable read-only Physics Section,” 1 November 193, Studies” showed that while there had introduced new computer systems using memory (EPROM). The advantage these 10, 21-23, Box 9 Binder, FCI. been only five papers in that area 1024 bit MOS memory chips; these were products had, along with semiconductor 4. The early history of the work on published in 1960, by 1966 there were implemented in older, static memory memory, was that they were capable of surface states is reviewed in Lillian 181. George Warfield of Princeton cells because Dennard’s one-device cell almost limitless expansion into chips Hoddeson et al., Out of the Crystal University, the guest editor of a special posed too many technical challenges where each successive generation used Maze: Chapters from the History of issue on MOS structures in the 1967 at the time. MOS technology found more and more transistors, and so could Solid-State Physics, p. 467, Oxford Fig.4. Drawing from Robert Dennard’s U.S. patent IEEE Transactions on Electron Devices an organizational home at IBM in a follow the curve that Gordon Moore University Press, New York (1992); on a dynamic random-access memory device. Fig. 5. The Intel 1103 memory chip, one of noted that “this field has progressed plant in Burlington, Vermont, that was had described in 1965, now enshrined the first commercially successful semiconductor and Michael Riordan and Lillian Silicon is designated by 32 in this figure and the from its black magic phase, in which new to semiconductor technology and Hoddeson, Crystal Fire: The Birth of oxide layer by 34. as Moore’s law, suggesting that the memories. (Photograph courtesy of Intel various and sundry mysterious potions not dominated by bipolar technology number of transistor on an integrated Corporation.) the , p. 120, p. 126, coupled with assorted witchcraft were in the same way as IBM’s East Fishkill circuit would double every year. 15 W. W. Norton, New York (1997). See used to achieve ‘good’ devices, and has facility.12 also John Bardeen, Phys. Review, 71,

48 The Electrochemical Society Interface • Fall 2007 The Electrochemical Society Interface • Fall 2007 49 Bassett reached technological and scientific By 1968 Fairchild’s MOS program Conclusion About the Author (continued from previous page) maturity.”11 was like a computer that had locked Information exchanges took place up. Fairchild had too many MOS and By 1974 MOS technology was firmly Ross Bassett, an associate professor bipolar programs that were contending established as a viable commercial In late 1966 Robert Dennard, a in a myriad of other ways besides of history at North Carolina State for the same commercial territory and technology. Intel had introduced its 4 researcher at IBM, came up with a conferences. Researchers called up their University (Raleigh), earned a bachelor’s thus squeezing each other out. When kilobit memory chip and its second- fundamental breakthrough in MOS friends who worked for other firms. degree in electrical engineering at Robert Noyce and Gordon Moore left generation microprocessor, the popular memory technology (see Figs. 3 and 4). Companies bought devices from other the University of Pennsylvania. After to found Intel that year, they reset 8080. But more important than the The basic unit in a semiconductor companies and either put them on working at IBM for eight years, he left the system. They would not have so achievements made in MOS was its memory is the cell, a configuration that test or reverse engineered them to get to pursue a PhD in history at Princeton, many competing processes running at future potential, as clearly described in stores one bit of information. Up until a sense of their competitive standing. which he received in 1998. He is author once. They had the ability to choose a a 1974 paper by Robert Dennard and his then, cells had been built using four Researchers carried information with of To the Digital Age: Research Labs, technology or two that they considered IBM colleagues. In it they detailed the or six transistors. Dennard discovered them as they moved from one company Start-Up Companies and the Rise of MOS most promising and concentrate on it. principles of device scaling, showing a way to build a cell using only a to another. In Silicon Valley, of course, Technology (Johns Hopkins University They could also choose the people and that as one reduced the size of an MOS single transistor. After attending an Fig. 3. The MOS integrated circuit developed by workers might stop after work for drinks Press, 2002). He is currently working organizational structure they wanted. transistor by a given factor, the delay IBM Research conference and being Wanlass at Fairchild in 1963. (F. M. Wanlass, at the Wagon Wheel Bar near Fairchild’s on the history of the Indian Institutes With so much research work having of the transistor decreased by the same impressed with the simplicity of the “Metal-Oxide Semiconductor Field Effect Transistor production facility in Mountain View. of Technology and the history of Indian and Microcircuitry,” Wescon 1963 Technical been done on MOS technology, Moore factor, while the power consumption various magnetic memories proposed, In these early years, when the future of faculty and students at MIT. He may be Papers, Session 13.2 Figure 6. © 1963 IEEE.) and Noyce decided that they would not per circuit decreased as the square of he began exploring analogs to these MOS technology was uncertain, these reached at [email protected]. need a research organization at Intel. that factor. Dennard’s group reported technologies in MOS technology. information exchanges helped advance Those who came to Intel from Fairchild its work on one-micron MOS devices, Ultimately Dennard focused on the The MOS Community the cause of MOS technology throughout References Research, like Grove, arrived with a and there was good reason to think simplest possible cell structure, a the semiconductor industry. new function: they would no longer be that MOS technology had a long run capacitor that stored a charge (thus Although the fate of the MOS researchers. ahead of it.16 1. Frank Wanlass, interview by author, producing a voltage) and an MOS IBM, Intel, and the Establishment transistor ultimately hinged on its For their core technology, Moore The 1960s research on the chemistry 18 October 1994. For other accounts transistor that would connect or success or failure at specific companies, of MOS Technology and Noyce decided to concentrate on and physics of MOS structures had of his MOS work at Fairchild, not connect the cell to the sensing its development was a cooperative effort, silicon-gate MOS technology, which been necessary, but not sufficient, see Michael J. Riezenman, IEEE circuitry. The one-device cell, as it came sometimes intentionally, sometimes By 1968 there were companies had been described by engineers and for the success of MOS technology. Spectrum, p. 44 (May 1991); and to be known, would occupy much less not, between companies who were such as American Microsystems or the scientists from Bell Labs. Silicon-gate It also required the development of Clifford Barney, Electronics, p. 64 (8 area than other cell configurations, nominally competitors. This industry- semiconductor operations of General technology had a self-aligning feature new products such as memories and October 1984). His CMOS patent is and therefore allow a single chip to wide effort benefited all who worked on Instrument that were focused on MOS that gave it density advantages over microprocessors, where its technical Frank M. Wanlass, “Low Stand-by hold many more memory bits. The MOS technology, through the transfer technology, but it was far from clear other MOS processes, but it had yet to advantages could come into play. And Power Complementary Field Effect one-device cell was not widely used of information and the creation of that these firms—or the technology be manufactured in large quantities. MOS technology also needed new Circuitry,” U.S. Patent No. 3,356,858, commercially until the 1970s, but since a supportive atmosphere for the new itself—had the strength to endure. It Before Noyce and Moore left Fairchild, organizational structures in which to filed 18 June 1963. then it has been the dominant memory technology. Information transfers was one large established firm, IBM, and Federico Faggin at R&D had done some flourish without being overwhelmed 2. Bruce Deal, interview by Henry cell for main computer memories. In occurred through conferences, inter- one small new firm, Intel, that would preliminary investigations of silicon- by the more mature bipolar technology. Lowood, 9 June 1988, 12 July 1988, 1988 C. T. Sah called the one-device firm meetings, confidential exchange prove to be critical in establishing MOS gate transistors. Intel took this new Fred Seitz called William Shockley the Stanford Oral History Project, cell “the most abundant man-made agreements, acquisition of artifacts, technology. IBM’s role was important process, which had been described on Moses of Silicon Valley for having a Department of Special Collections, object on this planet earth.”9 and the movement of personnel. Major because as the world’s largest computer paper and been used to make a few vision but being unable to get to the Stanford University Libraries, research labs received information from company it set standards that other devices, and through a series of subtle promised land of a broader commercial Stanford, California; Joshua Cooper 10 new start-ups, and vice-versa. companies were likely to follow. Intel’s steps made it capable of producing large success. In much the same way, Fairchild Ramo, Time, p. 54 (29 December The most important conference for role was to develop an assortment of numbers of consistently stable MOS Semiconductor Corporation became the 1997-5 January 1998); Linda the description of MOS work in the innovative MOS products that opened devices. Moses of MOS technology. Geppert, IEEE Spectrum, p. 34 (June first half of the 1960s was the Solid up new markets for the technology. But a stable process would be of little 2000); George Gilder, Microcosm: The State Device Research Conference. This In 1968 Edward Davis, an IBM manager benefit without a product that customers Quantum Revolution in Economics and invitation-only conference was held who was responsible for developing new wanted to buy in quantity. Moore and Technology, p. 83, Simon & Schuster, every summer, typically on a university computer memories, made the decision Noyce had focused the company on New York (1989); Ed Snow, interview campus, and would attract around 500 that—because of the greater densities semiconductor memory as an area of by author, 9 January 1996. One part attendees. The conference produced possible in semiconductor memories and components that customers would buy of Snow’s dissertation research was no formal publications, as intended what was then regarded as saturation in in large volumes and would not require published as E. H. Snow and P. Gibbs, to promote the more open exchange the performance of magnetic memories— an extensive design effort. After an J. App. Physics, 35, 2368 (1964). of information. The Solid State Device all of IBM’s computer systems would use unsuccessful 256 bit MOS memory chip, 3. “Progress Reports—Solid State Research Conference would typically semiconductor devices for their main a team at Intel led by Robert Abbott Physics Section,” 1 June 1963, 23, be dominated by a single topic that was computer memories. He decided further and John Reed developed a 1 kilobit Box 8, Binder, Fairchild Camera of particular interest to the community that because MOS technology offered memory chip for Honeywell that struck and Instrument Technical Papers of researchers; in the years 1964 and a threefold advantage in density over paydirt. The 1103 chip (see Fig. 5), and Progress Reports, Collection 1965, that topic was the MOS transistor. bipolar technology, IBM computers as it was called, became the standard M1055, Department of Special Researchers gave papers, cornered one would employ MOS memories after an semiconductor memory chip for non- Collections, Stanford University, another in the hallways, and engaged initial phase-in period using bipolar IBM computer manufacturers and Stanford, California (hereafter in spirited rump sessions. technology. In 1972, after a successful established Intel as a viable concern.14 FCI); “Progress Reports—Solid Many of these conference program transferring MOS technology Intel also developed a range of other State Physics Department,” 1 presentations did result in journal articles from IBM Research into development products based on MOS technology, most October 1963, 18, Box 9, Binder, at later dates. A 1967 bibliography of and manufacturing, which involved famously the microprocessor but also FCI; “Progress Reports—Solid State work in “Metal-Insulator Semiconductor thousands of people at five sites, IBM the erasable, programmable read-only Physics Section,” 1 November 193, Studies” showed that while there had introduced new computer systems using memory (EPROM). The advantage these 10, 21-23, Box 9 Binder, FCI. been only five papers in that area 1024 bit MOS memory chips; these were products had, along with semiconductor 4. The early history of the work on published in 1960, by 1966 there were implemented in older, static memory memory, was that they were capable of surface states is reviewed in Lillian 181. George Warfield of Princeton cells because Dennard’s one-device cell almost limitless expansion into chips Hoddeson et al., Out of the Crystal University, the guest editor of a special posed too many technical challenges where each successive generation used Maze: Chapters from the History of issue on MOS structures in the 1967 at the time. MOS technology found more and more transistors, and so could Solid-State Physics, p. 467, Oxford Fig.4. Drawing from Robert Dennard’s U.S. patent IEEE Transactions on Electron Devices an organizational home at IBM in a follow the curve that Gordon Moore University Press, New York (1992); on a dynamic random-access memory device. Fig. 5. The Intel 1103 memory chip, one of noted that “this field has progressed plant in Burlington, Vermont, that was had described in 1965, now enshrined the first commercially successful semiconductor and Michael Riordan and Lillian Silicon is designated by 32 in this figure and the from its black magic phase, in which new to semiconductor technology and Hoddeson, Crystal Fire: The Birth of oxide layer by 34. as Moore’s law, suggesting that the memories. (Photograph courtesy of Intel various and sundry mysterious potions not dominated by bipolar technology number of transistor on an integrated Corporation.) the Information Age, p. 120, p. 126, coupled with assorted witchcraft were in the same way as IBM’s East Fishkill circuit would double every year. 15 W. W. Norton, New York (1997). See used to achieve ‘good’ devices, and has facility.12 also John Bardeen, Phys. Review, 71,

48 The Electrochemical Society Interface • Fall 2007 The Electrochemical Society Interface • Fall 2007 49 Bassett (continued from previous page) Gate Dielectric Process Technology for the Sub-1 nm

717 (1947); A. S. Grove, B. E. Deal, Equivalent Oxide Thickness (EOT) Era E. H. Snow, and C. T. Sah, Solid-State Electronics, 8, 145 (1965); and B. E. Deal, M. Sklar, A. S. Grove, and E. H. Future Technical Meetings Snow, J. Electrochem. Soc., 114, 266 by L. Colombo, J. J. Chambers, and H. Niimi (1967). 5. A review of the group’s work is he semiconductor industry is now where t is the physical thickness of constant of Si N , ~7.5) to the higher provided in B. E. Deal, A. S. Grove, E. SiO2 3 4 in its third generation of gate the interface layer, potentially without dielectric constant of the new gate H. Snow, and C. T. Sah, Trans. Metall. May 18-23, 2008 Oct. 4-9, 2009 dielectrics. The first generation nitrogen, t is the physical thickness dielectric (κ ≈ 15-20 for the HfO and Soc. AIME, 233, 524 (1965). For T SiON 2 was the silicon dioxide (SiO ) era from of the SiON film, κ is the dielectric ~4-24 for Hf based HfSiON. This third the papers by Fairchild researchers Phoenix, Arizona Vienna, Austria 2 the early 1960s to about the mid-1990s. constant of the nitrogen containing regime of gate dielectric technology during this time, see “Index to The benefits of SiO noted in the earlier SiO film, and the summation over “i” has also required the replacement of Published Technical Papers,” Bruce 2 2 articles in this issue of Interface included is for layers that do not have a uniform the doped poly-silicon gate electrodes Deal Papers, Collection M1051, Oct. 12-17, 2008 April 25-30, 2010 utilization as: (a) passivation of surface nitrogen profile. by metallic gate electrodes to both Department of Special Collections, dangling bonds and p-n junction Today, the era of SiON films with an reduce the poly-silicon depletion effect Stanford University Libraries, PRiME 2008 Vancouver, BC interfaces, (b) pattern/diffusion masking equivalent oxide thickness (EOT) in the and to improve the work function Stanford, California. ability, and (c) insulator supporting range of ~1.0-1.2 nm have been in mass match between the high-κ and gate 6. Bassett, To the Digital Age, Chapter 2. Honolulu, Hawaii medium for aluminum interconnects production for nearly a decade, meeting electrode materials, itself a significant 7. William E. Harding, IBM J. Res. Dev., Oct. 10-15, 2010 between sections of the integrated demanding transistor and reliability area of research. 25, 651 (1981); D. R. Kerr, J. S. Logan, circuit. In addition, the amorphous requirements. For SiON thickness less The methods utilized for the P. J. Burkhardt, and W. A. Pliskin, May 24-29, 2009 Las Vegas, NV SiO has a large energy gap (~9 eV) than ~1.0 nm, however, it was again fabrication of SiON for the second era IBM J. Res. Dev., 8, 376 (1964); J. M. 2 and a dielectric strength sufficient to found that direct tunneling leakage will be discussed in the first section, Eldridge and P. Balk, Trans. Metall. Soc. San Francisco, support electric field strengths of several currents became excessive through the status of HfSiON gate dielectric in AIME, 242, 539 (1968); J. M. Eldridge, California megavolts/cm, (106 V/cm); the latter the SiON. Scaling below ~1 nm for presented in the second section, and “A Thermochemical Evaluation of is especially critical for the required higher-performance devices and lower the appropriate issues will the Doping of SiO /Si with P O and 2 2 5 metal-oxide-semiconductor field than ~1.5 nm for lower-power devices briefly be summarized in the third B O ,” Research Report (8 December 2 3 effect transistor (MOSFET) operation. also became limited by poly-silicon section. 1966), IBM Research, Yorktown These latter benefits have enabled depletion as well as gate dielectric Heights, New York; P. Balk, “Effects the semiconductor industry to scale leakage. Initial solutions will require of Hydrogen Annealing on Silicon Silicon Oxynitride transistors down to about the 180 nm either higher content nitrogen in SiON Surfaces,” paper presented at the ECS technology node* , corresponding to an than currently in production or higher- 1965 spring meeting, San Francisco, SiON dielectrics were first SiO thickness of ~3 nm. At about this κ dielectric constant gate materials California, Extended Abstracts of For more information on these future meetings, contact ECS 2 introduced at an EOT of about 3 nm oxide thickness, direct tunneling leakage (κ ~ 10-20). the Electronics Division, Vol. 14, and at that time required less than Tel: 609.737.1902 Fax: 609.737.2743 currents rather than source-drain or Serious attention for alternate, higher Abstract 109, pp. 237-40; P. Balk, P. about 10 atomic percent nitrogen to substrate leakage currents, reached levels dielectric constant materials began in J. Burkhardt, and L. V. Gregor, Proc. www.electrochem.org minimize boron penetration and gate that were a significant portion of the about 1995 and have resulted in an IEEE, 53, 2133 (1965). leakage. Controlled incorporation of allowable device leakage (~ 33%). In extensive literature.3-5 Hafnium-based 8. “CTB Minutes, October 29, 1965,” nitrogen, especially concentration addition, at the sub 3 nm film thicknesses gate dielectrics have emerged as the 4 November 1965, 3, Box TAR 242, and depth profile, was critical to the regime, boron dopant penetration broad industrial choice for the high-κ IBM Technical History Project, IBM introduction of SiON. The key process from the boron-doped p+ poly-silicon gate dielectric material. The principal Archives, Somers, NY. that enabled the industry to introduce electrode through the SiO into the motivation for moving to high-κ gate 9. C. T. Sah, Proc. IEEE, 76, 1301 2 the first reliable SiON gate dielectric The Benefits of Membership Can Be Yours! channel region was a serious issue for dielectrics was the need to reduce the 6 (1988); Robert H. Dennard, IEEE was plasma nitridation of SiO2. pMOSFETs. An extensive literature has direct tunneling gate leakage currents. Trans. Electron Devices, 31, 1549 The advantage of plasma nitridation been published on the increasing SiO2 This was achieved by increasing the gate (1984). Dennard’s patent is Robert of SiO2 is its ability to (a.) control the gate leakage and boron penetration1,2 dielectric’s physical thickness inasmuch H. Dennard, “Field Effect Transistor dielectric layer thickness, (b.) precisely Join now for • Journal of The Electrochemical Society with decreasing oxide thickness and as the direct tunneling leakage current Memory,” U.S. Patent No. 3,387,286, control the content and/or location their impact on continued device is drastically reduced due to its filed 14 July 1967. of the nitrogen, and (c.) improve • Electrochemical and Solid-State Letters scaling as enunciated in Moore’s law. exponential dependence on the physical 10. This section is based on Bassett, To exceptional reliability. Plasma nitrided SiO2 can be The continuance of scaling gate dielectric thickness. Concurrently, the Digital Age, Chapter 5. created by (i.) top surface nitridation, • Interface, the ECS Member methodologies, however, has been the gate dielectric constant, κ, is 11. George Warfield, “Introduction,” (ii.) homogeneous SiON formation, discounts on all achieved by the incorporation of increased. These two concurrent IEEE Trans. Electron Devices, 14, 727 Magazine where the nitrogen is distributed nitrogen (~10 atomic %) in the SiO , changes result in the approximate (1967); Earl S. Schlegel, IEEE Trans. 2 uniformly through the oxide, and • Professional Development and whereby silicon oxynitride (SiON) constancy of the MOSFET’s capacitance, Electron Devices, 14, 728 (1967). ECS publications, (iii.) incorporation of nitrogen at the thicknesses ranging from ~3.0 nm a significant parameter controlling the 12. Bassett, To the Digital Age, Chapter 7. Education Si-SiO2 interface. Simulations have down to ~1.0 nm or so were achieved. speed of the device. This is achieved 13. J. C. Sarace, R. E. Kerwin, D. L. Klein, shown that top surface nitrogen page charges, The introduction of nitrogen in SiO by appropriately adjusting the ratio of and R. Edwards, Solid-State Electronics, • Discounts on Meetings and 2 profiles have better boron blocking to form SiON increases the effective the selected gate dielectric constant, κ, 11, 653 (1968); L. L. Vadasz, A. S. characteristics than other profiles. The Publications dielectric constant, κ. The higher to the dielectric’s physical thickness. Grove, T. A. Rowe, and G. E. Moore, meetings, and short bulk nitrogen concentration in SiON effective dielectric constant leads to a As a result the MOSFET electrically IEEE Spectrum, p. 28 (1969); F. Faggin has to be critically controlled, but even • Honors and Awards Program lower effective gate dielectric thickness behaves as though its gate dielectric and T. Klein, Solid-State Electronics, more critical is the concentration of N courses. that is referred to as the equivalent oxide thickness is smaller than its physical 13, 1125 (1970). • Career Center atoms at the Si-dielectric interface. thickness (EOT) described in general thickness (to reduce direct tunneling 14. Brian Santo, IEEE Spectrum, p. 108 The concentration of nitrogen by Eq. 1. leakage current) by the ratio of the at the interface has to follow the (November 1988). n dielectric constant of SiON (generally bonding constraint theory7 in order 15. Gordon E. Moore, Daedalus, 125, 55 EOT= + 3.9 (tSiON)i tSiO2 (1) slightly more than SiO ’s value of ~3.9 (Spring 1996). S ki 2 to minimize defects. According to i but significantly less than the dielectric 16. R. H. Dennard, F. H. Gaensslen, H-N. the bonding constraint theory, whose Yu, V. L. Rideout, E. Bassous, and A. www.electrochem.org R. LeBlanc, IEEE J. Solid-State Circuits, * The technology node refers to the metal line-to-line spacing—½ pitch—for a particular DRAM generation. The physical channel length 9, 256 (1974). for a logic chip consonant with the DRAM generation is about 45% of the technology node as discussed in the International Technology Roadmap for , ITRS.

50 The Electrochemical Society Interface • Fall 2007 The Electrochemical Society Interface • Fall 2007 51