10Information: Displays and Memory Devices (1981–2007)
Ovshinsky’s most important energy technologies, thin-film solar cells and NiMH batteries, were major commercial successes. But his information technologies—which were more radically innovative and based on his most original discoveries, the switching effects he first observed in the early 1960s—failed to realize their full commercial potential for ECD. The flat panel displays that Ovshinsky had envisioned in 1968, and which ECD’s subsidiary OIS (Optical Imaging Systems) contributed greatly to developing, ended up enriching other companies. Ovonic optical memories, such as rewritable CDs and DVDs, enjoyed a period of commercial success but again mostly profited others. And while many in the semiconductor industry recognized the enormous promise of Ovshinsky’s electrical phase-change memory, it lay dormant for years because it was not considered commercially viable. Finally, his innovative cognitive computer, based on a further extension of his phase-change technology, never advanced beyond its research phase.
This chapter about ECD’s information research between 1981 and 2007 thus appears at first to be mainly a story of missed opportunities and unrealized possibilities. From a later vantage point however, the story looks quite different. As of this writing in 2016, it seems that the time for Ovonic phase-change memory has finally come (a story we briefly outline in the epilogue) and that Ovshinsky’s information technologies, based on his crucial discovery of the Ovshinsky effect, may end up having the most impact of all his inventions.1 Those technologies thus both grow out of his early efforts as an independent inventor working on his own and also depend on his later collaborative approach, inventing with and through others as he did in ECD’s energy technologies.
Flat Panel Displays
Despite Ovshinsky’s prediction in 1968 of flat TVs that could hang on the wall, it took some time for him to begin developing them, partly because, as in the case of solar cells, he hoped they could be made from his Ovonic chalcogenide materials.2 That was a reasonable hope, because liquid crystal displays depend on a grid of thin-film switches that allow an electric field to rotate the crystals and let light through. But as with the solar cells, chalcogenides proved unsuitable.
In 1981 Robert Johnson, the former Burroughs executive who had served in the 1970s as a consultant (see chapter 6), became ECD’s senior vice president in charge of developing thin-film technologies other than solar cells. Johnson recognized that the amorphous silicon material ECD was now starting to produce for its solar cells could also be used to make either diodes or transistors as switches for displays. He convinced Ovshinsky to set up a program of building active-matrix liquid crystal displays (LCDs) using diodes, which he saw as a unique opportunity because other researchers were using transistors.3
Johnson had hoped that Dick Flask would lead ECD’s development of the LCDs, but “the first disaster,” from Johnson’s point of view, was that in early 1982 Flasck left ECD.4 Hoping to learn more about the state of the art in active matrix displays, Johnson then contacted Professor J. William Doane of the Liquid Crystal Institute at Kent State University in Ohio, who suggested that ECD invite Zvi Yaniv to give a talk. A bright and ambitious Romanian-born physicist educated in Israel, Yaniv was writing his PhD thesis on order parameters in liquid crystals. He planned to return to Israel to become president of the newly established Practical Engineering College at Hebrew University in Jerusalem and so was surprised when the day after his talk Johnson brought him to see Ovshinsky, who offered him a well-paying research position at ECD.5 Yaniv started working at ECD early in 1983.
He began by studying ECD’s amorphous silicon solar cells with Vin Canella. Yaniv recalled, “I said to myself my God these guys know how to make these diodes very, very well.” At Johnson’s suggestion, Yaniv and Canella started developing the amorphous silicon diodes for LCDs. Both physicists were “emotionally built,” which made them compatible. “There were times when we’d be in an office talking,” Canella recalled, “but we’d be shouting and screaming and laughing, and people would knock on the door and say, ‘Is everything okay?’”
After about a year, Yaniv and Canella, working with consultants Marvin Silver and Mel Shaw, had developed a 32 × 32 pixel prototype LCD using diodes. Johnson then convinced Ovshinsky that it was time to create an ECD subsidiary to commercialize diode-driven LCDs, and in May 1984, Ovshinsky created Ovonic Display Systems (ODS), with Johnson as president, Yaniv as vice president, and Ovshinsky serving as chairman of the board. Not long after that, Yaniv secured $300,000 to support the work of ODS from the large Israeli defense company Elbit. The growing staff of ODS also included Canella, John McGill, and Meera Vijan.6 They started making rudimentary prototype displays based on diodes, and “Zvi had ideas for mass-producing them,” Johnson recalled.
The new company proved fairly successful in attracting government research contracts because the military was interested in flat panel displays for aircraft cockpits, and ODS had the advantage of a unique approach. Many companies, including GE, Sharp, Canon, Toshiba, Philips, IBM Japan, Seiko-Epson, Hitachi, and AT&T, were developing amorphous silicon transistors for driving liquid crystal displays, but only ODS’s displays used diodes.
In 1985 Ovshinsky renamed the company Ovonic Imaging Systems (OIS), because it had begun to explore using the diodes for other imaging devices as well as for displays.7 In the summer of 1986, a new 50/50 joint venture called Quartet Ovonics was formed with the Chicago-based Quartet Manufacturing Co. The venture produced and marketed OIS’s first product using the amorphous silicon diode, the electronic whiteboard. This highly successful technology, still on the market today, digitized writing or drawings made during presentations. About a year later, OIS followed up with the “wand,” a small handheld scanner.8
At this point, however, losses of funding from Sohio in the solar program and from ANR in the battery program led to a steep decline in ECD’s stock and drastic layoffs (see chapters 8 and 9). Ovshinsky needed to pull another rabbit out of his hat. To raise new funds, he followed Nancy Bacon’s suggestion to take OIS public, and in December 1986, Ovshinsky and Yaniv traveled to New York to meet with potential investors.9
In the Park Lane Hotel overlooking Central Park, OIS demonstrated its imaging and display technology, including prototypes of a small liquid crystal display, image digitizers, and fax machines. “I remember every half hour we had another group on another floor and we were explaining to them how great we are,” Yaniv said. In a room with a large window commanding a view of Central Park, a potential investor asked how large they thought the future TV displays could become. Ovshinsky turned to the big window and said, “I think we can make it as large as this window—correct Zvi?” Yaniv gasped and whispered, “Probably.” It was another of Ovshinsky’s visionary claims that seemed wildly improbable at the time. (Today, flat screens based on transistors rather than diodes can be even larger.) His typical exuberance and confidence in the technology helped raise substantial funding for OIS.
One of the investors at the meeting was William Manning of the Manning and Napier Investment Company of Rochester, a firm that managed several billion dollars in assets. Starting in the late 1970s, Manning had taken a liking to ECD’s technologies and had many of his clients invest heavily in the company. “Bill Manning was always fascinated by technology,” Canella said. “Stan could spin a story, throw out the hook, and Bill Manning would bite.” Now, impressed with OIS’s imaging and display possibilities, Manning invested roughly $15 million in OIS, which made the public offering successful, and about the same amount in another ECD subsidiary, OSMC (Ovonic Synthetic Materials Company), whose work included magnets and x-ray mirrors.10 Yaniv remembered that at the end of the last talk he gave in New York, Manning put his hand on his shoulder and promised, “You'll have your money.” Manning thus became ECD’s largest shareholder.
Losing Control of OIS
As a condition for his investment, however, Manning required Ovshinsky to sign an agreement that soon became the focus of an intense legal dispute. Manning claimed to be concerned with Ovshinsky’s management of ECD, which had repeatedly lost money and, to reduce his control, he sought to take away Ovshinsky’s loaded vote, which at that point stood at 25 votes per Class A share.11 The claim seemed plausible; others also complained about ECD’s unprofitability. “Why do companies keep giving money to Stan Ovshinsky,” Forbes magazine later asked, “the inventor who can create anything but profits?”12 Yaniv put it more admiringly: “Stan was the Robin Hood of scientists. He was taking money from the rich people and hiring three hundred to five hundred scientists. No one else in the world did this.” Investors like Manning would complain that Ovshinsky chose to plow all of ECD’s profits back into research instead of paying dividends to shareholders. But ECD didn’t actually have the profits with which to pay dividends. Since it was not a mature existing business with regularly recurring revenue, there was no way to quantify the future benefits of its research. Conservative accounting rules therefore required treating the company’s large research costs as expenses instead of additions to its capital. Ovshinsky’s ambitious business model with its several concurrent and interdependent R&D programs meant that research expenses almost always exceeded revenue, so ECD seldom showed a profit.
It soon became clear, however, that Manning was actually attempting to take over the whole of ECD, and when Ovshinsky refused to comply with the agreement, Manning sued to enforce it and gain control of the loaded vote. ECD filed a countersuit for violation of the 1934 Securities Act. The ensuing arbitration struggle ended with a “divorce” settlement in which Manning agreed to sell all his ECD stock and not buy more for ten years, but in return he gained all of OSMC and a controlling share of OIS.
Ovshinsky had managed to prevent Manning’s attempted takeover, but at a steep cost. Instead of gaining new funding through OIS, ECD was left with even less resources and had to make painful sacrifices. The company downsized from close to five hundred employees to about a hundred, reducing the staff so severely that it was difficult to function. In the midst of this terrible time in 1987, Ovshinsky had heart surgery in New York City at St. Luke’s hospital, where his daughter Robin then worked.13
While in the hospital Ovshinsky handled some work from his bed, mostly by phone. One piece of business was negotiating a license with the Korean company Samsung for OIS’s hand-held displays, thus initiating Samsung’s entry into the television display market.14 Stan bitterly recalled licensing it “for nothing because we were under a lawsuit.”
Meanwhile, under Yaniv’s direction, OIS had been working on making larger flat panel displays with amorphous silicon diodes, but when the displays grew larger than six inches they suffered serious problems. Moreover, because competitors like IBM, Matsushita, and Mitsubishi were all using transistors, Yaniv explained, ECD would have had to develop its own production capability for diode-driven displays, an extremely expensive proposition. At the time, Yaniv was also negotiating for support for OIS from Sharp, which required them to work on transistors as well as diodes. Quietly, while most of the OIS staff worked on diodes, Yaniv had one researcher, Mohshi Yang, experiment with thin-film transistors (TFTs); he demonstrated “a superb three-inch color TV,” Yaniv recalled.
When Samsung and the avionics industry (the market OIS was mainly aimed at) offered significant support to develop TFTs, that tipped the balance. On his own, Yaniv decided to change from diodes to transistors, announcing his decision in 1988 at a crucial management meeting with OIS officers. As he expected, they were dismayed. “These people developed the diode with me from time minus ten, and to come to a point where I gave up for commercial reasons, they couldn't understand it. I remember their faces.” The change from diodes to transistors may have been necessary and inevitable, but it destroyed Ovshinsky’s patent advantage. (OIS held the patents for switching with diodes, while using transistors was in the public domain.) Ovshinsky never forgave Yaniv for making the change.
Even though OIS had given up the advantage of using diodes, it still held patents for the design of active-matrix LCDs that would soon prove to be of value.15 In July 1989 Yaniv negotiated a second, far more lucrative licensing agreement with Samsung, this one for $2.5 million, ten times larger than the initial license that Ovshinsky had approved from his hospital bed in 1987.16 By this time OIS was virtually independent of ECD. Aiming to making a prototype three-inch color TV, Samsung sent personnel to Michigan to be trained by OIS staff. But the Samsung team behaved strangely, Vin Canella recalled. “They believed that we were trying to cheat them. We’d find people rummaging through the dumpster, trying to find the secret papers. We were very open and honest with them.” Marv Siskind also remembered seeing the Samsung visitors searching the dumpster “for any of our paper that we threw out.” The explanation for this strange behavior eventually became clear: the team from Samsung were television, not semiconductor, people. They only knew about cathode ray tubes, and when they returned to Korea and tried to use what they had learned at OIS, the devices they built failed. Only after an independent review by a Japanese TFT display expert did Samsung recognize that the failure did not result from OIS’s withholding information but rather from their team’s inappropriate background. Once Samsung replaced them with semiconductor people, they began making displays successfully.
From that point, Samsung went on to make ever-larger displays, eventually becoming the world’s largest manufacturer of TVs and LCDs. Yet something of Samsung’s collaboration with ECD and OIS remained embedded in its display design; much later, when the electronics systems designer Guy Wicker looked inside a Samsung display, he found that it retained the same pattern of connecting the transistors and other components that OIS had developed. And when the computer scientist and entrepreneur Tyler Lowrey visited Samsung and asked his hosts how they were able to develop their huge liquid crystal displays, the reply, as Lowrey later told Ovshinsky, was “You wouldn’t know a man named Stan Ovshinsky, in Detroit? We got our license from him.”17
By early 1991 OIS was seriously underfunded again and sought a partner with deep pockets. “By this point OIS is not doing Stan any good,” Canella explained, and he was “looking to dump the company,” for while ECD still owned a substantial share of OIS, the settlement with Manning had deprived Ovshinsky of so much control that he had no reason to continue. The successful business leader, William (Bill) Morse Davidson, owner of the Detroit Pistons and then among the richest men in Michigan, was an interested buyer. His company, Guardian Industries, was a large architectural and automotive glass manufacturer, and as Yaniv said, he believed it could make “anything built on glass.” Davidson purchased OIS and immediately butted heads with Yaniv. “Displays are not just pieces of glass, just as a microelectronic chip is not just a piece of silicon,” insisted Yaniv, who also let Davidson know that he believed he was underestimating the competition.18 “He thought that because Guardian Industries competed with Asia in making glass, they could also compete with Asia in making displays.” Yaniv also felt that Guardian had an exaggerated view of the size of the market and disagreed with Davidson’s plan to build a huge factory to make avionic and military displays.
The result of this unwelcome advice was that Yaniv was asked to step down the day after Guardian took over, though he remained a paid consultant for the next two years.19 Davidson went on to build his state of the art facility in nearby Northville, the first large-volume LCD plant in the United States. But when the Japanese and Koreans invested tens of billions in their TFT active matrix industries, Guardian’s displays were too expensive to compete. (Indeed, no American display maker could compete.) Guardian OIS was bankrupt by September 18, 1998.
There was one more chance for ECD to play a role in the display industry. Roughly a decade after ECD had licensed its technology for LCDs to Samsung, Tatung, the largest Taiwanese electronics company, was having trouble scaling up its thin-film transistors. Guy Wicker and Rosa Young convinced them to try using Ovshinsky’s threshold switches instead of TFTs. Representatives from Tatung who visited ECD were ready to offer funding to develop prototypes, but Ovshinsky disagreed with their plan to start small and work within a two- or three-year timeline. “He thought the timeline should be compressed to a much shorter time,” Young recalled, and nothing came of the discussion. “I was really very unhappy with Stan’s decision,” she said. It was yet another missed opportunity.
Phase-Change Memory
The creation of phase-change memory in the 1960s is arguably the most important invention of Ovshinsky’s career (see chapter 5). But as Marv Siskind said, “The world wasn’t ready for it.”20 During the 1970s, after the failure of the West Coast memory company OMI that Keith Cunningham had started (see chapter 6), ECD’s phase-change memory research languished because of inadequate funding. In the late 1970s, however, funding from IBM for imaging technologies had supported a small program in optical memory that allowed ECD to continue work on the materials.
In the mid-1980s the optical program got a boost from a lawsuit settlement. Matsushita, the largest Japanese semiconductor company, had introduced a rewritable optical memory, which they claimed was their invention. Angered by the infringement of his patents from the 1960s and 1970s, Ovshinsky sued the giant Osaka-based corporation in May 1983. With Momoko Ito handling the negotiations, the suit was settled out of court. She also persuaded Matsushita to collaborate with ECD on optical memory. ECD got $1.5 million for a two-year development program; Matsushita got a license to use the technology developed both before and during the program. It was not a large settlement, but Ovshinsky was satisfied to be acknowledged as the inventor of the technology.
Ovshinsky divided the optical memory work between groups headed by Dave Strand, who had been leading the program since 1980, and Rosa Young, who had recently joined ECD. Strand’s group focused on the basic physics of the materials, building and operating test equipment, while Young’s group focused on developing new materials and tailoring the properties of the phase-change alloys using Ovshinsky’s principle of chemical modification, which had been developed in the early stages of the photovoltaic program (see chapter 6). She was assisted by Eugenia (Genie) Mytilineou, a professor of physics at the University of Patras in Greece, who often worked at ECD in summers and during her sabbaticals, and who would become one of Young’s closest friends.
By systematically changing the recipe for the chalcogenide materials, Young’s group managed to increase the speed of the optical memories and improve the cycle life. In Strand’s group, work on recording and erasing led by Mike Hennessey significantly improved the process and resulted in a patent that was licensed to Matsushita and others.21 Both Strand and Young recalled how smoothly the collaboration with Matsushita went, and as always, Ovshinsky kept closely in touch with their progress and kept up the pressure.
When the two-year Matsushita program ended in 1986, Ovshinsky continued to fund the optical memory research. By 1988 the work had paid off with the development of the much-faster 225 alloy (consisting of germanium, antimony, and tellurium, Ge2Sb2Te5). With its repeatable switching time of 50 nanoseconds or less, it was around a thousand times faster than the earlier alloys, and required less energy for switching.22 Additional work done primarily by Japanese companies greatly increased the storage density.23 As a result, rewritable optical memory discs (CD-RW and DVD-RW) based on technology from ECD were widely used in the 1980s and 1990s and are still in use today. But while ECD received about $1 million in royalties from several Japanese licensees, when production later moved to China, the relatively low return on the discs made it impractical to enforce ECD’s patents.
Meanwhile, in 1985, while the Matsushita collaboration was still going on, Ovshinsky took steps to revive ECD’s electrical phase-change memory program.24 Named NGEN (Next GENeration of computers), the new program received funding from two Japanese companies: $1.2 million from NTT (Nippon Telegraph Technology) for 3D phase-change memory development and $4.5 million from NSC (Nippon Steel Corporation) for threshold switching logic. Both companies discontinued their support when the team’s effort to make a 3D memory failed, but ECD now had a clean room with deposition and lithography equipment that would be used continuously for further development of the threshold and memory switches.
In the mid-1980s, however, electrical phase-change memory had two fundamental shortcomings: it required too much current to switch, and it wasn’t fast enough. It also faced the formidable competition of flash memory, which had been introduced in 1984.25 The solution to increasing its speed came from the fast 225 alloy developed by the optical memory group. Ovshinsky was excited when it showed 50-nanosecond optical switching speeds and believed it would work even faster in an electrical memory. Many of his advisers disagreed, maintaining that the 225 alloy was a different class of materials and would not have suitable electrical behavior. But Ovshinsky insisted on trying it, so Wally Czubatyj, who by 1988 had become manager of the electronics group, assigned Pat Klersy and Dave Beglau to make the electrical devices using the 225 alloy. Guy Wicker, assigned to test them, was amazed to find he could set them with a 10-nanosecond pulse. Wicker emphasized, “It was Stan who motivated the use of fast optical alloys for electronic memory. Everyone thought he was crazy for insisting on it, but it was the biggest single improvement in the memory in more than twenty years.”26 It was another instance—and one of the most important—of Ovshinsky’s strategy of cross-fertilization among research programs. By 1989 the electronics group had developed a good working model of a three-dimensional electrical phase-change memory.27
Now Ovshinsky was ready to commit more resources to the program. On a sunny New Year’s Day, January 1, 1990, as Ben Chao recalled, Ovshinsky held a special meeting for roughly twenty of ECD’s scientific staff in his home to announce the new effort. He told them that he wanted to make electrical phase-change the next memory device and predicted a time when all computers would use it. At this point, the optical memory group and the electronics group, which had been working primarily on threshold switches, merged. Initially Dave Strand and Czubatyj led the new group. Strand brought familiarity with the materials, and Czubatyj brought familiarity with fabrication.28
For about three years the group worked on improving the electrical phase-change memory. The working model was fast, but it was too large and still required too much current. Klersy and Beglau processed wafers in the clean room, trying to make the devices smaller. While developing an insulator etching process, they inadvertently left a residue of carbon and fluorine polymer that broke down after a single pulse. The resulting memory bits needed two orders of magnitude less current. While they had managed to produce a device that could compete in speed and current with existing ones, it took some time to learn how to produce it reliably. By 1993, ECD had developed a consistent process that clearly showed the potential for making a competitive memory device.
Ovonyx (Ovonic Unified Memory)
At this point, the commercialization of electrical phase-change memory required more funding and a larger network of associates. In 1994, during Strand’s continuing efforts to raise money for the program, he made a cold call to Micron Semiconductors in Boise, Idaho, then the largest DRAM (dynamic random-access memory) manufacturer in the United States.29 Strand was surprised when Micron’s chief scientist, Tyler Lowrey, answered the phone himself, and he was absolutely elated when Lowrey expressed real interest in ECD’s electrical phase-change memory.
Well-known in semiconductor circles for developing more efficient and lower cost DRAM fabrication processes without lowering the pay for labor, Lowrey was “a visionary in his own right,” remarked Steve Hudgens. He had “the ear of the semiconductor industry,” said Guy Wicker, and could easily get an audience with senior people at leading firms. Because of his deep knowledge of semiconductors, Lowrey understood the importance of nonvolatile memory. As he later explained, “When you turn off the power on your PC you’ve got to reboot it every time because it’s got to bring in all the programs from the hard drive. You want an instant on. You want to be able ten years from now to turn it on and have it be right back where it was. Plus if you crash you want the data to be stored, not gone.” Lowrey also understood that flash memory, the computer industry standard, would eventually run into problems because it could only be scaled down to roughly fifty or a hundred stored electrons. Chalcogenide phase-change memory does not have that limitation; indeed, since it needs less power it functions better as it scales down.30 Moreover, it can be cycled many more times than flash memory, which wears out after some fifty thousand rewrite cycles. The many advantages of phase-change over flash memory also include its roughly a hundred times faster speed, greater efficiency, lower power, hardness to radiation (allowing the memories to function in space and in military applications) and greater potential for use with processors.
Lowrey recalled the original demonstration of the ECD devices at Micron. “We marched on down to the lab and we ran them, and we said, ‘Man, these things work.’ It was about a million [write/erase] cycles in a second, which was more than the lifetime of a flash memory. We signed a deal right on the spot. It was like a one-page agreement.” Micron subsequently finalized it as a joint venture at a meeting that included Micron’s CEO Joe Parkinson, Stan and Iris Ovshinsky, Marv Siskind, and Dave Strand. Once the joint venture began, Micron’s advanced fabrication lines were used to produce devices a hundred times smaller than could be made with ECD’s simpler equipment.31 But after two years of making progress, Micron faced a crisis when the price of DRAM, its main business, plummeted. It could no longer support the joint venture and pulled out.
Figure 10.1 The Ovonyx team in 1999. Left to right: Sergey Kostylev, Wally Czubatyj, Tyler Lowrey, Steve Hudgens, Pat Klersy, Boil Pashmakov, Guy Wicker.
Lowrey, however, continued to be enthusiastic about phase-change memory. When he left Micron in 1997, Ovshinsky invited him to join ECD. For eighteen months he was restrained by a noncompetition clause, but “the day after the eighteenth month was over, I was back in Michigan,” he recalled. By the end of 1999, he and Ovshinsky had formed a separate company to develop and commercialize electrical phase-change memory. They called the new company Ovonyx, short for Ovonic Unified Memory (OUM), a name chosen because of the memory’s versatility: “It can be used optically, electronically, thermally, whatever,” Steve Hudgens explained. The new company was created as a joint venture between Lowrey and ECD.32 Lowrey hired all the people at ECD who had been working in phase-change memory and were willing to join him, including Steve Hudgens, Guy Wicker, Boil Pashmakov, Sergey Kostylev, Patrick Klersy, and Wally Czubatyj, “Everybody was very happy,” said Pashmakov. “Tyler was the biggest name in flash memory worldwide,” and he knew what ECD didn’t know yet, “how you actually make a memory chip.”33
Ovonyx initially aimed to survive on the fees for licensing the technology to as many semiconductor manufacturers as possible. In these arrangements, it would be Ovonyx’s job to do R&D, while the licensees were to commercialize the technology. In November 1999, Ovonyx formed a funding agreement with Lockheed Martin Space Electronics and Communications (now British Aerospace, BAE), and in February 2000 with Intel.34 Part of the team began work in a Silicon Valley Intel facility in Santa Clara.35 “Intel gave us a laboratory and all the money and resources we needed to develop this,” recalled Wicker. And once Intel got involved in Ovonyx, other electronics firms wanted to follow suit. Among those who took licenses was the major Italian flash memory supplier, ST Microelectronics. When the dot-com bubble burst in March 2000, however, many companies were “scrapping to survive,” Lowrey recalled, and could not afford to invest in new projects. Ovonyx spent this time improving the technology, working with its partners.
In 2001, ECD closed its Troy headquarters and moved into much larger facilities in nearby Rochester Hills, but as a result the Michigan-based Ovonyx team no longer had a clean room. Ovshinsky took responsibility for providing a state-of-the-art fabrication facility, which he had Klersy design with a $7 million budget. Instead of charging Ovonyx for the use of the facility, Ovshinsky used it to renegotiate his agreement with Lowrey, regaining some of the intellectual property that had been transferred to Ovonyx involving neural network applications of the threshold and memory switches, property that became the basis of the cognitive computer (discussed next). Upon completion of the new clean room, most of the Santa Clara Ovonyx staff moved back to Rochester Hills, leaving Hudgens to manage the California program.36 The new clean room became Ovonyx’s main asset, enabling the company to build its patent portfolio.
When the market started to recover after 2002, Ovonyx signed up other licensees, many joining as a protective move. In 2005 Ovonyx added the Japanese memory company ELPIDA (created by Hitachi and NEC). About a year later, Samsung, “the biggest memory supplier in the world,” also signed up, and by 2007 Ovonyx had between ten and fifteen different licensees, “sort of all but Toshiba,” Hudgens said. The hope was not only to replace flash memory, then a $10–$12 billion yearly market, but also to displace DRAM, whose market was estimated at $25 billion a year.
Excited by the development and prospective commercialization of his phase-change technology, Ovshinsky interacted with both Lowrey and Hudgens as much as possible. He was impressed that Lowrey “follows me, where other people I lose.” In turn, Lowrey continued to bring questions to Ovshinsky, finding him “still sharp as a tack.” But as Joi Ito noted, only “the really smart people” were able to understand Ovshinsky and work with him.37
A continuing problem was that the corporate giants were pouring billions of dollars into research to improve flash memory, which was also getting cheaper. It was “like trying to jump on a moving train,” Hudgens remarked. Ovonyx researchers knew that phase-change is a superior nonvolatile memory, and they believed that it would eventually replace silicon-based flash memory. Cell phone manufacturers especially liked phase-change memory because battery-operated cell phones need nonvolatile memory with low power requirements.38 But the fact that phase-change memory was better than flash in many ways made no difference because, as Hudgens observed, “the marketplace wants cheap and good enough,” and for more than two decades flash had been cheap and good enough. Only the few who were committed to phase-change memory opted to wait in frustration, and more or less in limbo, for the time when MOSFETs (metal-oxide semiconductor field-effect transistors) had scaled down to their limit, a time they thought might not come for three or four decades.
Envisioning a Cognitive Computer
While Ovonyx was struggling to commercialize electrical phase-change memory, Ovshinsky worked with a small team to develop an even more ambitious information technology based on chalcogenide switching. Exploring new possibilities, it also drew on the whole history of his information work: his efforts to probe the nature of human and machine intelligence, his nerve cell studies, his Ovitron, threshold, and memory switches. All culminated in his attempt between 2003 and 2007 to build what he called his cognitive computer.
Ovshinsky had thought for years about expanding his nerve cell studies to model a synapse, and perhaps eventually a human brain. The possibility of actually doing that emerged when Boil Pashmakov discovered that the energy of the set pulse triggering the phase-change memory device could be composed of a number of lower energy pulses. That enabled cumulative memory storage. Just as the neurons in the brain collect pulses from many inputs and fire when their sum reaches a threshold, the cognitive devices in the computer would accumulate information from different inputs, and when these added up to the threshold for switching from amorphous to crystalline, the device would “fire.” Ovshinsky saw in the cognitive computer yet another physical confirmation of his belief that energy and information are two sides of the same coin. As he put it, “You’re adding energy, adding energy, adding energy, adding energy, and you’re making little crystallized regions. And when these regions connect to form a percolation path, then it fires just like a nerve cell.”39 Combinations of these devices would offer a capability for parallel processing like the processing of the human brain, which similarly sums up the information it receives.
The key to modeling this threshold feature was to make the Ovonic memory such that it didn’t have to be in either an “on” (conducting) or “off” (nonconducting) state, corresponding to the “one” or “zero” states in an ordinary binary computer.40 Instead, the memory could have many intermediate states. For example, if the pulse packages each consisted of 10% of the energy needed to switch from the amorphous to crystalline state, one would need ten pulses to reach the threshold, or if the packages were twice as big, one would need only five pulses to reach the percolation limit. This opened up the exciting possibility of performing arithmetic calculations within a single nanostructure. It also allowed encryption, for if you had a device requiring ten successive pulses of a particular width and height to produce a percolation path and you wanted to store a three, you would put three of these pulses on the device, and to read the three would take seven pulses. But if you found that it took six pulses then you would know that a four had been stored there. Only a person who knew how the encryption worked in a particular device could read the encrypted information, because having the wrong pulse width and height would destroy the information already there. The cumulative memory of the cognitive device also enabled storing different intermediate resistance states by applying different intermediate size pulses. That offered the possibility of a multi-bit memory in a single device, which would not only greatly increase memory storage density but also allow the development of new computer architectures: instead of binary logic, they could use decimal, hexadecimal, or other bases.
Another device that Ovshinsky and Pashmakov developed to provide enhanced capabilities for a cognitive computer was the three-terminal threshold switch, sometimes called the Ovonic Quantum Control Device. In this device a signal applied to the third terminal, analogous in function to the third terminal of a transistor, could change and control the threshold voltages at which the cognitive device would fire. Invented during the summer of 2006, the three-terminal device was designed to work in the same circuit with the cognitive device as part of a chain in which the output of one device affects the input of the next. Pashmakov led the development but, Dave Strand emphasized, “the inspiration and the direction on how to make it came from Stan.” As Pashmakov explained, in a normal computer “the processor has to interact with the memory all the time to retrieve data or send data to memory, and that’s what slows down computers.” Having everything in the same location eliminates such delay, and the greatly increased speed allows for many more sophisticated applications, like pattern or voice recognition, that require very fast processing of large amounts of data. “We’ve found that the device actually has the functionality of an artificial neuron,” Pashmakov said.
As Ovshinsky envisioned it, the cognitive computer would first perform relatively simple cognitive functions, such as pattern recognition, before progressing to more-sophisticated functions like inference. He looked forward to integrating groups of circuits to create an analog of human intelligence. He would emphasize that all his information inventions were only models, not actual synapses or brains, but he believed that “evolution is going to have to happen,” and that his cognitive computer could evolve into a device “that could make decisions and could learn” based on its interactions with external devices. The most advanced version of the cognitive computer that Ovshinsky achieved had sixteen linked devices, which he liked to call “synapses,” but he remained certain that in time he would be able to build a machine with a thousand or more synapses, approaching the cognitive power of the human brain. “Will it have built-in consciousness? No, but it will have a great deal of intelligence,” Ovshinsky said.
Once again, Ovshinsky’s ability to see the potential of something that no one else would consider remarkable was the key to developing the notion of the cognitive computer. Pashmakov recalled Ovshinsky’s excitement when he showed him that the device took two or more pulses to crystallize, recognizing that this “was how a neuron works.” Ovshinsky remarked, “That is what I always thought it was going to do.” He added, “We have to make a patent.”41
Progress was slow, however, because the small ECD cognitive computer team working between 2003 and 2007 was badly underfunded and understaffed, consisting mainly of Ovshinsky, Strand (who acted as the group’s manager), and Pashmakov (who carried out much of the experimental work).42 The biggest problem, however, was that designing such a computer on a large scale required developing a new architecture different from the von Neumann paradigm for binary computers. While Ovshinsky and his team took some steps toward this end in the early 2000s, they did not solve the problem. Strand later summarized the work on the cognitive computer as providing “really good bricks to build a house. But if you didn’t have a plan to build a house, you couldn’t get there with just a pile of bricks.”
Despite its innovative concepts and the promising steps taken toward realizing the cognitive computer, the research did not progress beyond its early stages. Perhaps because Ovshinsky was now eighty-three, he had trouble interesting others in the invention. When he called a press conference on June 6, 2006, to announce the new Ovonic Quantum Control Device, “one guy showed up.” Less than a year later, ECD’s new board would abruptly terminate the cognitive computer program (see chapter 11). Ovshinsky had hoped to continue the project when he set up Ovshinsky Innovation in 2007 (see chapter 12), but he never had the time or funding to do so.
