    |
 |  |  |  |  |  |  |  |  |  |  |  |  |
Texas Instruments: a DSP successby David Lammers Back in 1977, after having conquered the calculator — temporarily — the consumer products group at Texas Instruments was searching for another mountain to climb. Paul Breedlove suggested a Speak and Spell product for children. As Breedlove took the proposal around to TI's notoriously tough-minded management, it was considered "so far-fetched" that it barely qualified for what Gene Frantz, a TI senior fellow who managed the system design on the Speak and Spell project, described as "wild hair money."Richard Wiggins, an engineer who was brand-new to TI at the time, took a look at Breedlove's idea and pronounced: "I think we can do it." And they did: 18 months later, the product was a Christmas season hit. The Speak and Spell product included a 4-bit microcontroller, two 128-kbit ROMs and a speech synthesis chip (TMC0281) that paved the way for the single-chip DSPs that are the center of TI's semiconductor business today. TI's success in the DSP chip market is built on its early work in oil exploration and defense-related signal processing. In the IEEE-published book Signal Processing, The Emergence of a Discipline that appeared in 1998, historian Frederick Nebeker interviewed Enders Robinson, a professor at the Massachusetts Institute of Technology. Enders said offshore oil exploration in the early 1960s led to the first major use of digital techniques, as it became clear that analog signal processing could not filter the frequencies bouncing off the ocean. The movement to digital signal processing came "when Texas Instruments, which was Geophysical Services Inc., said 'give us your exploration records from the Gulf of Mexico and we'll remove the reverberation by deconvolution.' They showed the results, and everyone flip-flopped" from analog to digital signal processing, Nebeker quotes Enders as saying. Harvey Cragon, a computer architect who retired from TI in 1984, recalled that "the first TI computers processed analog signals, modified raw seismic data and generated visual outputs for interpretation by human operators. Mark Smith, a graduate of MIT who studied under Norbert Weiner, joined TI in 1954 and began studying the application of digital signal processing to the seismic data-reduction problem. A hybrid analog-digital computer, called seisMAC, was designed by J. Fred Bucy and others in 1955." The oil companies had money, Enders noted, and that helped fund much of the basic filtering and transform algorithm research. Panos Papamichalis, director of the TI Tsukuba R&D center in Japan, said one example of such research is the work John Burg, who developed the Burg time series analysis method as part of his early work in seismic signal processing techniques. Submarine detection and other wartime research projects fueled the mathematical and engineering underpinnings of today's DSP industry.
"But the big evolution came after the Speak and Spell came out," Papamichalis noted. "The single chip vocoder there was a big breakthrough. I was finishing up my PhD at the time at Georgia Tech, and I remember the big panic we all felt when we read about that chip. It really paved the way to the single-chip DSP." Cragon recreated the meetings and debates that preceded TI's first single-chip DSP in a fascinating article, "The early days of the TMS320 family," published in 1996 in the TI Technical Journal. After designing computers that used TTL circuits to perform signal processing, Cragon went to a TI strategic planning conference in 1979 and showed how a dedicated DSP (then called an SPC, for Signal Processing microComputer) might be created. "The design goal was to define a leadership product that would apply the low cost of VLSI microcomputers to commercial applications. All trade-offs were to favor digital signal processing," he said. It is interesting to note that the specification did not include a multiply-add instruction, which is considered fundamental to today's DSPs. "The reason for this design decision is a little hazy to me today as the processor had both a multiplier and adder that could operate concurrently," said Cragon. "I now believe that specifying two addresses in a single instruction would have widened the instruction ROM and the program bus, increasing the chip area and presenting difficult routing problems." Bob Hewes, a vice president and director of the DSPS R&D Center in Dallas, participated in the early meetings that Cragon describes. Hewes was asked to study the possibility of implementing the design in CMOS at 2-3 micron design rules, rather than NMOS. Hewes said in an interview that the TMS32010-which was presented at the 1982 ISSCC conference-was implemented in 5-MHz NMOS, and the 320C25 moved to 10 MHz in CMOS technology. Hewes said AT&T's desire to digitize the phone network played an important part in the evolution of DSP work at TI. Researchers were building hard-wired signal processors to filter voice signals, and TI applied programmable DSP technology that proved beneficial. As DSPs began appearing more frequently in repeaters, as a way to amplify voice signals, in echo cancellation and in modems, the communications industry began to see wider uses for single-chip DSPs, recalled Papamichalis, who serves as the president of the IEEE Signal Processing Society for 2000-2001. It was in the mid-1980s that the modem "went digital." In Houston, TI's DSP researchers realized that "at 1,200 bits per second, using analog became too tricky. We applied DSPs then, and at 2,400 baud it became more realistic. Companies went digital, and then programmable digital." Houston was the hub Hewes, who worked in Dallas at TI's central research lab, said "it would not be fair to say that TI's DSP chips were developed in our R&D lab. The MPU designers in Houston developed the requirements, and the architecture was done in Houston." TI's relationship with universities has played an important role in the past 15 years, both with key academics in the field, who wrote textbooks about DSP technology, and with students who learned how to apply DSPs. A group of professors at Georgia Institute of Technology had created a company, Atlanta Signal Processing Inc., to apply DSP technology, and TI engineers "talked with them a lot, and we used some of their ideas for our chip development projects," said Papamichalis, who earned his PhD there in 1978. Ron Schafer, Tom Barnwell, Jim McClellan and Russ Mersereau developed DSP technology, and sent dozens of their students to work at TI. Ray Simar, a DSP architect at TI, credits Rice University in Houston with a major role, too. Rice professors, including Sid Burrus, now the dean of engineering at Rice; Tom Parks, who moved north to Cornell University to teach, and others wrote three books about DSPs. "The university programs played a much bigger role than many people realize," said TI senior fellow Frantz. "We realized very quickly that DSP had to be moved to the undergraduate engineering curriculum, lower than the doctoral level, where it was at the time." Papamichalis said many of the best speech-processing algorithms were developed by customers that influenced TI's early instruction-set architectures and memory configurations. That openness to customer suggestions paid off in spades in the cell-phone market, when the C54X was designed to meet the needs of specific cell-phone customers in Europe. "Talking to customers was so important in the early days, and still is. Back in 1985, we had several key meetings where customers would say, 'Ah, it would be nice if we could do this or that.' That kind of interaction resulted in instructions for adaptive filtering, such as bit-reversed addressing, that saved us extra cycles," Papamichalis said. Frantz said much of TI's DSP success has come because the company recognized the importance of the "nontechnical" parts of developing the technology. Funding the publication of books, working with development-tool partners and searching for new applications continually are the keys to TI's success. The Century of the Engineer: Companies that made a difference
|
||||||||||||||||
RSS
 |
Home | About | Editorial Calendar | Feedback | Subscriptions | Newsletter | Media Kit | Contact | Reprints|  RSS| 
Digital|  Mobile |
|
| Network Websites |
|
|
International |
|
|
Network Features |
|
|
|
All materials on this site Copyright © 2010 TechInsights, a Division of United Business Media LLC All rights reserved. Privacy Statement | Terms of Service | About |
