Detailed history of Qualcomm
- Transfer
Considering the recent dramatic events that have happened to Qualcomm, I decided that it would be interesting to update the company's history, which will be published in the Mobile Unleashed book. Publish the full chapter from the book here.
Unlike other similar firms that started work in other segments of electronics, and then transferred to the field of devices for communications, Qualcomm has always focused on wireless technology, reliably transferring data between two points. Its CDMA technology has become a leap forward for mobile devices - if it could be made small enough, and if it were possible to convince providers to switch to it with D-AMPS and GSM.
NASA Jet Propulsion Laboratory
The roots of Qualcomm can be traced to one of the smartest people in the academic environment at leading technological universities in the United States who, as young engineers, came together for the first time to work on a space program. The depth of technical knowledge gained from satisfying the needs of demanding customers in digital data transmission systems has created a foundation on which patents, chips and devices then appeared.
The fruitful article “Mathematical Theory of Communications”, published by Claude Shannon from Bell Laboratories in 1948, laid the foundation for information theory. Together with the invention of the transistor and the advancements of digital programming and computing, Shannon's theorem and his work at MIT inspired a whole generation of mathematicians and scientists.
In June 1957, Andrew Viterbi, a MIT graduate with a master’s degree in electronic engineering, joined the Jet Propulsion Laboratory (JPL) in Pasadena, pc. California. At that time JPL was owned by the California Institute of Technology, but it worked under the auspices and with the money of the US ballistic missile agency.
Viterbi worked in Section 331 for communications led by Solomon Golomb. They developed telemetry for rockets and satellites. Golomb pioneered the theory of shift registers with linear feedback, which was used to encode digital messages for reliable transmission at high noise levels. Viterbi was working on closed phase loops, an element of this theory, critical for synchronizing a digital radio receiver with a transmitter — this was required in order for the information flow to be decrypted.
October 4, 1957 the USSR launched the " Sputnik-1"The next day, the Mylston-Hill radar belonging to the Lincoln Laboratories at MIT (MITLL) - where Irvin Reed, the most famous Reed-Solomon code reader, worked - discovered the Satellite in low orbit. amateur radio to track the satellite signal, which has grown and faded away every 96 minutes, according to the orbit movement of the satellite.
Space race has begun. The US Navy began to hurry with the response of their project "Avant-garde". On December 6, 1957, Test Vehicle 3 was launched, the third test vehicle with a satellite weighing 1.3 kg. He climbed the disgraceful 1.2 m, lost his cravings and fell back onto the launch pad, exploding. The payload landed nearby, in the bushes of Cape Canaveral, but did not stop the broadcast. “This is our competitor,” said Golomb.
On January 31, 1958, the Project Deal, known worldwide as Explorer 1, reached orbit. Life magazine posted a photo of Golomb and Viterbi in the JPL flight control room on the cover. On July 29, 1958, President Eisenhower signed a decree on national aeronautics and space flight, creating NASA. The JPL requested and received a transfer under the auspices of NASA in December 1958.
Viterbi enrolled at the University of Southern California (USC) to defend his doctoral thesis - this was the only institution that allowed him to continue working at JPL for a full day. He graduated in 1962 and went to teach at the University of California, Los Angeles (UCLA). He recommended Golomb to join the staff of USC faculty, where Reed was already (who joined Santa Monica’s Rand Corporation in 1960), Lindsey (joined JPL in 1962), Eberhart Rechtin, Lloyd Welch and others.
Many years later, Lindsay joked, "I think this group was created by God." Rehtin said that as a result of joint work, this group was able to do more in the field of digital communications than any of them could have done alone. Their work has influenced countless other people.
In 1963, at the national electronics conference in Chicago, the award for the best work went to Viterbi and Irwin Jacobs, a professor from MIT, whose office was located next to the office of Claude Shannon. Jacobs and Viterbi had already met in 1959, when Jacobs came to JPL for an interview, and each of them knew about the work of the other through the links of JPL and MIT.
At the next meeting at the conference in 1963, Jacobs told Viterbi that he would soon have his sabbatical and asked if it was interesting to work at JPL. Viterbi assured that the way it is. Jacobs was denied a request to take him to work, but Viterbi put in a word for him to the head of the unit Rehtin, and Jacobs eventually was hired as a researcher and sent to Pasadena. Viterbi taught at UCLA and advised at JPL, and the two became friends while Jacobs worked at JPL from 1964 to 65.
Having published the important for the history of the work "Principles of communication technology" together with John Wosenkraft in 1965, Jacobs moved to the west coast in 1966. One of his teachers in Corneille, Henry Bucker, persuaded him to join the new engineering department at the University of California at San Diego ( UCSD). Professors were then appreciated, and digital communications consultants were also in demand. One day, in early 1967, Jacobs went to a conference at the NASA Ames Research Center. On the way home on the plane, he found himself flying with Viterbi and another MIT graduate, Len Kleinrock, who joined the UCLA staff in 1963 and made friends with Viterbi. They began to talk, and Jacobs casually noted that he had accumulated more consulting work than he could handle.
Viterbi was finishing his masterpiece. He was looking for ways to simplify the theory of isolating weak digital signals from strong noise — so that his students at UCLA would have a better understanding of it than in the complex course that existed at the time. He came up with a general concept in March 1966, and refined the idea for a whole year before publishing. In April 1967, Viterbi described his approach in an article in the IEEE Transactions on Information Theory magazine entitled “Convolution Codes Error Bounds and Asymptotic Optimum Decoding Algorithm”.
Viterbi Algorithmpromotes "soft" solutions. The hard decision about whether a signal is zero or one can be made by observing each noisy bit received (or a group of bits encoded into a symbol), with a high probability of error. Viterbi considered probabilistic information contained in possible state transitions, and is known based on how the characters are encoded by the transmitter. Sequence analysis of the received symbols and state transitions using the “add-compare-select” operation (add-compare-select, ACS) determines the path of maximum likelihood, and more precisely coincides with the transmitted sequence.
It was just a theory, or so Viterbi thought at first. The algorithm reduced the amount of computation and the number of errors compared to others, but it still needed to be performed in real time, and it was believed that for a sufficiently small percentage of errors it would take “several thousand registers”. This work was picked up by several other researchers, among whom it is worth noting Jim Massey, David Forney and Jim Omur. They were convinced of its optimality. Jerry Heller, one of Jacobs' graduate students at MIT who joined him in San Diego, worked at JPL. He decided to drive away a few simulations, and from 1968 to 1969, he discovered that Viterbi was evaluating his theory too pessimistically; A fairly good advantage was given only 64 registers. But for that time, it still was a rather large cabinet of computing equipment.
Entrepreneurial ideas associated with a consulting firm did not let Jacobs go. In October 1968, Linkabit was born with a share capital of $ 1,500 (each of the founders contributed about $ 500) and an address that coincides with the Kleinrock home address in Brentwood. Soon the offices moved to a building located in Westwood, near UCLA. First, Jacobs, Kleinrock, and Viterbi, who taught full-time, spent one day a week at their firm.
However, the company has more cases than expected. The first engineer hired by the company in September was Jerry Heller, which was soon followed by Andrew Cohen, Klein Gilhausen and Jim Dunn. Len Kleinrock retired for several months, doing his favorite project - setting up the first end nodes of the ARPANET network and sending the first message on it in October 1969. If you believe him, when he tried to return to Linkabit, he was immediately deployed, issuing as a dismissal a certain percentage of the value of the company. In the absence of Kleinrock, and in view of the fact that Viterbi did not want to move for a few more years, Jacobs moved Linkabit's office in Sorrento Valley, one of the corners of San Diego’s Golden Triangle, in 1970. After that, he hired office manager Di Koffman immediately after he finished studying in high school.
"Programming is dead." Several speakers spoke on this topic at the IEEE Communication Theory Workshop conference, which was held in St. Petersburg in 1970, pcs. Florida. Irwin Jacobs stood in the far corner of the room, holding a 14-pin DIP chip - a simple 4-bit shift register, possibly 7495 from the TTL family (transistor-transistor logic). "This is the current state of digital technology, it will allow us to create all of this."
In the early days, Linkabit was a think tank, not a hardware manufacturer. Her first clients were the NASA and JPL Ames Research Center, as well as the Naval Electronic Laboratory at Pont-Loma and DARPA. Linkabit research related to the decoding of Viterbi resulted in a messaging system for deep space that was used in the Voyager project and other programs. However, soon compact versions of Viterbi decoders and other signal processing equipment will make Linkabit and its follower legendary.
Heller and Jacobs introduced a 2 Mbps Viterbi decoder with 64 states and a depth of 7 in October 1971. It was based on a commercial decoder made for military satellites. Linkabit Model 7026, or LV7026, used about 360 TTL chips on 12 boards in a 19-inch case, and was 4.5 U (7.9 ") high and 22" deep. Compared with previous versions of the equipment involved in processing the Viterbi algorithm, and occupying several racks the size of each refrigerator, this was a breakthrough.
The speed of work was also a problem. Viterbi talks about an early attempt by Linkabit to integrate a single ACS-state decoder on a chip containing only 100 logic elements - it was an average integrated circuit, or MSI. In his words, such an attempt “almost bankrupted the company” due to several consecutive problems with suppliers. Almost bankrupt? This seems an exaggeration until we consider the then available alternatives to TTL. Based on the company's 1971 report and Magnavox's 1974 document, Linkabit was played with the fast-working, but very capricious, “emitter-coupled logic” (ECL) technology, trying to increase the clock frequency of critical areas. Many companies could not do anything with ECL. Viterbi did not mention specific names, but among the suspects are Fairchild, IBM, Motorola and Signetics.
The change of direction brought more success. Klein Gilhausen started playing with the concept of a Linkabit Microprocessor (LMP) microprocessor, microcode architecture implementing the functions of a satellite modem. Gilhausen, Sheffie Worboes, and Franklin Antonio completed the LMP prototype board using mostly TTL chips and commercial high-speed ICUs and LSIs by May 1974. She worked at a speed of 3 MIPS. She had 32 instructions and four software stacks, one for processing, and one for monitoring. It was partly RISC (before the appearance of such a thing), partly DSP.
Jacobs began writing code and promoting LMP, lecturing at MITLL and several other institutes, talking about the ideas underlying digital signal processing by the satellite modem. The United States Air Force invited Linkabit to demonstrate its technology on LES-8/9 experimental satellites. TRW had a head start in a few years to create an extended spectrum modem within the K-Band SATCOM AN / ASC-22 system, but their solution was expensive and terribly voluminous.
Linkabit struck the MITLL team by placing its relatively small system consisting of several rack-mount enclosures of 19 "each and setting it up for data transfer in just an hour - the lab staff would probably take a few days only to launch the basic mode. After another three hours they found an error in the MITLL specifications, corrected it by simple reprogramming, and set up data reception. And despite TWL certification and the readiness of its product, the general in charge of the program decided to finance Linkabit, a company that never produced equipment in the amount required for the defense industry - so that it finishes the development of its modem.
In addition to the excellent work of the LMP, the United States Air Force became interested in its other aspects, which became known in 1978. The real product requirement was the ability to install a dual modem [dual modem] on aerial platforms like the Boeing EC-135 and airplanes of the strategic command of the US Air Force, including the Boeing B-52. The solution, which gradually developed into a modem and a data processor for a command post (CPM / P), using several LMPs for dual duplex modems and transmission of control commands, as a result, fit into three strong 1/2 ATR boxes .
Linkabit grew by 60% per year. For the expansion of the company, additional capital was required, and they considered the option of selling shares, but then they received a proposal from another company engaged in radio technology, M / A-COM. In August 1980, the purchase of the company was completed. This radically changed Linkabit’s culture, and the free flow of ideas throughout the organization was replaced by a hierarchical process control framework. But this did not stop innovation. There were several important commercial products. One of them is a small satellite earth station Very Small Aperture Terminal (VSAT), a small business satellite communication system using plates with a diameter of 120 to 240 cm. Among the main firms that bought this technology were 7-11, Holiday Inn, Schlumberger and Wal-Mart. Another technology is VideoCipher, Satellite TV encryption system, which worked in HBO and other broadcasting corporations. Jerry Heller tracked the development and growth of VideoCipher technology throughout her life.
Jacobs and Viterbi discussed the acquisition of the company with M / A-COM director Larry Gould. As Jacobs wrote, "We found a common language, but Gould had a midlife crisis." Gould wanted to change the management system or merge with other companies - and his ideas didn't make much sense. The board of directors dismissed Gould (officially, “retired”) from the post of director in 1982. Jacobs was a member of the council, but he traveled around Europe and could not influence the decision-making about the new organizational structure in the way he wanted. Then he tried to split the company and put Linkabit pieces back, going so far in this that he even vetoed the deal with investors. At the last moment, the M / A-COM board of directors changed its mind and did not keep its promise to allow Linkabit to secede. Having finished work on the three chips of the commercial version of the VideoCipher II descrambler, Jacobs suddenly "retired" on April 1, 1985. Within a week, Viterbi left M / A-COM, and soon others followed him.
But as a result, everything that happened was not like a pension. For a man who did not want to do day-to-day management at Linkabit, Irwin Jacobs did an excellent job as a director. Shortly after he left M / A-COM, one of his colleagues asked him, “Why don't we try to do it again?”. Jacobs took his family, with whom he promised to spend more time on a car trip to Europe, promising to think about it.
On July 1, 1985, six people gathered in the Jacobs' house — six people — all of them who had recently left Linkabit. In addition to Jacobs, there were Franklin Antonio, Di Kofman, Andrew Cohen, Klein Gilhausen and Harvey White. The legends say that there were seven of them: Andrew Viterbi was mentally present there, although he actually was on a European cruise until mid-July, before leaving, agreeing with the ideas of Jacobs. The core team chose the Qualcomm name for the new company, an abbreviation for quality communications. They were going to combine elements of the theory of digital communications with practical design knowledge to improve code division multiple access, or CDMA.
In the Shannon-Hartley theorem on channel capacity, Shannon illustrates that technologies using extended spectrum can reliably transmit more digital data with a wider range with a lower signal-to-noise ratio. CDMA uses a pseudo-random digital code to propagate a given data transfer over the entire allocated range.
Various assigned codes allow you to create multiple CDMA data channels operating in the same band. For any single channel, all its neighbors, working with a different code, look like they speak another language and do not interfere with the conversation. For outsiders without a code, this whole system is difficult to interpret, it looks like background noise. This made CDMA much more protected from listening or jamming than the primitive pseudo-random frequency tuning ideas put forward by Nikola Tesla and later patented in 1942 by actress and inventor Hedy Lamarr and her friend composer George Anteil.
Unlike TDMA system, using fixed channels that determine the exact number of transmissions that the base station could accommodate in the allocated range, CDMA significantly expanded its capacity. With the help of complex encoding and decoding technologies, Reed-Solomon codes and Viterbi decoding came into play - CDMA could significantly increase the number of users, bringing it to an acceptable level of digital errors and inter-channel interference. CDMA even re-uses the capacity released during pauses in a call — an ideal characteristic for mobile voice communications.
Coding techniques also generate a solution for multipath propagation in extended spectra. The RAKE receiver, developed by Bob Price and Paul Green from MITLL, was originally designed for use in the radar field, and used many correlators, called “fingers”, that can synchronize with different versions of the signal and statistically combine the results. The RAKE receivers made CDMA almost immune to noise between channels.
US Air Force, who planned to launch Satcom satellite"the first were fascinated by all the advantages of CDMA, but in order to manage to process all data in real time, considerable computational resources were required. Jacobs and Viterbi realized that they had a very valuable technology in their hands, whose performance was proved by the digital signal processing capabilities of LMP and dual modem that reliably processed CDMA data for satellite communications. Could Qualcomm satisfy commercial demands?
From the very beginning, two things were obvious: in commercial projects, cost plays a much larger role, and regulators like the United States Federal Communications Commission (FCC) take the stage when designing communication networks. Therefore, Qualcomm found itself in the same position as Linkabit - they worked on government messaging projects, trying to make equipment smaller and faster.
Government projects led to the emergence of a Viterbi decoder on a single chip. Finally, the CMOS ASIC technology has mastered this task, and there is no need to use hundreds of TTL chips and exotic techniques like ECL. Qualcomm developed the first chip by September 1987: Q1401, a 17 Mbps, 80 states, K = 7, speed 1/2. LSI Logic manufactured it using a 1.5 micron technology, on a crystal with an area of 169 mm 2in ceramic PGA with 155 contacts. It was available in commercial and military versions, and in the second version its speed was slightly lower in favor of a wider range of operating temperatures.
Shortly before the opening of Qualcomm, Viterbi had an interesting telephone conversation. He was called by Allen Salmazi — who left JPL in 1984 to found OmniNet — asking if their firms could work together on a new truck tracking system.
In 1984, the FCC allocated frequencies for RDSS (satellite radiolocation service). OmniNet had one license for RDSS, its competitor Geostar had another. The concept of Geostar was to transfer the position and messages from the truck over the decimeter waves (L-band) to the satellite, which were relayed by the company owning the trucks. If OmniNet turned out to organize RDSS based on the truck itself, it would be a big advantage.
Qualcomm was not sure of this possibility. Salmazi gave them $ 10,000 to research the project - he had neither clients nor investors (no one believed it would work, even in Geostar rejected the offer of cooperation), there was only money "from family and friends." OmniNet had to commercialize the idea in order to survive, and Qualcomm was her last hope.
There were few satellites in the L-band and they were expensive, in particular, because their signal processing system needed to be adapted for each specific mission. Satellites in the centimeter range (K u -band) used for VSATand other tasks, it was complete, they were cheaper, they allowed to process the signal on the ground, gave two-way communication, but they had one drawback. The FCC issued a license to use the K u -band only for fixed terminals with large ground-based parabolic antennas, which had to be sent to the target with an accuracy of 1-2 degrees. The second use of the range allowed mobile use only when it did not interfere with the basic option. A smaller satellite dish, especially on a moving truck, had to have guidance and aperture problems that would almost certainly create interference. Then Klein Gilhausen said: "We will use CDMA."
In theory, CDMA and the extended spectrum should have solved any problems with the interference on the transmitter side, and if the antenna was directed accurately enough, then reception should work. But now they doubted the FCC about this. Qualcomm convinced the FCC to grant it an experimental license covering the use of 600 trucks. Jacobs and the team created a unique directional antenna with a diameter of 10 "and a height of 6", while being very accurate. The signal processing was handled by a 4 ”x8” x9 ”module, and the display contained four lines of 40 characters each, and it also had a small keyboard and indicators for the driver. By January 1988, the system began limited testing trips across the country.
Having failed to find customers, Salmazi squandered all the money - so Qualcomm bought him, his company and the entire system by launching the OmniTRACS system in August 1988. Having received no complaints of interference, the FCC allowed the system to operate without restrictions. By October, Qualcomm had the first serious customer, the Schneider company, which owns 10,000 trucks. OmniTRACS successfully grew, and today this system uses 1.5 million trucks. This first important victory gave Qualcomm the necessary capital that enabled it to enter the next major market for CDMA.
Gilhausen buzzed Jacobs and Viterbi all ears with his idea of entering the CDMA-enabled cell phone market. Viterbi this idea seemed familiar - he represented it in the work of 1982 on the extended spectrum. The transition from a network of military satellites to several hundred B-52 and EC-135, and from there to private satellite networks with tens of thousands of trucks, was fairly straightforward - but the public cellular network had a well-known problem.
Although CDMA signals reduced the interference of digital channels, it was necessary to take into account certain characteristics of radio waves in the case when multiple transmitters simultaneously communicate with one ground station. For satellite communications, all terminals on the surface of the Earth were located far enough, and under normal conditions, the signals from each of them were approximately of the same power.
In a cellular network with low-power handsets, distance mattered, and the far / near problem turned out to be serious. This problem is related to the dynamic range of the base station receiver. If all the tubes work with the same power, the nearest one locks the receiver and forgets the tubes that transmit from longer distances from the tower, which is why they are not heard in the noise.
Viterbi, Jacobs, Gilhausen and Butch Weaver set about finding out the details. In the meantime, they worked with CDMA simulations, the telecommunications industry association (TIA) at the meeting in January 1989 chose DMA based on D-AMPSas a standard 2G connection in the USA. D-AMPS has become an evolutionary development of AMPS, and some argue that in this choice there was a share of nationalism - an alternative was chosen to the dominant GSM in Europe, despite its rapid development. FDMA was considered to be a low-risk approach (Motorola, AT & T and other companies faced it), but TDMA already showed its technical superiority in GSM ratings.
Few in the industry took CDMA seriously. The Cellular Communications Industry Association (CTIA) insisted on increasing the number of users in the 2G standard by at least 10 times compared with the capabilities of AMPS, but also required a smooth transition period. DAMPS did not meet the capacity requirements, but was considered the fastest way to implement 2G.
Capacity issues gave Qualcomm a chance. Jacobs contacted CTIA, presented her the results of a CDMA study, and, after an initial refusal, obtained a speech at a meeting of members of the association in Chicago in June 1989. He expected that the assembled experts would make a lot of holes in his presentation, but this did not happen.
One of the reasons for the success of the presentation was that the company had tested its technology with PacTel Cellular since February 1989. After the TIA vote, Jacobs and Viterbi started asking for negotiations with regional operators. “Suddenly, once Irving Jacobs and Andy Viterbi came to my office. Honestly, I don’t even know how they got there, ”said PacTel Cellular director Jeff Haltman.
However, William Lee, chief scientific consultant for PacTel Cellular, knew why they came. PacTel Cellular in the Los Angeles market quickly grew the user base, and pretty soon the company had to face insufficient capacity. Lee has been studying the efficiency of the extended digital spectrum for years and problems with capacity, comparing FDMA and TDMA.
What he saw in CDMA technology was about 20 times superior to analog systems - and the risks of developing TDMA were enough to justify the $ 1 million rate to fund Qualcomm's research.
Lee, like many others, wanted to see a working solution to the far / middle problem and other problems.
A little less than six months later, on November 7, 1989, Qualcomm had a prototype. The CDMA “telephone” - and really it was 15 kg of equipment - was stuck in a van, ready to travel around San Diego. Two “base stations” were launched to demonstrate the transfer of a call between them.
The Qualcomm team, in particular: Andrew Viterbi (left), Irwin Jacobs (center), Butch Weeher and Klein Gilhausen (right) with a CDMA van, approx. 1989
Before the assembled directors of the cellular industry, of which there were at least 150, and according to some data, all 300, William Lee gave a presentation, then Jacobs and his own, and then Gilhausen began to describe what visitors should see today. And when they were about to disband the group and start the demonstration, Jacobs noticed that Butch Weaver was waving fiercely at him. Glitch GPS has violated the synchronization of base stations. Jacobs improvised, and continued to spread about CDMA technology for 45 minutes, until Weaver and the team brought the system to a working state.
Many visitors were amazed by what they saw. Critics said that CDMA would never work, that the theory would not withstand the full-scale deployment of real-world conditions, and one expert even stated that "it violates the laws of physics." In addition, there was still a small problem with the placement of all the equipment in a small tube - but Qualcomm was ready to cope with it. In addition to the need for miniaturization and basic tasks of expanding the spectrum by direct sequencing and channelization, Qualcomm developed solutions for the three main problems of CDMA.
The first was the far / near problem. Dynamic power control changes the levels so as to maintain an adequate signal-to-noise ratio. CDMA tubes, located closer to the base stations, usually use less energy for transmission, and those that are farther away - more. As a result, all signals come to the base station with approximately one signal-to-noise ratio. Lowering the transmission energy also weakened the interference and saved the battery. Qualcomm used an aggressive open-and-closed-loop energy regulation algorithm that adjusts 800 times per second (later this number increased to 1500), which was much higher than the few times per second that GSM was content with.
The second was the transfer of the call. In the TDMA system, calls were usually disrupted when the user switched from one base station to another due to a hard transmission. CDMA phones established a connection with the next base station, not yet disconnected from the current one.
The third was a variable coefficient vocoder. Instead of encoding on / off in GSM, the variable-ratio encoder quickly adapted to natural pauses and conversation resumption, reducing the number of bits transmitted by the handsets, which increased the total capacity of the base station. This property in TDMA was not, in it the channels were fixed, and they could not be divided.
In the case of CDMA commercialization, Hultman promised support from PacTel Cellular, but other deals had to be made to reach a critical mass. PacTel introduced Qualcomm to higher-level company directors from other companies that emerged after the separation of AT & T’s monopoly, and the main cellular infrastructure manufacturers, in search of markets where CDMA would be useful. Qualcomm management has also made a fateful decision regarding the business model: instead of self-fabricating all the equipment, the company will sell licenses to use intellectual property using CDMA to manufacturers.
New York, where the telephone company NYNEX was operating, became another cellular market with problems of increasing capacity. Qualcomm brought its CDMA prototypes to Manhattan for field trials in February 1990. NYNEX has already ordered AT & T to look for next-generation infrastructure, and in early July, AT & T and Qualcomm agreed on a license for CDMA base station technology. On July 31, 1990, Qualcomm published the first version of the CDMA specifications for the collection of industry comments - Common Air Interface. On August 2, NYNEX announced that it would spend $ 100 million to create a “second cellular telephone network” in Manhattan by the end of 1991, mainly to allow time for the placement of frequencies and the construction of base stations. $ 3 million were to be withdrawn by Qualcomm to manufacture CDMA phones.
Other companies were not in a hurry with the offers. The two largest providers of cellular infrastructure, Ericsson and Motorola, planned to deal with TDMA networks. Motorola insured by entering into an interlicensing agreement with Qualcomm in September 1990, but publicly expressed concerns about technical issues. Operators, such as McCaw Cellular (the predecessor of AT & T Wireless) and Ameritech, tried to postpone a decision on the large-scale adoption of CDMA. As for other places, Europe made a bid for GSM based on TDMA, and Japan developed its own cellular network based on TDMA.
In the column of the undecided was Korea, which had no digital solutions. Salmazi achieved that Li from PacTel presented the company in August 1990, which resulted in consecutive discussions, which ended in May 1991 with an agreement on the joint development of ETRI CDMA. And although this program received extensive funding and promised large royalties in the future, five years were spent on its deployment.
But even after these victories, Qualcomm from a financial point of view continued to balance on the edge. Every dollar of profit went to the salaries of workers, whose number had already increased to 600 by 1991 - and to research and development in the field of CDMA.
PacTel continued to work on CDMA plans, which led to the testing of CAP I capacity in November 1991 using Qualcomm CDMA ready for commercial use. During the two-year program, five ASICs were developed. Three integrated circuits were required for a CDMA phone: a modulator, a demodulator, and an advanced Viterbi decoder. The other two were created for the base station, which also used the Viterbi decoder. These chipsets were connected to an external microprocessor. The tests showed good performance of the CDMA technology on a large scale, and proved that it can achieve the predicted capacities.
Qualcomm CDMA chipsets, approx. 1991
On the heels of the announcement of the success of the CAP I and ASIC tests at the CTIA technology forum, Qualcomm made the first exit to the stock exchange, offering 4 million shares and collecting $ 68 million in December 1991. PacTel bought a stake in the open market, and added another $ 2.2 million for the purchase of warrants for 390,000 shares in addition to guarantee continued research and development in the field of CDMA.
By the beginning of 1992, in addition to the Korean joint project ETRI in developing the CDMA standard with Qualcomm four companies have already decided to participate: AT & T, Motorola, Oki and Nortel Networks. The licensee number five in April 1992 was none other than Nokia - it was the culmination of a year and a half negotiations between Jacobs and Jorma Ollila. Nokia watched PacTel with great interest, and opened its own research and development center in San Diego to get closer to what's happening with CDMA. One of the stumbling blocks was royalties: it is believed that Nokia paid about 3% of the average selling price of the handset under the terms of its first contract for 15 years.
On March 2, 1993, Qualcomm introduced the CD-7000, a CDMA / AMPS-enabled phone running on a single narrow-band transmission chip: Mobile Station Modem (MSM). The phone was a typical chocolate bar form factor, 178x57x25 mm in size and weighing just over 340 grams. The first client was US West, with an application for at least 36,000 phones. Also in March 1993, four manufacturers announced their plans for CDMA phones and infrastructure in Korea: Goldstar, Hyundai, Maxon and Samsung.
Qualcomm unveiled the details of the MSM narrowband transmission chip at the Hot Chips Symposium in August 1993. The three main functions of a CDMA modulator, demodulator and Viterbi decoder were placed on a single chip made using 0.8 micron technology with an area of 114 mm 2 . It had 450,000 transistors, it consumed 300 mW, and to work as a tube, it still needed an external processor and circuit for working with radio waves. Qualcomm hinted that the supply of spare parts involved several different factories, but did not disclose suppliers - later it was reported that one of them was IBM.
TIA finally conceded, endorsing CDMA in the first publication of the IS-95 specification in July 1993; Commercially, this option became known as cdmaOne. A choice of digital standards for 2G has emerged in the cellular markets: CDMA, D-AMPS and GSM.
In the CD-7000 phone, along with the MSM chip, an Intel 80C186 processor was installed. The next logical step was their integration, but Intel was not involved in intellectual property. At first, Intel rejected Qualcomm offers. But under the constant onslaught of the sales department in San Diego, Intel, located in Chandler, pcs. Arizona learned all about Qualcomm, CDMA technology, and market opportunities, before finally agreeing on the delivery of the 80C186 core.
The task of changing the design of Intel 80C186 to a more standard one for the industry has proved difficult. Qualcomm developed MSM using high-level hardware description language (HDL) techniques that could be quickly reconfigured to various libraries, databases for simulations, and test vectors. It quickly became clear that it would be easier to transfer Qualcomm MSM IP to the Intel technological process, and give Intel the entire chip production. Qualcomm agreed to this. Intel was going to simultaneously engage in mobile and manufacturing businesses.
On February 1, 1995, Qualcomm announced a Q5257 MSM2 with a Q186 core in a QFP with 176 contacts, as well as an integrated Q5312 chip (Analog Baseband Processor, BBA2), which replaced 17 individual chips in an 80-pin QFP. These two chips made up most of the CDMA phone, such as, for example, the QCP-800, which was announced the next day. Qualcomm, in preparation for the release of large volumes, cooperated with Sony for the release of a new phone that supports two communication standards, and with a doubled battery life, which lasted five hours of conversation. It was also announced the release of a single Q5160 Cell Site Modem chip (CSM) for CDMA base stations, in which there was no integrated processor.
In June 1996, Q5270 MSM2.2 was introduced. The main improvements were the 13 Kbps PureVoice vocoder, using QCELP and giving better sound quality without increasing power consumption. It was offered in QFP format with 176 contacts for commercial use and in a larger one, with 208 contacts for debugging circuits.
Reducing power consumption was the goal of developing the MSM2300, which was announced in March 1997. Searching for a signal using hardware DSP worked up to eight times faster than that of MSM2.2. The 176-pin QFP had backward compatibility, which allowed equipment to be directly upgraded.
With the deployment of CDMA around the world, the number of chipsets produced has grown at an explosive pace. Qualcomm claimed that, in the total, shipments of various MSM variants — mainly MSM2 and MSM2.2, produced by Intel — reached six million units by June 1997. Intel also promoted its low-power embedded 386EX processors for Nokia and Research in Motion tubes [future BlackBerry / approx. trans.]. What could go wrong?
Probably, such a question Qualcomm asked herself when Intel refused to do the planned update of the embedded kernel. Frankly, the complexity of the production of the 386EX was much higher, and it was necessary to somehow place more chips from Qualcomm. Intel probably found the design too risky, and decided that six million pieces would not be enough to justify.
Qualcomm tried to hurry events, asked for the possible cost of this solution, and received a very superficial answer, without any major improvements in processor speed. (Intel, most likely, then just sued DEC about the Alpha chip. If Qualcomm had a need for a new kernel a little later, and if Intel had figured out the intellectual property or production business model for StrongARM, then Intel’s role on the mobile the market could have been completely different). And while the supply of existing chips continued, the phase of working with Intel on the next generation of chips for Qualcomm was over.
The high-performance processor core was not searched for long. Many users of Qualcomm CDMA licenses, in particular, LSI Logic, Lucent Technologies (split from AT & T), Samsung and VLSI, were proponents of ARM. Officially, Qualcomm announced the release of the first license using ARM in July 1998.
New chipset launches accelerated, and Qualcomm became one of the most fruitful suppliers of ARM-based chips, and its products were actively used in thousands of mobile devices. Below we will list only the key chip models.
When the ARM deal became widely known, the MSM3000 chip was already in the process of being developed, and its release was announced in February 1998, and the core was replaced with ARM7TDMI. Other enhancements included the SuperFinger demodulator, which accelerated data transfer to 64 Kbps, and improved standby mode. It was produced by the process of 0.35 microns. For the first time, the products were manufactured by TSMC . To avoid confusion with the old models, since the new required completely different software, the QFP with 176 contacts had a completely different pinout.
At that time, there was one more core in production. For quite a long time, DAC chips were present in the product line, and in February 1999, the MSM3100 with the ARM7TDMI core and the independently manufactured QDSP2000 programmable core were introduced. The QDSP2000 operational unit had a computational pipeline with five stages and optimized instructions for implementing a variable compression ratio codec and other functions, such as eliminating echo.
The 3G technology debuted on the MSM5000 chip supporting the updated cdma2000 specifications. Its release was announced in May 1999, and while still working on the ARM7TDMI core, it reached a speed of 153.6 Kbps and had improved search capabilities. The following year, MSM5000 was used in cdma2000 field tests, and its High Data Rate (HDR) technology would later develop to 1xEV-DO.
Flirting with Palm and a pdQ CDMA phone in September 1998 led to a survey of operating systems for smartphones. In September 1999, Qualcomm announced plans to develop iMSM chips designed for Microsoft Windows CE and Symbian OSs, including the iMSM4100 with a dual core of ARM720T processors, one for data transfer and the other for the OS. With the advent of StrongARM and other solutions, the iMSM4100 at the time of launch was ahead of them in integration but was slower in speed. Qualcomm was well versed in data transfer, but she still had a lot to learn about application processors.
The evolution of data transfer chips from Qualcomm
By the mid-2000s, there were three families of chips in development: 2G cdmaOne, 3G cdma2000, and prototypes of application processors, such as MSP1000 (essentially, iMSM with only one ARM720T processor).
Against the background of the presence of many manufacturers of CDMA phones, Qualcomm retired from this business, selling it to Kyocera in February 2000. After many years, during which Andrew Viterbi gave birth to new ideas, he announced his retirement in March. In May, Qualcomm announced that the total supply of MSM chipsets exceeded 100 million.
In February 2001, Qualcomm developed an ambitious plan. The MSM6xxx family development plan included a wide range of products, ranging from the initial MSM6000 chip based on ARM7TDMI with support for only 3G cdma2000. A set of Launchpad applications based on the new BREW API helped manufacturers to more effectively develop software. Modules such as radioOne have also been added to increase the efficiency of the Zero Intermediate Frequency conversion, and gpsOne to improve the positioning.
At the other end of the scale was the MSM6500, running on the ARM926EJ-S with two QDSP4000 cores, supporting 3G cdma2000 1xEV-DO and GSM / GPRS plus AMPS, all on the same chip. MSM6500 came out almost two years later, it was made using 0.13 micron technology, packed in a case with 409 contacts CSP.661,662 2003, and it marked the beginning of the change of the company leader. In January, Don Schrock announced his retirement from the post of head of Qualcomm CDMA Technologies (QCT), giving way to Sanajay Jha, who led the development teams of MSM.
The next plan was the MSM7xxx family, it was first shown in May 2003, and plans for it were similar - a wide range of entry-level chips to expensive ones. The 90nm MSM7600 version carried the ARM1136J-S with a frequency of 400 MHz and QDSP5000 for applications, plus 274 MHz ARM926EJ-S and QDSP4000 for multi-mode messaging. Also on the chip was a Q3Dimension GPU, from the IMAGEON license agreement with ATI. In 2006, MSM7600A decreased to 65 nm and received a frequency of 528 MHz. MSM7600, still going under the MSM sign, marked the change of direction for future Qualcomm application processors.
In September 2003, Qualcomm reached a milestone of 1 billion chips from the MSM line - nine years after the first commercial release.
“Qualcomm has always been present in the semiconductor business,” was how Klein Gilhausen began his speech at Telecosm 2004. “We always knew that the key to implementing CDMA technology was a very aggressive policy for developing specialized chips.” Qualcomm's next steps were testing how aggressive a company could be.
Irving Jacobs resigned as director of Qualcomm on July 1, 2005 - the year of the 20th anniversary of the founding of the company - and became chairman of the board of directors. He was replaced by his son, Paul Jacobs, who was involved in speech compression algorithms, the launch of the pdQ smartphone, the BREW project and others. Stephen Altman, who led the licensing, replaced the outgoing president of the company, Tony Thornley. In general, the development strategy has not changed.
Paul Jacobs and Irwin Jacobs, approx. 2009
Many ARM licensees immediately supported the release of the new ARM Cortex-A8 core in October 2005. Instead of making the finished version, Sanjay Jha received the first architectural license for ARMv7 and revealed plans for the Scorpion processor core in November 2005. The loud headlines that this would be the first ARM core, operating at 1 GHz, were slightly exaggerated; Samsung has been promoting the ARM10 Halla design, operating at 1.2 GHz three years before. Nevertheless, Qualcomm has overtaken all competitors with its Scorpion, for example, TI OMAP 3 using Cortex-A8, and released its core design to the market two years earlier than Intrinsity Hummingbird.
The company's advantage came from the little-known acquisition of Xcella in August 2003 — it was a North Carolina company founded by former IBM employees, including Ron Tessitore and Tom Collopy. They made a huge contribution with their experience in developing processors.
The Scorpion used a 13-stage loading / storage pipeline, similar to the Cortex-A8, but it had two additional processing pipelines - one ten-step for simple arithmetic, and the other 12-step for multiplication with accumulation. SIMD operations in the VeNum multimedia engine had pipelines with a large number of stages, and the data width was doubled to 128 bits. Clock-do-Mania clocked logic, enhanced termination buffer, and other tweaks to optimize power consumption for the TSMC 65 nm LP process have led to energy savings of up to 40% compared to Cortex-A8.
DAC capabilities have also been improved. The Hexagon DAC core, also called QDSP6, also switched to the 65 nm process technology. It began production in the fall of 2004, and Hexagon used three techniques to save energy: Very Long Instruction Word (VLIW), multi-threading to reduce overhead in the absence of the necessary data in the L2 cache and a new set of instructions to maximize workload per package. The 64-bit vector execution unit processed up to eight simultaneous 16-bit multiplication operations with accumulation in one cycle. Three threads could run four instructions each cycle, two on double vector execution units, and two on double load / storage units.
Both cores under the new brand for application processors: Snapdragon. On November 14, 2007, Qualcomm showed a new QSD8250 with HSPA support and a dual-mode QSD8650 with CDMA2000 1xEV-DO and HSPA. Each had a Scorpion processor at 1 GHz and a Hexagon V1 DSP core at 600 MHz. Also on the chip were the Adreno 200 GPU (renamed after Qualcomm bought ATI mobile graphics assets from AMD in 2009), running at 133 MHz. Continued releases of the multi-mode combination ARM926EJ-S with QDSP4000.
Qualcomm flourished thanks to the fashion for "netbooks", and more and more often found itself competing with Intel and its Atom processor. WiMAX has become Intel's standard for broadband notebook connectivity, but it needed a new infrastructure. Taking advantage of the chance, Qualcomm unveiled its first Gobi chipset in October 2007 and used 65 nm MDM1000 to connect netbooks and similar non-phone devices to the Internet using EV-DO or HSPA over existing 3G networks.
Sales for use in PCs and netbooks immediately made Gobi a hit, and Snapdragon's popularity grew more slowly. Gobi began to inject resources. The MDM9x00 family development plan, announced in February 2008, contained a 45 nm process and a modem enhanced to support LTE, which later turned out to be based on ARM Cortex-A5. After Sanjay Jha left for Motorola in August 2008, Qualcomm raised Steve Mollenkopf to head of the QCT in order to maintain the former direction of development of the core strategy.
But it was time for a big change in mobile operating systems that Snapdragon was supposed to help. In September 2008, the T-Mobile G1, created by HTC, was the first Android-powered phone - and it worked on a Qualcomm MSM7201A chip. LG and Samsung worked on Android phones with Qualcomm chips to release them in 2009, and Sony Ericsson did not lag far behind them.
Snapdragon went further, to the second generation since. the 45nm process, introduced in November 2009. MSM7x30 were supposed to reduce the cost and energy consumption, and rolled back to use the 800 MHz core of the Scorpion with a QDSP5000 at a frequency of 256 MHz and a stranded Adreno 205 GPU. In preparation for dual cores, the 45nm version of Scorpion received debugging capabilities borrowed from ARM Cortex-A9 and improvements in the L2 cache. In June 2010, the third generation of Snapdragon MSM8260 and MSM8660 appeared, in which two Scorpion operated at 1.2 GHz, together with 400 MHz Hexagon V3, plus Adreno 220 GPU with improved efficiency. Shells were getting bigger; the MSM8x60 had 976 contacts, 14x14 mm in size and a nanoscale package (NSP) type.
Qualcomm's method of working with the announcement of new products was usually to inform the media about an early version of development plans, and then release the finished product two or three years later. When the World Mobile Congress (MWC) began in February 2011, Qualcomm had a pair of aces up his sleeve for use in presentations.
First: Gobi switched to the 28 nm process in the form of MDM9x25. Among the improvements are the addition of support for Category 4 speeds, up to 150 Mbps on LTE FDD and LTE TDD, and support for HSPA + Release 9. Trial batches of these third-generation chips appeared at the end of 2012.
The second has already been partially announced twice. A couple of MWCs before that, Qualcomm mentioned MSM8960, a new version of Snapdragon, designed for multi-mode operation, including LTE. At an analyst briefing in November 2010, this chip was identified as a 28nm transition to the process, using the next generation of processor cores on the new microarchitecture, as well as a faster Adreno GPU. At MWC 2011, the first ARM processor with a 28 nm core was named: Krait.
It was announced that Krait would be the core used in three different chips. At the lower end of the scale was a dual-core 1.2 GHz Krait MSM8930 with an Adreno 305 GPU. In the middle was the MSM8960, a dual-core 1.5 GHz Krait with a faster Adreno 225 GPU. At the top end was an APQ 8064 with a quad-core 1.5 GHz Krait with an Adreno 320 GPU.
Independent cores in voltage and frequency allowed Krait to significantly save energy, up to 25-40% compared to the SMP approach, such as the big.LITTLE with ARM Cortex-A15, depending on the load. Advantages in efficiency, in particular, were achieved thanks to the 3-wide instruction decode, compared to the 2-wid of the Scorpion, and also thanks to the out-of-order execution, the 7th execution ports as compared to the 3, and the doubled L2 cache increased to 1 MB. This allowed Krait to rise to 3.3 DMIPS / MHz.
Trying to understand the heap of nomenclature created by them, Qualcomm at a meeting of analysts in November 2011 defined a hierarchical branding scheme. The new Krait 28 nm based chips are called Snapdragon S4, and split into S4 Play, S4 Plus and S4 Pro. 65 nm Scorpion were designated Snapdragon S1, 45 nm single-core Scorpion - Snapdragon S2, and 45 nm dual-core Snapdragon - S3.
Sometimes marketers excel themselves. Hierarchy is good, but a confusing nomenclature, which is difficult to translate from English, is not very good. A second attempt at CES 2013 led to the emergence of the modern Snapdragon license plate.
It was announced that the flagship Snapdragon 800 for high-end phones will contain a quad-core Krait 400 CPU at 2.3 GHz and a Hexagon V5 at 600 MHz and Adreno 330 at 450 MHz, as well as an LTE modem. The Snapdragon 600 had a quad-core Krait 300 CPU at 1.9 GHz with a Hexagon V4 at 500 MHz and an Adreno 320 GPU at 400 MHz, without a modem, for reasons of economy.
The launches that followed with CES 2013 fall into the category of Snapdragon 200 for phones, Snapdragon 400 for phones and tablets, Snapdragon 600 for devices of average performance, and Snapdragon 800 for high performance. The Snapdragon 200 line uses the ARM Cortex-A7 core to save money.
There was another example of not very successful marketing. Shortly after the unexpected launch of the Apple A7 chip with 64-bit support in September 2013, Qualcomm chief marketer Anand Chandraseker responded with great skepticism about its value to users. In the process of further study (and, possibly, after several annoyed calls from ARM), Chandrasekera was scolded, and his statements a week later were officially recognized as “inaccurate”.
The crisis was avoided, but it was not possible to come up with an answer. At a research meeting in November 2013, Qualcomm showed a fourth-generation Gobi development plan, moving to 20 nm from 9x35, supporting LTE Category 6 and carrier aggregation. In December 2013, a hurried presentation of the quad-core Snapdragon 410 with ARM Cortex-A53 brought Qualcomm back to the arena of 64-bit application processors.
It may have so simply coincided in time, but a few days after the presentation of the Snapdragon 410, there was a serious rotation of managers. Paul Jacobs announced that he would quit the Qualcomm director post, remaining chairman of the board of directors, and Steve Mollenkopf was temporarily promoted to director on December 12, 2013, with a permanent appointment next March, if shareholders approve his candidacy.
Steve Mollenkopf
In April 2014, the Snapdragon 810 was shown on TSMC 20 nm. Eight cores and a big.LITTLE scheme had four ARM Cortex-A57 cores at 2 GHz and four Cortex-A53 cores at 1.5 GHz. Also inside was the returning Hexagon V5 and its dynamic multithreading at 800 MHz, the Adreno 430 GPU at 600 MHz and the new support for LPDDR4 memory. Also included were a Cat 9 LTE modem, full 4K Ultra HD video support, and two graphics processors for computational photography. His younger brother, the Snapdragon 808, used two ARM Cortex-A57 cores instead of four, a simpler GPU, Adreno 418, and only supported LPDDR3.
Gobi chips on the fifth generation of the fifth generation became the main subject of discussion for analysts in November 2014. Gobi 9x45 supported LTE Advanced Category 10. This assumed a download speed of 450 Mbps by combining carriers in LTE.
On the development plan of Qualcomm, apparently, there was an A / B strategy - take the intellectual property of ARM where it is, add core of internal development, repeat the cycle. This is the only reasonable way to compete on a wide range of four options, from the lowest to the highest level. The Snapdragon 200 line combats an influx of ARM Cortex-A5 based chips from Taiwan and China, and the Snapdragon 800 and Gobi are fighting monsters like Apple, Intel, Samsung and many others.
Qualcomm's relentless improvement in chip design has brought tremendous success to CDMA and Android. In the growing market of cell phones, growing by 11%, 80% of which is occupied by Android, Qualcomm faces new problems previously unseen. Instead of celebrating the 30th anniversary, in July 2015, Qualcomm announced a reduction in the number of employees by 15%. Experts believe that this sad news was due to the fact that the 64-bit wave launched by Apple caught Qualcomm off guard, followed by a Snapdragon 810 overheating scandal from LG and Samsung.
Qualcomm vice president of marketing Tim McDonough has his own view of the Snapdragon 810 overheating history, saying that all decisions about phones are made 18 months before the public sees them - and, as we have seen, the main decisions on development plans Chips are accepted 18 months before. The latter is controlled by Qualcomm. The first is getting shorter than you might want Qualcomm. The source code hints that LG has switched from Snapdragon 810 to a weaker version of Snapdragon 808 - leaving the same LTE implementation - just a few months before the release of the LG G4 product. McDonough claimed that the problems were with the pre-release Snapdragon 810 (which had since been updated, after which overheating reports were gone), and that manufacturers are switching to the Snapdragon 808 because that they do not need support for full-size 4K video. The longest is the LTE modem evaluation, and this process has already started for a long time. That would make the transition - if it took place at LG - a quick and less painful one. Samsung could have had its own interests in indicating the existence of this problem - at that moment the company was preparing to launch the flagship Exynos 8 Octa.
The main parameters of mobile processors Qualcomm
Perhaps recent events are forced to be more cautious when placing development plans for all to see. At MWC 2015 in March, the main theme of the presentation was Snapdragon 820 with Kryo, the new 64-bit ARMv8-A CPU core. Details about the four cores and the clock frequency of 2.2 GHz appear (as well as rumors of even higher speeds) and the new production partner of Samsung with their 14 nm FinFET process. In August, plans were made for the Adreno 530 GPU and the new Spectra image processor for the Snapdragon 820; In addition, a new Hexagon 680 DAC is being developed.
Qualcomm November 10, 2015, when communicating with the media, reiterated that Snapdragon 820 consumes 30% less energy than Snapdragon 810. They also mentioned support at the system level, Cat 12 LTE, 802.11ad Wi-Fi and anti-malware. learning. Their marketing is moving away from the specifications of intellectual property to examples of using the capabilities of chips, which is good news.
Kryo creates a possible point for entering the emerging market of 64-bit server ARM. Competition with Intel and AMD on their field can turn into an interesting adventure. Qualcomm also pursues the Internet of things, having technologies as a result of the acquisition of Atheros and CSR, and develops software in AllJoyn. How Qualcomm will change the business model based on licensing complex communication algorithms will determine whether the company will remain the leader among non-manufacturing firms. Can they develop intellectual property supporting a new application segment, for example, drones? Is there more work on 4G LTE cellular, and how quickly will 5G technology be deployed?
Attempts by investors to divide the company into a division, one of which will deal with intellectual property, and the other with chips, appear ill-conceived. Although the part of the business that licenses intellectual property has a legacy in the form of cash flow from CDMA, the chip business benefits from strict adherence to the action plan. Without this synergy, what will fuel the chip business?
While mobile devices will use wireless connections, Qualcomm is not going anywhere. In the near future, complex strategic issues are to be resolved, and this can lead to a serious and extensive influence on the production strategy and competition in the application segments.
Chapter 9: Press Q to connect
Unlike other similar firms that started work in other segments of electronics, and then transferred to the field of devices for communications, Qualcomm has always focused on wireless technology, reliably transferring data between two points. Its CDMA technology has become a leap forward for mobile devices - if it could be made small enough, and if it were possible to convince providers to switch to it with D-AMPS and GSM.
NASA Jet Propulsion Laboratory
The roots of Qualcomm can be traced to one of the smartest people in the academic environment at leading technological universities in the United States who, as young engineers, came together for the first time to work on a space program. The depth of technical knowledge gained from satisfying the needs of demanding customers in digital data transmission systems has created a foundation on which patents, chips and devices then appeared.
Real rocket scientists
The fruitful article “Mathematical Theory of Communications”, published by Claude Shannon from Bell Laboratories in 1948, laid the foundation for information theory. Together with the invention of the transistor and the advancements of digital programming and computing, Shannon's theorem and his work at MIT inspired a whole generation of mathematicians and scientists.
In June 1957, Andrew Viterbi, a MIT graduate with a master’s degree in electronic engineering, joined the Jet Propulsion Laboratory (JPL) in Pasadena, pc. California. At that time JPL was owned by the California Institute of Technology, but it worked under the auspices and with the money of the US ballistic missile agency.
Viterbi worked in Section 331 for communications led by Solomon Golomb. They developed telemetry for rockets and satellites. Golomb pioneered the theory of shift registers with linear feedback, which was used to encode digital messages for reliable transmission at high noise levels. Viterbi was working on closed phase loops, an element of this theory, critical for synchronizing a digital radio receiver with a transmitter — this was required in order for the information flow to be decrypted.
October 4, 1957 the USSR launched the " Sputnik-1"The next day, the Mylston-Hill radar belonging to the Lincoln Laboratories at MIT (MITLL) - where Irvin Reed, the most famous Reed-Solomon code reader, worked - discovered the Satellite in low orbit. amateur radio to track the satellite signal, which has grown and faded away every 96 minutes, according to the orbit movement of the satellite.
Space race has begun. The US Navy began to hurry with the response of their project "Avant-garde". On December 6, 1957, Test Vehicle 3 was launched, the third test vehicle with a satellite weighing 1.3 kg. He climbed the disgraceful 1.2 m, lost his cravings and fell back onto the launch pad, exploding. The payload landed nearby, in the bushes of Cape Canaveral, but did not stop the broadcast. “This is our competitor,” said Golomb.
On January 31, 1958, the Project Deal, known worldwide as Explorer 1, reached orbit. Life magazine posted a photo of Golomb and Viterbi in the JPL flight control room on the cover. On July 29, 1958, President Eisenhower signed a decree on national aeronautics and space flight, creating NASA. The JPL requested and received a transfer under the auspices of NASA in December 1958.
Viterbi enrolled at the University of Southern California (USC) to defend his doctoral thesis - this was the only institution that allowed him to continue working at JPL for a full day. He graduated in 1962 and went to teach at the University of California, Los Angeles (UCLA). He recommended Golomb to join the staff of USC faculty, where Reed was already (who joined Santa Monica’s Rand Corporation in 1960), Lindsey (joined JPL in 1962), Eberhart Rechtin, Lloyd Welch and others.
Many years later, Lindsay joked, "I think this group was created by God." Rehtin said that as a result of joint work, this group was able to do more in the field of digital communications than any of them could have done alone. Their work has influenced countless other people.
San Diego Related
In 1963, at the national electronics conference in Chicago, the award for the best work went to Viterbi and Irwin Jacobs, a professor from MIT, whose office was located next to the office of Claude Shannon. Jacobs and Viterbi had already met in 1959, when Jacobs came to JPL for an interview, and each of them knew about the work of the other through the links of JPL and MIT.
At the next meeting at the conference in 1963, Jacobs told Viterbi that he would soon have his sabbatical and asked if it was interesting to work at JPL. Viterbi assured that the way it is. Jacobs was denied a request to take him to work, but Viterbi put in a word for him to the head of the unit Rehtin, and Jacobs eventually was hired as a researcher and sent to Pasadena. Viterbi taught at UCLA and advised at JPL, and the two became friends while Jacobs worked at JPL from 1964 to 65.
Having published the important for the history of the work "Principles of communication technology" together with John Wosenkraft in 1965, Jacobs moved to the west coast in 1966. One of his teachers in Corneille, Henry Bucker, persuaded him to join the new engineering department at the University of California at San Diego ( UCSD). Professors were then appreciated, and digital communications consultants were also in demand. One day, in early 1967, Jacobs went to a conference at the NASA Ames Research Center. On the way home on the plane, he found himself flying with Viterbi and another MIT graduate, Len Kleinrock, who joined the UCLA staff in 1963 and made friends with Viterbi. They began to talk, and Jacobs casually noted that he had accumulated more consulting work than he could handle.
Viterbi was finishing his masterpiece. He was looking for ways to simplify the theory of isolating weak digital signals from strong noise — so that his students at UCLA would have a better understanding of it than in the complex course that existed at the time. He came up with a general concept in March 1966, and refined the idea for a whole year before publishing. In April 1967, Viterbi described his approach in an article in the IEEE Transactions on Information Theory magazine entitled “Convolution Codes Error Bounds and Asymptotic Optimum Decoding Algorithm”.
Viterbi Algorithmpromotes "soft" solutions. The hard decision about whether a signal is zero or one can be made by observing each noisy bit received (or a group of bits encoded into a symbol), with a high probability of error. Viterbi considered probabilistic information contained in possible state transitions, and is known based on how the characters are encoded by the transmitter. Sequence analysis of the received symbols and state transitions using the “add-compare-select” operation (add-compare-select, ACS) determines the path of maximum likelihood, and more precisely coincides with the transmitted sequence.
It was just a theory, or so Viterbi thought at first. The algorithm reduced the amount of computation and the number of errors compared to others, but it still needed to be performed in real time, and it was believed that for a sufficiently small percentage of errors it would take “several thousand registers”. This work was picked up by several other researchers, among whom it is worth noting Jim Massey, David Forney and Jim Omur. They were convinced of its optimality. Jerry Heller, one of Jacobs' graduate students at MIT who joined him in San Diego, worked at JPL. He decided to drive away a few simulations, and from 1968 to 1969, he discovered that Viterbi was evaluating his theory too pessimistically; A fairly good advantage was given only 64 registers. But for that time, it still was a rather large cabinet of computing equipment.
Entrepreneurial ideas associated with a consulting firm did not let Jacobs go. In October 1968, Linkabit was born with a share capital of $ 1,500 (each of the founders contributed about $ 500) and an address that coincides with the Kleinrock home address in Brentwood. Soon the offices moved to a building located in Westwood, near UCLA. First, Jacobs, Kleinrock, and Viterbi, who taught full-time, spent one day a week at their firm.
However, the company has more cases than expected. The first engineer hired by the company in September was Jerry Heller, which was soon followed by Andrew Cohen, Klein Gilhausen and Jim Dunn. Len Kleinrock retired for several months, doing his favorite project - setting up the first end nodes of the ARPANET network and sending the first message on it in October 1969. If you believe him, when he tried to return to Linkabit, he was immediately deployed, issuing as a dismissal a certain percentage of the value of the company. In the absence of Kleinrock, and in view of the fact that Viterbi did not want to move for a few more years, Jacobs moved Linkabit's office in Sorrento Valley, one of the corners of San Diego’s Golden Triangle, in 1970. After that, he hired office manager Di Koffman immediately after he finished studying in high school.
Programming modem
"Programming is dead." Several speakers spoke on this topic at the IEEE Communication Theory Workshop conference, which was held in St. Petersburg in 1970, pcs. Florida. Irwin Jacobs stood in the far corner of the room, holding a 14-pin DIP chip - a simple 4-bit shift register, possibly 7495 from the TTL family (transistor-transistor logic). "This is the current state of digital technology, it will allow us to create all of this."
In the early days, Linkabit was a think tank, not a hardware manufacturer. Her first clients were the NASA and JPL Ames Research Center, as well as the Naval Electronic Laboratory at Pont-Loma and DARPA. Linkabit research related to the decoding of Viterbi resulted in a messaging system for deep space that was used in the Voyager project and other programs. However, soon compact versions of Viterbi decoders and other signal processing equipment will make Linkabit and its follower legendary.
Heller and Jacobs introduced a 2 Mbps Viterbi decoder with 64 states and a depth of 7 in October 1971. It was based on a commercial decoder made for military satellites. Linkabit Model 7026, or LV7026, used about 360 TTL chips on 12 boards in a 19-inch case, and was 4.5 U (7.9 ") high and 22" deep. Compared with previous versions of the equipment involved in processing the Viterbi algorithm, and occupying several racks the size of each refrigerator, this was a breakthrough.
The speed of work was also a problem. Viterbi talks about an early attempt by Linkabit to integrate a single ACS-state decoder on a chip containing only 100 logic elements - it was an average integrated circuit, or MSI. In his words, such an attempt “almost bankrupted the company” due to several consecutive problems with suppliers. Almost bankrupt? This seems an exaggeration until we consider the then available alternatives to TTL. Based on the company's 1971 report and Magnavox's 1974 document, Linkabit was played with the fast-working, but very capricious, “emitter-coupled logic” (ECL) technology, trying to increase the clock frequency of critical areas. Many companies could not do anything with ECL. Viterbi did not mention specific names, but among the suspects are Fairchild, IBM, Motorola and Signetics.
The change of direction brought more success. Klein Gilhausen started playing with the concept of a Linkabit Microprocessor (LMP) microprocessor, microcode architecture implementing the functions of a satellite modem. Gilhausen, Sheffie Worboes, and Franklin Antonio completed the LMP prototype board using mostly TTL chips and commercial high-speed ICUs and LSIs by May 1974. She worked at a speed of 3 MIPS. She had 32 instructions and four software stacks, one for processing, and one for monitoring. It was partly RISC (before the appearance of such a thing), partly DSP.
Jacobs began writing code and promoting LMP, lecturing at MITLL and several other institutes, talking about the ideas underlying digital signal processing by the satellite modem. The United States Air Force invited Linkabit to demonstrate its technology on LES-8/9 experimental satellites. TRW had a head start in a few years to create an extended spectrum modem within the K-Band SATCOM AN / ASC-22 system, but their solution was expensive and terribly voluminous.
Linkabit struck the MITLL team by placing its relatively small system consisting of several rack-mount enclosures of 19 "each and setting it up for data transfer in just an hour - the lab staff would probably take a few days only to launch the basic mode. After another three hours they found an error in the MITLL specifications, corrected it by simple reprogramming, and set up data reception. And despite TWL certification and the readiness of its product, the general in charge of the program decided to finance Linkabit, a company that never produced equipment in the amount required for the defense industry - so that it finishes the development of its modem.
In addition to the excellent work of the LMP, the United States Air Force became interested in its other aspects, which became known in 1978. The real product requirement was the ability to install a dual modem [dual modem] on aerial platforms like the Boeing EC-135 and airplanes of the strategic command of the US Air Force, including the Boeing B-52. The solution, which gradually developed into a modem and a data processor for a command post (CPM / P), using several LMPs for dual duplex modems and transmission of control commands, as a result, fit into three strong 1/2 ATR boxes .
Linkabit grew by 60% per year. For the expansion of the company, additional capital was required, and they considered the option of selling shares, but then they received a proposal from another company engaged in radio technology, M / A-COM. In August 1980, the purchase of the company was completed. This radically changed Linkabit’s culture, and the free flow of ideas throughout the organization was replaced by a hierarchical process control framework. But this did not stop innovation. There were several important commercial products. One of them is a small satellite earth station Very Small Aperture Terminal (VSAT), a small business satellite communication system using plates with a diameter of 120 to 240 cm. Among the main firms that bought this technology were 7-11, Holiday Inn, Schlumberger and Wal-Mart. Another technology is VideoCipher, Satellite TV encryption system, which worked in HBO and other broadcasting corporations. Jerry Heller tracked the development and growth of VideoCipher technology throughout her life.
Jacobs and Viterbi discussed the acquisition of the company with M / A-COM director Larry Gould. As Jacobs wrote, "We found a common language, but Gould had a midlife crisis." Gould wanted to change the management system or merge with other companies - and his ideas didn't make much sense. The board of directors dismissed Gould (officially, “retired”) from the post of director in 1982. Jacobs was a member of the council, but he traveled around Europe and could not influence the decision-making about the new organizational structure in the way he wanted. Then he tried to split the company and put Linkabit pieces back, going so far in this that he even vetoed the deal with investors. At the last moment, the M / A-COM board of directors changed its mind and did not keep its promise to allow Linkabit to secede. Having finished work on the three chips of the commercial version of the VideoCipher II descrambler, Jacobs suddenly "retired" on April 1, 1985. Within a week, Viterbi left M / A-COM, and soon others followed him.
"Let's do it again."
But as a result, everything that happened was not like a pension. For a man who did not want to do day-to-day management at Linkabit, Irwin Jacobs did an excellent job as a director. Shortly after he left M / A-COM, one of his colleagues asked him, “Why don't we try to do it again?”. Jacobs took his family, with whom he promised to spend more time on a car trip to Europe, promising to think about it.
On July 1, 1985, six people gathered in the Jacobs' house — six people — all of them who had recently left Linkabit. In addition to Jacobs, there were Franklin Antonio, Di Kofman, Andrew Cohen, Klein Gilhausen and Harvey White. The legends say that there were seven of them: Andrew Viterbi was mentally present there, although he actually was on a European cruise until mid-July, before leaving, agreeing with the ideas of Jacobs. The core team chose the Qualcomm name for the new company, an abbreviation for quality communications. They were going to combine elements of the theory of digital communications with practical design knowledge to improve code division multiple access, or CDMA.
In the Shannon-Hartley theorem on channel capacity, Shannon illustrates that technologies using extended spectrum can reliably transmit more digital data with a wider range with a lower signal-to-noise ratio. CDMA uses a pseudo-random digital code to propagate a given data transfer over the entire allocated range.
Various assigned codes allow you to create multiple CDMA data channels operating in the same band. For any single channel, all its neighbors, working with a different code, look like they speak another language and do not interfere with the conversation. For outsiders without a code, this whole system is difficult to interpret, it looks like background noise. This made CDMA much more protected from listening or jamming than the primitive pseudo-random frequency tuning ideas put forward by Nikola Tesla and later patented in 1942 by actress and inventor Hedy Lamarr and her friend composer George Anteil.
Unlike TDMA system, using fixed channels that determine the exact number of transmissions that the base station could accommodate in the allocated range, CDMA significantly expanded its capacity. With the help of complex encoding and decoding technologies, Reed-Solomon codes and Viterbi decoding came into play - CDMA could significantly increase the number of users, bringing it to an acceptable level of digital errors and inter-channel interference. CDMA even re-uses the capacity released during pauses in a call — an ideal characteristic for mobile voice communications.
Coding techniques also generate a solution for multipath propagation in extended spectra. The RAKE receiver, developed by Bob Price and Paul Green from MITLL, was originally designed for use in the radar field, and used many correlators, called “fingers”, that can synchronize with different versions of the signal and statistically combine the results. The RAKE receivers made CDMA almost immune to noise between channels.
US Air Force, who planned to launch Satcom satellite"the first were fascinated by all the advantages of CDMA, but in order to manage to process all data in real time, considerable computational resources were required. Jacobs and Viterbi realized that they had a very valuable technology in their hands, whose performance was proved by the digital signal processing capabilities of LMP and dual modem that reliably processed CDMA data for satellite communications. Could Qualcomm satisfy commercial demands?
From the very beginning, two things were obvious: in commercial projects, cost plays a much larger role, and regulators like the United States Federal Communications Commission (FCC) take the stage when designing communication networks. Therefore, Qualcomm found itself in the same position as Linkabit - they worked on government messaging projects, trying to make equipment smaller and faster.
Government projects led to the emergence of a Viterbi decoder on a single chip. Finally, the CMOS ASIC technology has mastered this task, and there is no need to use hundreds of TTL chips and exotic techniques like ECL. Qualcomm developed the first chip by September 1987: Q1401, a 17 Mbps, 80 states, K = 7, speed 1/2. LSI Logic manufactured it using a 1.5 micron technology, on a crystal with an area of 169 mm 2in ceramic PGA with 155 contacts. It was available in commercial and military versions, and in the second version its speed was slightly lower in favor of a wider range of operating temperatures.
Space truckers
Shortly before the opening of Qualcomm, Viterbi had an interesting telephone conversation. He was called by Allen Salmazi — who left JPL in 1984 to found OmniNet — asking if their firms could work together on a new truck tracking system.
In 1984, the FCC allocated frequencies for RDSS (satellite radiolocation service). OmniNet had one license for RDSS, its competitor Geostar had another. The concept of Geostar was to transfer the position and messages from the truck over the decimeter waves (L-band) to the satellite, which were relayed by the company owning the trucks. If OmniNet turned out to organize RDSS based on the truck itself, it would be a big advantage.
Qualcomm was not sure of this possibility. Salmazi gave them $ 10,000 to research the project - he had neither clients nor investors (no one believed it would work, even in Geostar rejected the offer of cooperation), there was only money "from family and friends." OmniNet had to commercialize the idea in order to survive, and Qualcomm was her last hope.
There were few satellites in the L-band and they were expensive, in particular, because their signal processing system needed to be adapted for each specific mission. Satellites in the centimeter range (K u -band) used for VSATand other tasks, it was complete, they were cheaper, they allowed to process the signal on the ground, gave two-way communication, but they had one drawback. The FCC issued a license to use the K u -band only for fixed terminals with large ground-based parabolic antennas, which had to be sent to the target with an accuracy of 1-2 degrees. The second use of the range allowed mobile use only when it did not interfere with the basic option. A smaller satellite dish, especially on a moving truck, had to have guidance and aperture problems that would almost certainly create interference. Then Klein Gilhausen said: "We will use CDMA."
In theory, CDMA and the extended spectrum should have solved any problems with the interference on the transmitter side, and if the antenna was directed accurately enough, then reception should work. But now they doubted the FCC about this. Qualcomm convinced the FCC to grant it an experimental license covering the use of 600 trucks. Jacobs and the team created a unique directional antenna with a diameter of 10 "and a height of 6", while being very accurate. The signal processing was handled by a 4 ”x8” x9 ”module, and the display contained four lines of 40 characters each, and it also had a small keyboard and indicators for the driver. By January 1988, the system began limited testing trips across the country.
Having failed to find customers, Salmazi squandered all the money - so Qualcomm bought him, his company and the entire system by launching the OmniTRACS system in August 1988. Having received no complaints of interference, the FCC allowed the system to operate without restrictions. By October, Qualcomm had the first serious customer, the Schneider company, which owns 10,000 trucks. OmniTRACS successfully grew, and today this system uses 1.5 million trucks. This first important victory gave Qualcomm the necessary capital that enabled it to enter the next major market for CDMA.
Keep talking
Gilhausen buzzed Jacobs and Viterbi all ears with his idea of entering the CDMA-enabled cell phone market. Viterbi this idea seemed familiar - he represented it in the work of 1982 on the extended spectrum. The transition from a network of military satellites to several hundred B-52 and EC-135, and from there to private satellite networks with tens of thousands of trucks, was fairly straightforward - but the public cellular network had a well-known problem.
Although CDMA signals reduced the interference of digital channels, it was necessary to take into account certain characteristics of radio waves in the case when multiple transmitters simultaneously communicate with one ground station. For satellite communications, all terminals on the surface of the Earth were located far enough, and under normal conditions, the signals from each of them were approximately of the same power.
In a cellular network with low-power handsets, distance mattered, and the far / near problem turned out to be serious. This problem is related to the dynamic range of the base station receiver. If all the tubes work with the same power, the nearest one locks the receiver and forgets the tubes that transmit from longer distances from the tower, which is why they are not heard in the noise.
Viterbi, Jacobs, Gilhausen and Butch Weaver set about finding out the details. In the meantime, they worked with CDMA simulations, the telecommunications industry association (TIA) at the meeting in January 1989 chose DMA based on D-AMPSas a standard 2G connection in the USA. D-AMPS has become an evolutionary development of AMPS, and some argue that in this choice there was a share of nationalism - an alternative was chosen to the dominant GSM in Europe, despite its rapid development. FDMA was considered to be a low-risk approach (Motorola, AT & T and other companies faced it), but TDMA already showed its technical superiority in GSM ratings.
Few in the industry took CDMA seriously. The Cellular Communications Industry Association (CTIA) insisted on increasing the number of users in the 2G standard by at least 10 times compared with the capabilities of AMPS, but also required a smooth transition period. DAMPS did not meet the capacity requirements, but was considered the fastest way to implement 2G.
Capacity issues gave Qualcomm a chance. Jacobs contacted CTIA, presented her the results of a CDMA study, and, after an initial refusal, obtained a speech at a meeting of members of the association in Chicago in June 1989. He expected that the assembled experts would make a lot of holes in his presentation, but this did not happen.
One of the reasons for the success of the presentation was that the company had tested its technology with PacTel Cellular since February 1989. After the TIA vote, Jacobs and Viterbi started asking for negotiations with regional operators. “Suddenly, once Irving Jacobs and Andy Viterbi came to my office. Honestly, I don’t even know how they got there, ”said PacTel Cellular director Jeff Haltman.
However, William Lee, chief scientific consultant for PacTel Cellular, knew why they came. PacTel Cellular in the Los Angeles market quickly grew the user base, and pretty soon the company had to face insufficient capacity. Lee has been studying the efficiency of the extended digital spectrum for years and problems with capacity, comparing FDMA and TDMA.
What he saw in CDMA technology was about 20 times superior to analog systems - and the risks of developing TDMA were enough to justify the $ 1 million rate to fund Qualcomm's research.
Lee, like many others, wanted to see a working solution to the far / middle problem and other problems.
A little less than six months later, on November 7, 1989, Qualcomm had a prototype. The CDMA “telephone” - and really it was 15 kg of equipment - was stuck in a van, ready to travel around San Diego. Two “base stations” were launched to demonstrate the transfer of a call between them.
The Qualcomm team, in particular: Andrew Viterbi (left), Irwin Jacobs (center), Butch Weeher and Klein Gilhausen (right) with a CDMA van, approx. 1989
Before the assembled directors of the cellular industry, of which there were at least 150, and according to some data, all 300, William Lee gave a presentation, then Jacobs and his own, and then Gilhausen began to describe what visitors should see today. And when they were about to disband the group and start the demonstration, Jacobs noticed that Butch Weaver was waving fiercely at him. Glitch GPS has violated the synchronization of base stations. Jacobs improvised, and continued to spread about CDMA technology for 45 minutes, until Weaver and the team brought the system to a working state.
Many visitors were amazed by what they saw. Critics said that CDMA would never work, that the theory would not withstand the full-scale deployment of real-world conditions, and one expert even stated that "it violates the laws of physics." In addition, there was still a small problem with the placement of all the equipment in a small tube - but Qualcomm was ready to cope with it. In addition to the need for miniaturization and basic tasks of expanding the spectrum by direct sequencing and channelization, Qualcomm developed solutions for the three main problems of CDMA.
The first was the far / near problem. Dynamic power control changes the levels so as to maintain an adequate signal-to-noise ratio. CDMA tubes, located closer to the base stations, usually use less energy for transmission, and those that are farther away - more. As a result, all signals come to the base station with approximately one signal-to-noise ratio. Lowering the transmission energy also weakened the interference and saved the battery. Qualcomm used an aggressive open-and-closed-loop energy regulation algorithm that adjusts 800 times per second (later this number increased to 1500), which was much higher than the few times per second that GSM was content with.
The second was the transfer of the call. In the TDMA system, calls were usually disrupted when the user switched from one base station to another due to a hard transmission. CDMA phones established a connection with the next base station, not yet disconnected from the current one.
The third was a variable coefficient vocoder. Instead of encoding on / off in GSM, the variable-ratio encoder quickly adapted to natural pauses and conversation resumption, reducing the number of bits transmitted by the handsets, which increased the total capacity of the base station. This property in TDMA was not, in it the channels were fixed, and they could not be divided.
Climb and hold on
In the case of CDMA commercialization, Hultman promised support from PacTel Cellular, but other deals had to be made to reach a critical mass. PacTel introduced Qualcomm to higher-level company directors from other companies that emerged after the separation of AT & T’s monopoly, and the main cellular infrastructure manufacturers, in search of markets where CDMA would be useful. Qualcomm management has also made a fateful decision regarding the business model: instead of self-fabricating all the equipment, the company will sell licenses to use intellectual property using CDMA to manufacturers.
New York, where the telephone company NYNEX was operating, became another cellular market with problems of increasing capacity. Qualcomm brought its CDMA prototypes to Manhattan for field trials in February 1990. NYNEX has already ordered AT & T to look for next-generation infrastructure, and in early July, AT & T and Qualcomm agreed on a license for CDMA base station technology. On July 31, 1990, Qualcomm published the first version of the CDMA specifications for the collection of industry comments - Common Air Interface. On August 2, NYNEX announced that it would spend $ 100 million to create a “second cellular telephone network” in Manhattan by the end of 1991, mainly to allow time for the placement of frequencies and the construction of base stations. $ 3 million were to be withdrawn by Qualcomm to manufacture CDMA phones.
Other companies were not in a hurry with the offers. The two largest providers of cellular infrastructure, Ericsson and Motorola, planned to deal with TDMA networks. Motorola insured by entering into an interlicensing agreement with Qualcomm in September 1990, but publicly expressed concerns about technical issues. Operators, such as McCaw Cellular (the predecessor of AT & T Wireless) and Ameritech, tried to postpone a decision on the large-scale adoption of CDMA. As for other places, Europe made a bid for GSM based on TDMA, and Japan developed its own cellular network based on TDMA.
In the column of the undecided was Korea, which had no digital solutions. Salmazi achieved that Li from PacTel presented the company in August 1990, which resulted in consecutive discussions, which ended in May 1991 with an agreement on the joint development of ETRI CDMA. And although this program received extensive funding and promised large royalties in the future, five years were spent on its deployment.
But even after these victories, Qualcomm from a financial point of view continued to balance on the edge. Every dollar of profit went to the salaries of workers, whose number had already increased to 600 by 1991 - and to research and development in the field of CDMA.
PacTel continued to work on CDMA plans, which led to the testing of CAP I capacity in November 1991 using Qualcomm CDMA ready for commercial use. During the two-year program, five ASICs were developed. Three integrated circuits were required for a CDMA phone: a modulator, a demodulator, and an advanced Viterbi decoder. The other two were created for the base station, which also used the Viterbi decoder. These chipsets were connected to an external microprocessor. The tests showed good performance of the CDMA technology on a large scale, and proved that it can achieve the predicted capacities.
Qualcomm CDMA chipsets, approx. 1991
On the heels of the announcement of the success of the CAP I and ASIC tests at the CTIA technology forum, Qualcomm made the first exit to the stock exchange, offering 4 million shares and collecting $ 68 million in December 1991. PacTel bought a stake in the open market, and added another $ 2.2 million for the purchase of warrants for 390,000 shares in addition to guarantee continued research and development in the field of CDMA.
By the beginning of 1992, in addition to the Korean joint project ETRI in developing the CDMA standard with Qualcomm four companies have already decided to participate: AT & T, Motorola, Oki and Nortel Networks. The licensee number five in April 1992 was none other than Nokia - it was the culmination of a year and a half negotiations between Jacobs and Jorma Ollila. Nokia watched PacTel with great interest, and opened its own research and development center in San Diego to get closer to what's happening with CDMA. One of the stumbling blocks was royalties: it is believed that Nokia paid about 3% of the average selling price of the handset under the terms of its first contract for 15 years.
On March 2, 1993, Qualcomm introduced the CD-7000, a CDMA / AMPS-enabled phone running on a single narrow-band transmission chip: Mobile Station Modem (MSM). The phone was a typical chocolate bar form factor, 178x57x25 mm in size and weighing just over 340 grams. The first client was US West, with an application for at least 36,000 phones. Also in March 1993, four manufacturers announced their plans for CDMA phones and infrastructure in Korea: Goldstar, Hyundai, Maxon and Samsung.
Qualcomm unveiled the details of the MSM narrowband transmission chip at the Hot Chips Symposium in August 1993. The three main functions of a CDMA modulator, demodulator and Viterbi decoder were placed on a single chip made using 0.8 micron technology with an area of 114 mm 2 . It had 450,000 transistors, it consumed 300 mW, and to work as a tube, it still needed an external processor and circuit for working with radio waves. Qualcomm hinted that the supply of spare parts involved several different factories, but did not disclose suppliers - later it was reported that one of them was IBM.
TIA finally conceded, endorsing CDMA in the first publication of the IS-95 specification in July 1993; Commercially, this option became known as cdmaOne. A choice of digital standards for 2G has emerged in the cellular markets: CDMA, D-AMPS and GSM.
Obstacle on six millions
In the CD-7000 phone, along with the MSM chip, an Intel 80C186 processor was installed. The next logical step was their integration, but Intel was not involved in intellectual property. At first, Intel rejected Qualcomm offers. But under the constant onslaught of the sales department in San Diego, Intel, located in Chandler, pcs. Arizona learned all about Qualcomm, CDMA technology, and market opportunities, before finally agreeing on the delivery of the 80C186 core.
The task of changing the design of Intel 80C186 to a more standard one for the industry has proved difficult. Qualcomm developed MSM using high-level hardware description language (HDL) techniques that could be quickly reconfigured to various libraries, databases for simulations, and test vectors. It quickly became clear that it would be easier to transfer Qualcomm MSM IP to the Intel technological process, and give Intel the entire chip production. Qualcomm agreed to this. Intel was going to simultaneously engage in mobile and manufacturing businesses.
On February 1, 1995, Qualcomm announced a Q5257 MSM2 with a Q186 core in a QFP with 176 contacts, as well as an integrated Q5312 chip (Analog Baseband Processor, BBA2), which replaced 17 individual chips in an 80-pin QFP. These two chips made up most of the CDMA phone, such as, for example, the QCP-800, which was announced the next day. Qualcomm, in preparation for the release of large volumes, cooperated with Sony for the release of a new phone that supports two communication standards, and with a doubled battery life, which lasted five hours of conversation. It was also announced the release of a single Q5160 Cell Site Modem chip (CSM) for CDMA base stations, in which there was no integrated processor.
In June 1996, Q5270 MSM2.2 was introduced. The main improvements were the 13 Kbps PureVoice vocoder, using QCELP and giving better sound quality without increasing power consumption. It was offered in QFP format with 176 contacts for commercial use and in a larger one, with 208 contacts for debugging circuits.
Reducing power consumption was the goal of developing the MSM2300, which was announced in March 1997. Searching for a signal using hardware DSP worked up to eight times faster than that of MSM2.2. The 176-pin QFP had backward compatibility, which allowed equipment to be directly upgraded.
With the deployment of CDMA around the world, the number of chipsets produced has grown at an explosive pace. Qualcomm claimed that, in the total, shipments of various MSM variants — mainly MSM2 and MSM2.2, produced by Intel — reached six million units by June 1997. Intel also promoted its low-power embedded 386EX processors for Nokia and Research in Motion tubes [future BlackBerry / approx. trans.]. What could go wrong?
Probably, such a question Qualcomm asked herself when Intel refused to do the planned update of the embedded kernel. Frankly, the complexity of the production of the 386EX was much higher, and it was necessary to somehow place more chips from Qualcomm. Intel probably found the design too risky, and decided that six million pieces would not be enough to justify.
Qualcomm tried to hurry events, asked for the possible cost of this solution, and received a very superficial answer, without any major improvements in processor speed. (Intel, most likely, then just sued DEC about the Alpha chip. If Qualcomm had a need for a new kernel a little later, and if Intel had figured out the intellectual property or production business model for StrongARM, then Intel’s role on the mobile the market could have been completely different). And while the supply of existing chips continued, the phase of working with Intel on the next generation of chips for Qualcomm was over.
Roundabout search for improved nuclei
The high-performance processor core was not searched for long. Many users of Qualcomm CDMA licenses, in particular, LSI Logic, Lucent Technologies (split from AT & T), Samsung and VLSI, were proponents of ARM. Officially, Qualcomm announced the release of the first license using ARM in July 1998.
New chipset launches accelerated, and Qualcomm became one of the most fruitful suppliers of ARM-based chips, and its products were actively used in thousands of mobile devices. Below we will list only the key chip models.
When the ARM deal became widely known, the MSM3000 chip was already in the process of being developed, and its release was announced in February 1998, and the core was replaced with ARM7TDMI. Other enhancements included the SuperFinger demodulator, which accelerated data transfer to 64 Kbps, and improved standby mode. It was produced by the process of 0.35 microns. For the first time, the products were manufactured by TSMC . To avoid confusion with the old models, since the new required completely different software, the QFP with 176 contacts had a completely different pinout.
At that time, there was one more core in production. For quite a long time, DAC chips were present in the product line, and in February 1999, the MSM3100 with the ARM7TDMI core and the independently manufactured QDSP2000 programmable core were introduced. The QDSP2000 operational unit had a computational pipeline with five stages and optimized instructions for implementing a variable compression ratio codec and other functions, such as eliminating echo.
The 3G technology debuted on the MSM5000 chip supporting the updated cdma2000 specifications. Its release was announced in May 1999, and while still working on the ARM7TDMI core, it reached a speed of 153.6 Kbps and had improved search capabilities. The following year, MSM5000 was used in cdma2000 field tests, and its High Data Rate (HDR) technology would later develop to 1xEV-DO.
Flirting with Palm and a pdQ CDMA phone in September 1998 led to a survey of operating systems for smartphones. In September 1999, Qualcomm announced plans to develop iMSM chips designed for Microsoft Windows CE and Symbian OSs, including the iMSM4100 with a dual core of ARM720T processors, one for data transfer and the other for the OS. With the advent of StrongARM and other solutions, the iMSM4100 at the time of launch was ahead of them in integration but was slower in speed. Qualcomm was well versed in data transfer, but she still had a lot to learn about application processors.
The evolution of data transfer chips from Qualcomm
By the mid-2000s, there were three families of chips in development: 2G cdmaOne, 3G cdma2000, and prototypes of application processors, such as MSP1000 (essentially, iMSM with only one ARM720T processor).
Against the background of the presence of many manufacturers of CDMA phones, Qualcomm retired from this business, selling it to Kyocera in February 2000. After many years, during which Andrew Viterbi gave birth to new ideas, he announced his retirement in March. In May, Qualcomm announced that the total supply of MSM chipsets exceeded 100 million.
In February 2001, Qualcomm developed an ambitious plan. The MSM6xxx family development plan included a wide range of products, ranging from the initial MSM6000 chip based on ARM7TDMI with support for only 3G cdma2000. A set of Launchpad applications based on the new BREW API helped manufacturers to more effectively develop software. Modules such as radioOne have also been added to increase the efficiency of the Zero Intermediate Frequency conversion, and gpsOne to improve the positioning.
At the other end of the scale was the MSM6500, running on the ARM926EJ-S with two QDSP4000 cores, supporting 3G cdma2000 1xEV-DO and GSM / GPRS plus AMPS, all on the same chip. MSM6500 came out almost two years later, it was made using 0.13 micron technology, packed in a case with 409 contacts CSP.661,662 2003, and it marked the beginning of the change of the company leader. In January, Don Schrock announced his retirement from the post of head of Qualcomm CDMA Technologies (QCT), giving way to Sanajay Jha, who led the development teams of MSM.
The next plan was the MSM7xxx family, it was first shown in May 2003, and plans for it were similar - a wide range of entry-level chips to expensive ones. The 90nm MSM7600 version carried the ARM1136J-S with a frequency of 400 MHz and QDSP5000 for applications, plus 274 MHz ARM926EJ-S and QDSP4000 for multi-mode messaging. Also on the chip was a Q3Dimension GPU, from the IMAGEON license agreement with ATI. In 2006, MSM7600A decreased to 65 nm and received a frequency of 528 MHz. MSM7600, still going under the MSM sign, marked the change of direction for future Qualcomm application processors.
In September 2003, Qualcomm reached a milestone of 1 billion chips from the MSM line - nine years after the first commercial release.
Scorpion, Hexagon and Gobi
“Qualcomm has always been present in the semiconductor business,” was how Klein Gilhausen began his speech at Telecosm 2004. “We always knew that the key to implementing CDMA technology was a very aggressive policy for developing specialized chips.” Qualcomm's next steps were testing how aggressive a company could be.
Irving Jacobs resigned as director of Qualcomm on July 1, 2005 - the year of the 20th anniversary of the founding of the company - and became chairman of the board of directors. He was replaced by his son, Paul Jacobs, who was involved in speech compression algorithms, the launch of the pdQ smartphone, the BREW project and others. Stephen Altman, who led the licensing, replaced the outgoing president of the company, Tony Thornley. In general, the development strategy has not changed.
Paul Jacobs and Irwin Jacobs, approx. 2009
Many ARM licensees immediately supported the release of the new ARM Cortex-A8 core in October 2005. Instead of making the finished version, Sanjay Jha received the first architectural license for ARMv7 and revealed plans for the Scorpion processor core in November 2005. The loud headlines that this would be the first ARM core, operating at 1 GHz, were slightly exaggerated; Samsung has been promoting the ARM10 Halla design, operating at 1.2 GHz three years before. Nevertheless, Qualcomm has overtaken all competitors with its Scorpion, for example, TI OMAP 3 using Cortex-A8, and released its core design to the market two years earlier than Intrinsity Hummingbird.
The company's advantage came from the little-known acquisition of Xcella in August 2003 — it was a North Carolina company founded by former IBM employees, including Ron Tessitore and Tom Collopy. They made a huge contribution with their experience in developing processors.
The Scorpion used a 13-stage loading / storage pipeline, similar to the Cortex-A8, but it had two additional processing pipelines - one ten-step for simple arithmetic, and the other 12-step for multiplication with accumulation. SIMD operations in the VeNum multimedia engine had pipelines with a large number of stages, and the data width was doubled to 128 bits. Clock-do-Mania clocked logic, enhanced termination buffer, and other tweaks to optimize power consumption for the TSMC 65 nm LP process have led to energy savings of up to 40% compared to Cortex-A8.
DAC capabilities have also been improved. The Hexagon DAC core, also called QDSP6, also switched to the 65 nm process technology. It began production in the fall of 2004, and Hexagon used three techniques to save energy: Very Long Instruction Word (VLIW), multi-threading to reduce overhead in the absence of the necessary data in the L2 cache and a new set of instructions to maximize workload per package. The 64-bit vector execution unit processed up to eight simultaneous 16-bit multiplication operations with accumulation in one cycle. Three threads could run four instructions each cycle, two on double vector execution units, and two on double load / storage units.
Both cores under the new brand for application processors: Snapdragon. On November 14, 2007, Qualcomm showed a new QSD8250 with HSPA support and a dual-mode QSD8650 with CDMA2000 1xEV-DO and HSPA. Each had a Scorpion processor at 1 GHz and a Hexagon V1 DSP core at 600 MHz. Also on the chip were the Adreno 200 GPU (renamed after Qualcomm bought ATI mobile graphics assets from AMD in 2009), running at 133 MHz. Continued releases of the multi-mode combination ARM926EJ-S with QDSP4000.
Qualcomm flourished thanks to the fashion for "netbooks", and more and more often found itself competing with Intel and its Atom processor. WiMAX has become Intel's standard for broadband notebook connectivity, but it needed a new infrastructure. Taking advantage of the chance, Qualcomm unveiled its first Gobi chipset in October 2007 and used 65 nm MDM1000 to connect netbooks and similar non-phone devices to the Internet using EV-DO or HSPA over existing 3G networks.
Sales for use in PCs and netbooks immediately made Gobi a hit, and Snapdragon's popularity grew more slowly. Gobi began to inject resources. The MDM9x00 family development plan, announced in February 2008, contained a 45 nm process and a modem enhanced to support LTE, which later turned out to be based on ARM Cortex-A5. After Sanjay Jha left for Motorola in August 2008, Qualcomm raised Steve Mollenkopf to head of the QCT in order to maintain the former direction of development of the core strategy.
But it was time for a big change in mobile operating systems that Snapdragon was supposed to help. In September 2008, the T-Mobile G1, created by HTC, was the first Android-powered phone - and it worked on a Qualcomm MSM7201A chip. LG and Samsung worked on Android phones with Qualcomm chips to release them in 2009, and Sony Ericsson did not lag far behind them.
Snapdragon went further, to the second generation since. the 45nm process, introduced in November 2009. MSM7x30 were supposed to reduce the cost and energy consumption, and rolled back to use the 800 MHz core of the Scorpion with a QDSP5000 at a frequency of 256 MHz and a stranded Adreno 205 GPU. In preparation for dual cores, the 45nm version of Scorpion received debugging capabilities borrowed from ARM Cortex-A9 and improvements in the L2 cache. In June 2010, the third generation of Snapdragon MSM8260 and MSM8660 appeared, in which two Scorpion operated at 1.2 GHz, together with 400 MHz Hexagon V3, plus Adreno 220 GPU with improved efficiency. Shells were getting bigger; the MSM8x60 had 976 contacts, 14x14 mm in size and a nanoscale package (NSP) type.
Krait, Thirs and A / B Strategy
Qualcomm's method of working with the announcement of new products was usually to inform the media about an early version of development plans, and then release the finished product two or three years later. When the World Mobile Congress (MWC) began in February 2011, Qualcomm had a pair of aces up his sleeve for use in presentations.
First: Gobi switched to the 28 nm process in the form of MDM9x25. Among the improvements are the addition of support for Category 4 speeds, up to 150 Mbps on LTE FDD and LTE TDD, and support for HSPA + Release 9. Trial batches of these third-generation chips appeared at the end of 2012.
The second has already been partially announced twice. A couple of MWCs before that, Qualcomm mentioned MSM8960, a new version of Snapdragon, designed for multi-mode operation, including LTE. At an analyst briefing in November 2010, this chip was identified as a 28nm transition to the process, using the next generation of processor cores on the new microarchitecture, as well as a faster Adreno GPU. At MWC 2011, the first ARM processor with a 28 nm core was named: Krait.
It was announced that Krait would be the core used in three different chips. At the lower end of the scale was a dual-core 1.2 GHz Krait MSM8930 with an Adreno 305 GPU. In the middle was the MSM8960, a dual-core 1.5 GHz Krait with a faster Adreno 225 GPU. At the top end was an APQ 8064 with a quad-core 1.5 GHz Krait with an Adreno 320 GPU.
Independent cores in voltage and frequency allowed Krait to significantly save energy, up to 25-40% compared to the SMP approach, such as the big.LITTLE with ARM Cortex-A15, depending on the load. Advantages in efficiency, in particular, were achieved thanks to the 3-wide instruction decode, compared to the 2-wid of the Scorpion, and also thanks to the out-of-order execution, the 7th execution ports as compared to the 3, and the doubled L2 cache increased to 1 MB. This allowed Krait to rise to 3.3 DMIPS / MHz.
Trying to understand the heap of nomenclature created by them, Qualcomm at a meeting of analysts in November 2011 defined a hierarchical branding scheme. The new Krait 28 nm based chips are called Snapdragon S4, and split into S4 Play, S4 Plus and S4 Pro. 65 nm Scorpion were designated Snapdragon S1, 45 nm single-core Scorpion - Snapdragon S2, and 45 nm dual-core Snapdragon - S3.
Sometimes marketers excel themselves. Hierarchy is good, but a confusing nomenclature, which is difficult to translate from English, is not very good. A second attempt at CES 2013 led to the emergence of the modern Snapdragon license plate.
It was announced that the flagship Snapdragon 800 for high-end phones will contain a quad-core Krait 400 CPU at 2.3 GHz and a Hexagon V5 at 600 MHz and Adreno 330 at 450 MHz, as well as an LTE modem. The Snapdragon 600 had a quad-core Krait 300 CPU at 1.9 GHz with a Hexagon V4 at 500 MHz and an Adreno 320 GPU at 400 MHz, without a modem, for reasons of economy.
The launches that followed with CES 2013 fall into the category of Snapdragon 200 for phones, Snapdragon 400 for phones and tablets, Snapdragon 600 for devices of average performance, and Snapdragon 800 for high performance. The Snapdragon 200 line uses the ARM Cortex-A7 core to save money.
There was another example of not very successful marketing. Shortly after the unexpected launch of the Apple A7 chip with 64-bit support in September 2013, Qualcomm chief marketer Anand Chandraseker responded with great skepticism about its value to users. In the process of further study (and, possibly, after several annoyed calls from ARM), Chandrasekera was scolded, and his statements a week later were officially recognized as “inaccurate”.
The crisis was avoided, but it was not possible to come up with an answer. At a research meeting in November 2013, Qualcomm showed a fourth-generation Gobi development plan, moving to 20 nm from 9x35, supporting LTE Category 6 and carrier aggregation. In December 2013, a hurried presentation of the quad-core Snapdragon 410 with ARM Cortex-A53 brought Qualcomm back to the arena of 64-bit application processors.
It may have so simply coincided in time, but a few days after the presentation of the Snapdragon 410, there was a serious rotation of managers. Paul Jacobs announced that he would quit the Qualcomm director post, remaining chairman of the board of directors, and Steve Mollenkopf was temporarily promoted to director on December 12, 2013, with a permanent appointment next March, if shareholders approve his candidacy.
Steve Mollenkopf
In April 2014, the Snapdragon 810 was shown on TSMC 20 nm. Eight cores and a big.LITTLE scheme had four ARM Cortex-A57 cores at 2 GHz and four Cortex-A53 cores at 1.5 GHz. Also inside was the returning Hexagon V5 and its dynamic multithreading at 800 MHz, the Adreno 430 GPU at 600 MHz and the new support for LPDDR4 memory. Also included were a Cat 9 LTE modem, full 4K Ultra HD video support, and two graphics processors for computational photography. His younger brother, the Snapdragon 808, used two ARM Cortex-A57 cores instead of four, a simpler GPU, Adreno 418, and only supported LPDDR3.
Gobi chips on the fifth generation of the fifth generation became the main subject of discussion for analysts in November 2014. Gobi 9x45 supported LTE Advanced Category 10. This assumed a download speed of 450 Mbps by combining carriers in LTE.
On the development plan of Qualcomm, apparently, there was an A / B strategy - take the intellectual property of ARM where it is, add core of internal development, repeat the cycle. This is the only reasonable way to compete on a wide range of four options, from the lowest to the highest level. The Snapdragon 200 line combats an influx of ARM Cortex-A5 based chips from Taiwan and China, and the Snapdragon 800 and Gobi are fighting monsters like Apple, Intel, Samsung and many others.
What will happen after the phones?
Qualcomm's relentless improvement in chip design has brought tremendous success to CDMA and Android. In the growing market of cell phones, growing by 11%, 80% of which is occupied by Android, Qualcomm faces new problems previously unseen. Instead of celebrating the 30th anniversary, in July 2015, Qualcomm announced a reduction in the number of employees by 15%. Experts believe that this sad news was due to the fact that the 64-bit wave launched by Apple caught Qualcomm off guard, followed by a Snapdragon 810 overheating scandal from LG and Samsung.
Qualcomm vice president of marketing Tim McDonough has his own view of the Snapdragon 810 overheating history, saying that all decisions about phones are made 18 months before the public sees them - and, as we have seen, the main decisions on development plans Chips are accepted 18 months before. The latter is controlled by Qualcomm. The first is getting shorter than you might want Qualcomm. The source code hints that LG has switched from Snapdragon 810 to a weaker version of Snapdragon 808 - leaving the same LTE implementation - just a few months before the release of the LG G4 product. McDonough claimed that the problems were with the pre-release Snapdragon 810 (which had since been updated, after which overheating reports were gone), and that manufacturers are switching to the Snapdragon 808 because that they do not need support for full-size 4K video. The longest is the LTE modem evaluation, and this process has already started for a long time. That would make the transition - if it took place at LG - a quick and less painful one. Samsung could have had its own interests in indicating the existence of this problem - at that moment the company was preparing to launch the flagship Exynos 8 Octa.
The main parameters of mobile processors Qualcomm
Perhaps recent events are forced to be more cautious when placing development plans for all to see. At MWC 2015 in March, the main theme of the presentation was Snapdragon 820 with Kryo, the new 64-bit ARMv8-A CPU core. Details about the four cores and the clock frequency of 2.2 GHz appear (as well as rumors of even higher speeds) and the new production partner of Samsung with their 14 nm FinFET process. In August, plans were made for the Adreno 530 GPU and the new Spectra image processor for the Snapdragon 820; In addition, a new Hexagon 680 DAC is being developed.
Qualcomm November 10, 2015, when communicating with the media, reiterated that Snapdragon 820 consumes 30% less energy than Snapdragon 810. They also mentioned support at the system level, Cat 12 LTE, 802.11ad Wi-Fi and anti-malware. learning. Their marketing is moving away from the specifications of intellectual property to examples of using the capabilities of chips, which is good news.
Kryo creates a possible point for entering the emerging market of 64-bit server ARM. Competition with Intel and AMD on their field can turn into an interesting adventure. Qualcomm also pursues the Internet of things, having technologies as a result of the acquisition of Atheros and CSR, and develops software in AllJoyn. How Qualcomm will change the business model based on licensing complex communication algorithms will determine whether the company will remain the leader among non-manufacturing firms. Can they develop intellectual property supporting a new application segment, for example, drones? Is there more work on 4G LTE cellular, and how quickly will 5G technology be deployed?
Attempts by investors to divide the company into a division, one of which will deal with intellectual property, and the other with chips, appear ill-conceived. Although the part of the business that licenses intellectual property has a legacy in the form of cash flow from CDMA, the chip business benefits from strict adherence to the action plan. Without this synergy, what will fuel the chip business?
While mobile devices will use wireless connections, Qualcomm is not going anywhere. In the near future, complex strategic issues are to be resolved, and this can lead to a serious and extensive influence on the production strategy and competition in the application segments.