Breakthrough Ideas (continued, with additions about design rules and the power of modularity informed by Baldwin and Clark (2000)).
From Necessity Are Born ASIC Solutions
Because Qualcomm’s 3G air interface evolved from its 2G spread spectrum, using the same basic set of design rules, parameters, and interfaces to ensure generational compatibility, it is built on a history of prior success, demonstrated excellence that proved it excellence in cdmaOne networks around the world.
CDMAOne had a simpler design structure, but its operating features and parameters are included within the set that comprises 1X to insure a form of hierarchically nested compatibly within CDMA2000’s augmented and more complete set of design rules and advances the performance of certain parameters. CDMA2000 can revert to cdmaOne network’s parameters and design rules when CDMA2000 is unavailable. The 3G terminals can revert to the simpler cdmaOne design rules and parameters when only a 2G spread spectrum network is available. And, the cdmaOne handset can operate as a somewhat simpler and isolated spread spectrum 2G module within 3G 1X networks by using the parameters and rules of 2G spread spectrum that are a nested block within the larger 3G set.
These abilities are derived from the history of computers based on the principle of abstraction, the isolation, and segregation of certain functions into modules. The power of modularity itself is a system-wide driver of high technology. To separate and isolate functions into modular architectures permits human beings to design complex systems beyond the reach of the skill-set, perhaps even, of the architect and certainly beyond the understanding of most contributors to the open system of technology. This modularity means that complex systems can then evolve at a modular level instead of a system level, rapidly increasing the fitness of the system.
According to Baldwin and Clark (2000, p. 63,) “A module is a unit whose structural elements are powerfully connected among themselves and relatively weakly connected to elements in other units. Clearly there are degrees of connection, thus there are gradations of modularity.”
The process of abstraction in design and task structures is simply a matter of distinguishing levels in hierarchy and “hiding” the higher order rules from the lower order modules. The principle is the same as in the classic examples: the design rules of software that interfaces with hardware are isolated and separated from the “hidden” design rules of the computer’s architecture. The design rules of an operating system contain information that is hidden from the application, which interfaces by using the operating system’s proscribed design rules. When I said, “simply” in the first sentence, I did not mean that this is not a complex human undertaking, but instead that I am offering a “conceptual abstraction” as a device to foster understanding of an immensely complex practical engineering and management problem that require tools like a design structure matrix and a isomorphic task structure matrix to foresee, coordinate and manage complex undertaking by breaking them into smaller design units and isomorphic tasks. I meant only that abstractions help us understand complexity.
Baldwin and Clark (p. 72) indicated, “For a modularization to work in practice, the architects must partition the design parameters into two separate categories: visible information and hidden information. The partition specifies which parameters will interact outside of their module, and how potential interactions across modules will be handled.”
Thus, modules, which are at a lower level of the hierarchy can interface with a higher-level module, which shields the lower-level module’s operation be using an interface with specified rules for interaction, while keeping hidden set of information about its design rules that order its own tightly integrated functioning. The isolated and separated functions in the higher-level module--its hidden architecture--controls the nested hierarchical system because it is the progenitor of basic global design rules for the architecture of the system, as well as its parameters and interfaces. Although, the hierarchically higher module hides its interior design rules, it design rule specify interfaces that integrate a system through a standard set of interactive parameters for plug-and-play connection with lower-level modules. Understanding these principle is crucial for understanding what I mean by “strategic architectural control.”
Baldwin and Clark (76-77 continued by making these key points that are crucial to understanding Qualcomm’s strategic architectural control and the historically unfolding NTT FOMA and European UMTS problems in spread spectrum:
“Perfectly modular designs do not spring fully-formed from the minds of architects. When a design is first modularized, the designer’s knowledge of interdependencies is usually imperfect, and as a result the initial set of design rules will be incomplete. Incomplete design rules give rise to unforeseen interdependencies, which in turn will require consultations and iterations between the hidden-module designers and the architects of the system. The integration and testing of these designs will be fraught with difficulty and the designs themselves have high risk of failure.”
When you compare a political process where there is insufficient time and no procedures for testing design rules with the experience-based knowledge derived from Qualcomm’s years of experience in formulating workable design rules for spread spectrum, you understand the significance of an integrated learning base. Whether for CDMA2000 MSMs or the 5200, 6200, or 6250 MSMs for UMTS, Qualcomm: (1) rigorously tests performance before standardization; (2) develops standards that are focused on maximizing performance; (3) achieves faster commercialization because it enhances its existing interfaces, and (4) provides a firm roadmap for continually improved performance into the future.
Not only is Qualcomm the CDMA pioneer and 3G technology leader, but also its technology division (QCT) offers a unique and proprietary wireless solution that is singular and complete, including integrated circuits, radio and power products, system software, development and testing tools, and customer support using the expertise of its experienced system design engineers. Don Shrock proudly proclaims that QTC offers complete systems, not point products.
Thus, CDMA2000 1x remains an evolutionary technology that evolved incrementally through sustained growth along traditional trajectories within the spheres of accelerating trajectories in performance computers, power management, software coding, and modulation. How does QCT do it? By superior engineering that is made possible by, to use Brian Arthur’s term, deep craft knowledge of RF spread spectrum and how to translate those features into ASIC designs. According to SI poster “engineer,” who is a former Qualcomm executive (emphasis added):
“In 1990, I sat in a room with a rather smug AT&T microelectronics group who I was asking to help me make a DSP core for the new CDMA ASICS. They told me that it would take all their highly talented group so many hours to make a core which would run at 400 mW and run about 30 MHz to perform the CDMA voice coding. They then told me it was not in their marketing plans and that their vast technological group would not get to 30 MHz speed until sometime projected in 1994 and only then would they be able to help me out.
Disappointed that the giant AT&T would not help me out with this new technology, I came home and we brainstormed for a week or two. My team then came up with a proprietary DSP core which had a unique architecture, consumed only 30 mW and ran at 10 MHz, was 3 times smaller than the AT&T core would have been, and we ran in an entire technology feature size larger. In todays MSM, this core runs at 13 MHz and draws less than 2 mW!!
Hence was born the Qualcomm QDSP core which is still in use today. Out of Necessity, the solution was born.”
This innovative QDSP core is elegant¾using relatively few MIPs to achieve the desired high performance with exceptionally low power consumption. Elegant solutions require superior algorithms. Only second-rate RF or ASIC engineers try to solve a problem by throwing more amplifying or processing power at it.
In Machine Beauty: Elegance and the Heart of Technology, the Yale computer scientist Dr. David Gelernter (1998, p. 55) saw the algorithm, which is a procedure for computing something, as the heart of software:
“A good algorithm has to be powerful: has to function well, which usually means running fast and not requiring too much memory. The best algorithms are simple, too: a simple algorithm is easier to capture in software¾easier to program correctly, to understand, analyze, and improve. In short the best algorithms are the beautiful ones.”
CDMA technologies possess deep beauty. As Gelernter (who survived an attack by the Unabomber) teaches, beauty lies in the happy marriage of power and simplicity. This is true in theories, machines, and architectures. If you resonate to beauty, you can esthetically appreciate CDMA’s elegant marriage of Power and Simplicity¾from it born was Beauty Bare. Mathematics’ simplicity is the anodyne to applied physic’s real-world complexities. Complexity is present in spread spectrum because the physics of simultaneous fading, location, velocity, vocoding, power control, and usage of the RF signal must be translated from precise mathematical formulae into elegant software algorithms. In turn, these algorithms are hardwired into ASICs to speed and integrate the system.
The luminosity of the QDSP core algorithms shines forth: an elegant and inspired marriage of mathematics’ abstract simplicity with powerful software algorithms. Also, aesthetic simplicity in mobiles requires just-good-enough processing power and memory. The evidence of a successful CDMA union was apparent from “engineer’s” happiness at its nuptials, delight over beauty expressed as so-few-MIPs-to-do-so-much-more-mathematical-integration-at-less-power-consumption-and-cost.
QCT has produced over 750M CDMA chips, and is working on its eighth and ninth generation chipsets within a clearly defined roadmap, a chip-plan in which CSMs and MSMs are sampled and delivered on time. In 2001, Qualcomm became the largest fabless semiconductor company in the world, outsourcing chipset production to IBM, Motorola, Texas Instruments, and Taiwan Semiconductor.
“Engineer” posted that his FPGA layout was transformed into an ASIC in 18 months, right on the industry target. However, as always, extensive cross verification was required. According to Qualcomm, the developmental process for any new wireless system requires these steps: (1) Build and test prototype system; (2) Establish standard specifications for system and handsets; (3) Make revisions to the standard to stabilize it for manufacturing; (4) Test performance of standard releases; (5) Optimize system and handset performance; (6) Test interoperability of phones and infrastructures; (7) Test interoperability of any multimode/multiband systems; (8) Prepare chips and software for initial launch; (9) Early commercial launch; (10) Finalize chips and software for full commercial launch; (11) Full deployment; (12) Add rich features for multimedia, and (13) Ramp volumes to reduce manufacturing costs.
Given Qualcomm’s knowledge of how long it takes Qualcomm’s experts to develop a new generation of chips, this knowledgeable “engineer” is incensed when rivals impugn Jacob’s integrity instead of accepting his predictions of the timelines for new WCDMA chips. He said, “For history, the CDMA ASICS as FPGA [first step of building a prototype by designing and setting the field programmable logic gates] were done in 1990, the first ASICS from this in 1991; the first MSM chip done in 1993, the first phones into the commercial world in 1995; and high volume rollout in 1996.” Also, “engineer” noted that contemporary hurdles are higher now because today’s chips must have a higher degree of integration, and market competition demands a bill of materials under $100 in a sub 100 gm phone. Still more daunting, he continued, you couldn’t even model today’s complex MSMs in an FPGA because this requires more transistors than an FPGA can provide. Qualcomm is being generous, not conservative, in predicting no significant (10 million) WCDMA rollout until late 2004, particularly given that the UMTS Release 99 pre-final standards remain embarrassingly unfinished and untested. (I speculate that Qualcomm may have speeded up the appearance of interoperable UMTS handsets and base stations by its own development of a handset for testing and its UMTS chipset.)
When “engineer” was in charge of building the first ASICs at Qualcomm, he had 16 CDMA-experienced PhDs as system engineers who worked along side the chip designers to ensure that the wiring design of every integrated circuit precisely matched the RF system-level-design rules, fully exploiting the advantages of spread spectrum. In addition, the CDMA operating system software and ASICs require one another to operate effectively, making the integrated system firmly proprietary and more difficult to reverse engineer.
Without extensive knowledge of the intricacies and synergies of CDMA RANs, and the reciprocal interdependencies in chipset design, rival chip designers cannot match, much less surpass, the ratio of power to simplicity embodied in Qualcomm’s chips. If you have not played the 2G spread spectrum game, you are ill prepared to jump into the 3G spread spectrum game.
Because rival’s engineers are learning a new spread spectrum skill-set from the trial and error of experience, they cannot not know in advance the practical problems inherent in the complexities of system-design-and-functioning. One way to proceed is to simplify the problem. Commonly this means, say, focusing on improving operation performance and system integration, leaving the problem of creating efficiency in power usage until later. Not an elegant solution but a practical one when creating a working design and task-structure overwhelms you as you try to get a long-delayed new product out the door to fulfill your marketing promises.
Yet, very complex systems invariable demonstrate non-linear effects and reciprocal interdependencies. (That is another reason why a new transdisciplinary science of complex adaptive systems has been forming.) Yet, beautiful designs stem from years of engineering experience solving similar problems or from an exceptional architect’s intuition ( like Qualcomm’s Viterbi) of fundamental scientific principles. A system architect carefully prescribes a set of design rules for a modular architecture that proscribes any exceptions to preserve modularity. This means that you cannot, gerrymander patches that temporarily solve an unexpected problem of unwanted interactions between two modules. To do so fails to preserve the integrity of hierarchical design levels, hidden information, and modular structure itself. All designs start as patchwork quilts but must evolve into an integrated mosaic of plug-and-play system functioning in which the parts produce a whole that is more than the sum of their values. Modular architectures evolve more rapidly because they permit modular level testing and the proliferation of competition for economic value at the modular level. If the history of third-generation spread spectrum has not overcome all of the complexities of radio wave physics in RAN design, it still can create beautiful ASICs that handle the heart of the practical problems. If not full modularity yet, Qualcomm is well down the road because they took a spread spectrum road less traveled. It made all the difference.
It is crucial that the investor understand that DSPs are general-purpose devises that are optimized to perform rapid multiplication and accumulation (MAC) because MACs facilitate useful mathematical functions (Fourier transforms) in digital communication. Whereas, Qualcomm’s ASICs are customized CDMA signal processors, using MACs where needed. Their expertise is the knowledge codified in their beautiful QDSP algorithms for spread spectrum, power control, soft handoffs, vocoders, and the like, and in the elegant designs of its ASIC and RAN system architectures.
Therefore, the architecture of spread spectrum is fully and elegantly integrated from theory to RAN design, from the mathematical formula of radio wave physics to algorithms in the MSM and CSM ASICs, from spread spectrum architecture into its expanding platform of augmenting and coherent architectures. Qualcomm’s manifest competence in generating various ingeniously powerful, but simple, solutions, when taken together, forms a significant barrier to entry. Because RF systems are always long lived¾with each new generation being 10 or more years apart, with some 1G systems still in place after about 20 years¾and its singular integrated and powerful proprietary RF system has little competition within the CDMA mode, Qualcomm has created high switching costs.
The specific design challenges in introducing CDMA ASICs were developing efficient algorithms and feature rich protocol stacks¾layers of protocols-as-language-codes that abstract and isolate the OSI functions that make the complexities of a modern communication system transparent to users. As always, the mobile context itself required reducing power consumption and feature size. Shrinking line- and feature-size itself within an IC reduces the power required to operate an MSM. Qualcomm uses an ARM CPU, whose design also maximizes processing power as it minimizes the draw on battery power, to handle basic computing and application functions. In a past joint venture with RF Microsystems, power management chips were designed with unique proprietary features to conserve power. Memory is also reduced to a minimum to save the draw on electrical power. In itself, this just-good-enough principle in conserving power and reducing form factors reveals that elegance is QCT’s design credo. But above all else, the elegant integration of the CDMA RF system adds intrinsic value.
This is why “engineer” was so pleased with Qualcomm’s accomplishment in creating a “unique architecture” that required only just-good-enough processing speed and little battery power, a power consumption that is currently much less than its competitors. Qualcomm’s MSMs and CSMs are indeed highly intelligent communicators because they have a unique low-power-consumption architecture that enabled advanced power control within a CDMA wireless solution that transforms “processing gain” into “interference reduction” across the spread spectrum. Taken together, their integrated wireless solutions resonate, creating a symphonic architectonic beauty.
Where will CDMA innovations go? Rapid progress is being made along traditional trajectories of chip performance. Following Moore’s law, the price/performance curve of DSPs doubles about every 12-18 months. Increased processing power permits more ingenious codes—using algorithms that are optimized for specific purposes. When combined together, advances in radio technology with advances in processors from placing transistors ever-closer together on ever-larger wafers, wireless technology promises continuing improvements, from intelligent antennae to BLAST, from boomers to spot beams, from diversity innovations to cancellation techniques.
At the 2000 Wireless and Optical Communications Conference, the program announcement stated, “Tremendous cost and size reductions in digital signal processing make it feasible to use very sophisticated and highly adaptive algorithms from modulation theory, information theory, and antenna array processing in this revolution which will enable wireless capacity to increase 100 fold.”
The increasing rate of discovery in science becomes one of the ultimate drivers of high technology and economic wealth creation. For instance, tomorrow’s installment presents the second breakthrough idea that transformed this already beautiful CDMA 1x wireless architecture by creating a sister! |
