|
|
 |
 |
When Microsoft announced plans to build the Xbox game console in March 2000, Sony Computer Entertainment, the group responsible for the Playstation2 console, knew that it faced an entire new level of competition in the $9-billion-a-year computer gaming market. Whereas Sony had built the Playstation2 around a specialized chip architecture, Microsoft intended to use standard Intel CPUs—the same architecture used in hundreds of millions of personal computers.
In the short term, Microsoft's approach made little difference, because the two consoles were roughly comparable in performance. Long-term, though, Sony knew that it was in trouble. As a chip design and manufacturing house, Intel had economies of scale that Sony could never match. Beating Intel at its own game isn't a trivial undertaking, as companies such as Motorola, Cyrix, and Transmeta have all learned. How would Sony remain competitive with Microsoft's future Xbox products based on Intel's low-priced chips?
 Sony's solution was to ally itself with Toshiba and IBM in what's arguably the most ambitious CPU chip development effort of the past decade. The three firms are designing what they're calling a supercomputer-on-a-chip, so fast and so powerful that it can (theoretically, at least) leapfrog Intel's technology and act as the brains not just for gaming consoles but also for an entire range of broadband devices. While the trio isn't making specific promises, the chip's performance is rumored to be in range of a teraflop, roughly 100 times more powerful than today's Pentium 4 chips.
Since the announcement of the alliance, in March 2001, the three partners have been holding their cards close to the vest, ignoring repeated information requests from multiple business and trade publications, according to IBM sources. Now, Electronics Design Chain takes you behind the scenes at the birth of a CPU chip design project that, if the partners involved can deliver on what they promise, has the potential to threaten the hegemony of the Microsoft/Intel alliance.
Phase 0: The Vision
Although Sony's Playstation2 game console looks likely to hold its own against Microsoft's Xbox, Sony executives were well aware that the Microsoft/Intel axis posed a major long-term threat. At stake was more than just the computer gaming business, though. Both Sony and Microsoft had their strategic eye on the home gateway, a successor to the game console that would someday offer a range of broadband services, from on-demand television to real-time video chats. For Sony, this long-heralded convergence of computers and consumer electronics represents the Holy Grail. Even though that convergence has been promised for many years by a multitude of different companies, the fact that Microsoft and Intel were now positioned to enter that market posed a serious threat to Sony's future plans.
 Faced with fighting the most successful partnership in the computer industry, Sony needed allies. Sony had previously allied itself with Toshiba to create the CPU chip for the Playstation2, but the current situation called for some extra muscle. Sony turned to IBM, a company with long experience in chip development. "The processor platform that people have only been able to imagine is now going to become a reality," promises Ken Kutaragi, president and CEO of Tokyo-based Sony Computer Entertainment Inc. (SCEI). The vision is to "combine IBM's processor technology, Sony's leadership in the computer entertainment world and Toshiba's total system LSI technology to develop a new type of microprocessor that is not described on any existing road map," explains SCEI vice president Kenichi Fukunaga. "Our target in this project is to develop a totally new kind of microprocessor that will become the de facto standard for broadband network equipment," echoes Katsuji Fujita, executive vice president of the Tokyo-based Toshiba Semiconductor Co. "We expect that it will be applied to any systems that connect to broadband networks, such as network servers, home gateways, mobile terminals, and PDAs."
Each of the partners brought key strengths to the project, according to Chekib Akrout, vice president of development for IBM Microelectronics. "The three companies are complementary in what they bring to the table," he explains, "Sony has a strong presence in the game market and content definition, Toshiba brings valuable perspective on the technical challenges of chip design for consumer electronics, and IBM is strong on high-performance processor development." The three companies decided in early 2001 to spend $400 million over five years to develop the new CPU, now code-named "Cell." IBM would lead a design team of 400 engineers, the majority of whom would come from IBM but with strong contingents from Sony and Toshiba.
 |
|
| "It's going to take a major marketing effort to forge the alliances and applications that are the preconditions for success." Rob Enderle, Research Fellow, Giga Information Group |
|
It was now up to these engineers to turn Sony's idea of a broadband-oriented powerhouse from a vision into a practical reality.
Phase 1: Conceptual Design
 After much consultation, the three partners tapped IBM Fellow Jim Kahle to head the design team, which is headquartered at IBM's STI Design Center in Austin, Tex. Kahle has extensive experience with chip design and was project manager for the portable version of the PowerPC processor in the Apple G3 PowerBook series. Soft-spoken and quietly competent, Kahle discovered he would need to be as much a diplomat as a manager in the difficult months ahead. He compares the partnership favorably with the relationship that IBM had with Motorola when Kahle was working on the PowerPC chip. "IBM and Motorola were too close to one another's business models to cooperate very effectively," he recalls, adding that because the three current partners are in such different markets, "we don't trip over each other's feet as much."
Kahle quickly realized that this project was unusual in that the designers were starting with a blank slate. "Most chip design projects involve some kind of legacy design and a requirement for backward compatibility," he says. In this case, however, the engineers were chartered to create a completely new design that could satisfy a speculative view of what might be needed in the future. "We were designing not to a preexisting set of requirements but to a vision of what the partners were trying to achieve." Because of this, the design team first needed to define, at a conceptual level, a completely new chip architecture that would actually do what the high-level executives from all three companies wanted.
Kahle turned to IBM Research, an organization responsible for numerous experimental products and concepts, including the famous Deep Blue computer that defeated world chess champion Garry Kasparov. "We were able to draw on the experience of engineers with twenty years' experience in experimental chip design," says Kahle.
The biggest challenge, according to Akrout, was coordinating inputs and ideas from different levels of all three companies. "The presence of engineers from other firms made the team unique within IBM, because they were bringing different corporate cultures into the mix," he explains. There were also cultural differences in the way the American engineers and the Japanese engineers approached the design process. "Japanese engineers prefer to have all the data in place before making a decision," explains Akrout. "The American attitude seems to be that 'when you go for the leading edge, you're always going to be close to the edge of the cliff.'" As a result, American engineers tend to generate more breakthrough products whereas Japanese engineers tend to be better at refining and turning already invented technology into products.
Kahle found that doing the conceptual design also required forging these very different corporate and geographical cultures into a team that could work together productively. To do this, he transformed the conceptual design phase into a daily dialogue between the various members of each company and culture, in the hope that by bringing all the elements together, he'd be able to combine the best of American enthusiasm with the best of Japanese meticulousness. "I've tried to create an environment that brings out the best in everyone," he says.
He believed that the only way to effectively manage this project would be for a single company, IBM, to take a leadership role, and he was able to convince the partners to place Sony and Toshiba engineers under IBM management at the Austin facility.
The selection of a single design methodology also helped bring the team together, according to Kahle. "Because IBM was taking the lead in the design effort, we decided to use a design chain methodology with which IBM was already familiar, based on its experience in designing PowerPC and other chip architectures," he explains, noting that "as the number of transistors on a chip has gone up, IBM has implemented a certain discipline in the design process that allows projects to proceed quickly through the various design stages."
|
|
| "The vision is to develop a new type of microprocessor that is not described on any existing road map." Kenichi Fukunaga, SCEI |
|
 The team began working with a variety of collaborative tools, ranging from industry-standard EDA to Lotus Notes for e-mail and collaboration, to specialized tools that IBM had previously developed especially for chip design. IBM is thus acting as the lead partner in the chip design, with Toshiba and Sony functioning as the manufacturers and consumers of the chip that share a vested interest in the scope of its final functionality.
Within a few months of the project launch, Kahle had successfully refereed a conceptual design that satisfied all the stakeholders. The stage was set for the engineers to begin digging into the actual work of designing the chip.
Phase 2: High-Level Design
It's a scenario that's all too familiar to design engineers: tackling an aggressive set of development milestones combined with requirements for rich functionality. Kahle realized that the only way to achieve these ambitious goals was to create an iterative design process in which quickly generated successive approximations swiftly closed in on an optimal design. This was no leisurely academic exercise: The engineers had to come up with a fast design—fast. "There's an incredible danger at this stage that the engineers might get locked into a particular way of thinking," he explains. "The last thing we needed was for engineers to start saying, 'I can't change it' and threatening to throw the schedule into jeopardy."
Kahle made it clear early on that rigid attitudes on the part of the engineers would not be tolerated. He emphasized a rule of thumb: "You've got to design at least three times before you get it right." Defining the design process in this way helped engineers from one part of the team accept more help from engineers on other parts of the team. For example, when one group of engineers who were developing a pipeline were suddenly confronted with a whiz-bang idea from an engineer outside the group, Kahle encouraged the original engineers to see the new idea as an asset, rather than a scheduling issue.
Kahle admits that it was something of a struggle to get some of the engineers to commit to that level of openness, but he remained convinced that design flexibility was essential to preventing any stovepiping that might otherwise limit the effectiveness of the final chip. "I've learned that if you optimize for only one part of the system, the entire system becomes suboptimal through the creation of bottlenecks," he explains, "I needed to make certain that 'not invented here' stayed out of the mix." Kahle also had to remain aware of the eventual manufacturability of the chip at this point, but elected to put the main burden of this part of the effort on the implementation phase that would follow.
Kahle used daily architecture brainstorming meetings as a forum for keeping ideas flowing. He directed some of his engineers to work from the bottom up, drafting circuit designs for specific functional elements of the chip, while other engineers were working from the top down, modeling the high-level architecture and providing the definition of the chip instruction set. "We needed both views because the project was so complex," he explains. He also instituted multiday design reviews at major project milestones, to be certain that the project remained on target. These meetings were relatively smooth, because the design engineers were working at the same location, making it easy for the teams to communicate freely.
While Kahle was refereeing the battles in the design center, Akrout was keeping top management informed of the project's progress. This was an important task, because the strategic nature of the project had attracted interest at the highest levels of all three companies. For example, SCEI president and CEO Ken Kutaragi has never missed a quarterly review meeting. Mitsuo Saito, a chief fellow in Toshiba Semiconductor Company who was instrumental in the creation of the Playstation2 CPU, has also remained heavily involved. "It's unusual to see such high-level executives take such a detailed interest in a project," says Kahle, "but Kutaragi-san and Saito-san have very clear ideas of where they want the project to go."
It was under these watchful eyes that Kahle moved the project to the next level, from the high-level design all the way down to the actual physical layout of the PROM.
Phase 3: Design Implementation
As of this writing, Kahle's team is busily building a set of complete schematics for the masks that the factories will use to manufacture the Cell chip. The manufacturing design will utilize 90nm geometry and silicon-on-insulator (SOI) technology, which increases transistor switching speed by reducing capacitance (build-up of electrical charges in the transistor's elements), and thus reducing the discharge time and power requirements. As with the high-level design, the implementation process is highly iterative, with the engineers building simulations of how each section of the chip will perform under real-life circumstances. "The trick is to achieve the right stability of the chip's smaller pieces," explains Kahle, "and then to build up bigger and bigger chunks, thereby improving the overall quality and stability of the design."
 When he speaks about the Cell project, Kahle exudes the kind of confidence one normally associates with a manager whose project is proceeding apace. Rather than worrying about schedule, Kahle's main concern is the competition. "We need to get this project completed before our competitors get a shot at building something similar," he says.
Indeed, even if the Cell design project is completed on time, the three companies may find it difficult to transform it into a successful business, according to Rob Enderle, research fellow at Giga Information Group, a Cambridge, Mass.-based research firm. "The market for CPU chips is already in overcapacity," he says, "and it's going to take a major marketing effort to forge the alliances and applications that are the preconditions for success." Enderle notes that attempts to displace the Intel architecture with alternatives such as the PowerPC haven't been wildly successful.
Enderle also points out that Sony has a spotty record in establishing de facto standards. "Don't forget about the Betamax," he says. "Sony's not a company that understands how to play by industry rules." He points out that problems may occur down the line when the three companies try to work together to bring the chip to market. For the time being, the potential problems with the partnership may be masked by the administrative decision to put IBM in charge of the design effort. There's always a danger that the relationships may become frayed when the project opens out into the larger market.
 These problems may be a few years out, though. IBM has not announced when it will release the final chip, but the Cell project was originally envisioned as taking five years, suggesting that the final product may not be ready until as late as 2007. Although Sony refuses to talk about its future plans for the chip, it does admit that the Cell chip will not be the CPU in the Playstation3, reportedly slated for release in 2005.
Dan Kusnetsky, vice president of Framingham, Mass.-based research firm IDC, believes that it would be a big mistake to sell the Sony/Toshiba/IBM alliance short. "The high-performance computing community is one of the few places where raw performance is one of the key factors in selecting a platform," he says, "A platform that is priced competitively has a chance to make some inroads into this market."
Brooks Gray, research director at Hampton, N.H.-based Technology Business Research, also believes that the chip may do well in the market. "This is a viable strategy if the chips are going into Sony devices," he says, warning, however, that Intel is also working on architectures that are conceptually similar to the Cell chip and will presumably compete head to head for the same applications.
The broader market is where the ultimate fate of the Cell chip will eventually be decided. But whatever happens, Kahle will always be enormously proud of his involvement. Even as he works to keep the design process on track, he's well aware that the biggest challenges probably lie in the future. "What we're doing—starting a whole new architecture from scratch—happens very rarely, perhaps once a decade," he says. "There are no easy blueprints for getting it right."
|
|
ABOUT THE AUTHOR |
|
 |
|
Geoffrey James (gjames@designchain.com) is a widely published high-tech journalist based in New Hampshire and the author of numerous books. His writing has appeared in Upside, Red Herring, CIO, and the New York Times. |
|
Back to the top
|
|
