These newer companies extend their economic reach not through big factories and big payrolls but by establishing alliances, both with one another and with the industry's largest corporations. So close are these ties -- and so numerous -- that the lines separating one business from another can begin to blur. Consider the examples of Altera, on the one hand, and Weitek on the other.
Altera, whose PLDs are widely considered state of the art, has no chip-production facilities of its own. Instead, it has long-term manufacturing agreements with such companies as Intel, Texas Instruments (TI), and Cypress. The manufacturers, in turn, get the right to make and sell some of Altera's innovative devices for their own accounts.
These relationships don't look much like conventional procurement contracts. In a conventional contract, the buyer hands the specs to the supplier and says we'll take delivery in six weeks, or maybe every six weeks for a year. End of discussion. When Altera signed its agreement with TI in 1989, by contrast, it initiated a relationship that would last a minimum of seven years. Altera had a design for a new family of products but knew that at least a year of testing and refinement would be necessary before the products could be brought to market.
So the two companies began to work on the products together, with people and information moving freely across corporate borders. "We're in the development phase with TI now," explains Altera president Rodney Smith. "We put runs through their fab, generate test programs, generate the software to support [the part]. Within the year we'll be figuring out the process technology, and which fab it'll be manufactured in." Engineers from both companies work jointly on all these issues. By the time the new product appears on the market, they will have spent thousands of hours in direct collaboration.
Plenty of companies contract out their manufacturing when they're young, then build a plant of their own once they get big enough. Altera, however, has become more rather than less intertwined with its manufacturing partners over time. Only last April, for example, Smith took the unusual step of buying equity in one of his suppliers: he paid $7.4 million -- and has agreed to go as high as $15 million -- for an interest in the Cypress subsidiary that runs the company's most up-to-date fabrication facility.
It's a deal with remarkable benefits for both sides. As a minority owner Altera gets a guaranteed fraction of the fab's output at cost plus, along with full access to the information it needs to determine what cost really is. It also gets early access to Cypress's next-generation manufacturing technology, an area in which Cypress is a leader. Cypress -- which already has the rights to produce and sell some of Altera's products -- gets a sizable cash investment, plus a chance to run its fab closer to capacity, thereby lowering costs. And it has rights to the flagship products Altera develops in the future.
Weitek, the specialist in math-intensive chips, presents a different but no less complex model of a relationship-based company. Like Altera, it has no fab of its own, relying instead on manufacturers both here and in Japan. But the company's relationships extend into customers' shops as well as into suppliers'. Even more than Altera's, they define its way of doing business.
Rather than sell to large numbers of customers all over the world, as Altera does, Weitek designs high-performance chips for specific applications. Its chips can be found in Hewlett-Packard computers, for example, and in the workstations produced by Sun Microsystems. So Weitek chip designers have to work closely with the computer engineers who will utilize their products.
Closeness like this we all should have. When Sun needed a so-called floating-point chip for its SPARCstation 1, it asked Weitek for a design -- then sent two engineers and two costly computers over to Weitek's building to help in the development. Hewlett-Packard (HP) began buying Weitek chips -- then offered to manufacture them for Weitek in its state-of-the-art fab. Not that Weitek president Art Collmeyer has allowed his 240-employee company to be swallowed up by Sun (11,000 employees) or HP (93,000). He happily sells chips to Sun's competitors as well as to Sun. And when he agreed to use HP's fab, he did so on the condition that he could sell chips made there to other customers. Giant HP now finds itself in the peculiar position of manufacturing chips for someone else -- Weitek -- that will wind up in a competitor's computers.
So where are Weitek's borders? Eric Larson, marketing manager for HP's circuit technology group, seems to define his own giant corporation almost as a continuation of the smaller company. "We recognized that teaming with [Weitek] could provide us with access to the leading-edge technology they had," says Larson. "We could do that by becoming a supplier to them, really an extension of their manufacturing operation."
* * *
In an environment as competitive as semiconductors, there are obvious benefits to specialization. Altera, for example, is by common consent the number-one company in the design of PLDs, and as such attracts the best and brightest programmable-logic engineers. Its $10-million R&D budget goes exclusively to new-product development in this field -- an expenditure matched by no one else. Relationships are a concomitant of specialization. Cypress can boost its own new-product portfolio by licensing some of Altera's, and so can spend more money on cash-eaters such as a new fab. Result: faster introduction of new products and new manufacturing processes all around.
But a relationship-based strategy has virtues of its own as well. Relationships allow an otherwise small company, for example, to leverage its presence in the marketplace. If Altera or Weitek had the same sales per employee as an integrated company such as Hewlett-Packard, the smaller companies' work forces could generate only a fraction of the revenue they do now. Relationships also help minimize fixed costs, no small matter in a cyclical business. Altera may have to spend as much as $15 million for a share of Cypress's fab. But to build a fab of its own would cost two or three times as much -- and the fab would have to be kept busy through thick times and thin.
