It was a lousy business and everybody knew it. Several semiconductor companies made custom integrated circuits but, except for American Microsystems Inc., it was not their main business. It was too engineering-intensive and there were too few customers. It was in the nature of integrated circuits in the 1960s that they made economic sense only for high-volume applications. So semiconductor manufacturers tried to develop circuits that lots of people would need, like logic circuits, amplifiers and A/D and D/A converters. Custom circuits made sense only for huge volumes and not many companies needed a large volume of anything. Most semiconductor companies turned away custom-circuit business because it diverted engineering from proprietary new-product development, which was likely to have a bigger payback.

When C. Lester Hogan left Motorola Semiconductor in the summer of 1968 to become president of Fairchild Semiconductor, he was joined by a group of leading Motorolans who became known as Hogan's Heroes. They found that Fairchild had a huge R&D department. Wilf Corrigan, one of the Heroes, remembers that there were hundreds of people in R&D. And one of the activities that absorbed some of their time was custom design.

They used a powerful IBM mainframe computer to automate IC design. The computer wasn't very powerful by standards of a couple of decades later, but at a cost of about $5 million, it was mighty powerful.

That custom-design effort evolved by the early 1970s into what Fairchild called MicroMosaic, an array of as many as 300 PMOS or NMOS gates. The customers for these huge arrays made calculators. And there, indeed, was volume to justify custom circuits. An individual calculator might have three, four, maybe as many as seven chips. And there were lots and lots of calculators.

It turns out that almost any competent semiconductor manufacturer could produce a single chip to do the basic add-subtract-multiply-divide functions of what were widely known as four-bangers. But no calculator manufacturer would consider it. Each demanded its own custom design, though customers mostly couldn't tell the difference. The calculator manufacturers insisted that they needed product differentiation.

When outfits like Texas Instruments and Mostek started offering standard chips for calculators, calculator makers insisted at first that they couldn't consider such chips, though they offered economies of time and dollars. But rather quickly the calculator manufacturers changed their attitudes and adopted standard chips.

So the semiconductor companies with custom-design ambitions looked to other markets. The computer market continued to want custom chips and it was a big market. The computer people wanted speed, and that meant emitter-coupled logic. But there were too few major computer makers to make this a major market worth pursuing.

The semiconductor manufacturers may not have thought of it in those terms, but it was obvious that they would create custom chips only for high-volume jobs. A customer would approach a semiconductor company with a requirement for a custom chip and the semi manufacturer would normally decline. The job took up too much engineering time, for which the customer did not pay. The payback was exclusively in manufacturing chips — lots of them if there was to be any profit in the job.

By 1974, Wilf Corrigan was president and CEO of Fairchild Camera & Instrument Corp., parent of Fairchild Semiconductor, and he saw that the market for custom MOS chips was dying. The calculator manufacturers were switching to standard chips. Fairchild continued to make custom circuits, using emitter-coupled logic, for computer manufacturers. The military, another possible customer for custom chips, looked upon MOS, a rather young and touchy technology, as a toy technology. It had lots of problems, it was hard to manufacture and it certainly wasn't reliable enough for military circuitry. The military was content to stick to standard TTL.

Corrigan reckoned that the custom-chip part of Fairchild had a capability that might prove handy from time to time, but it did not make sense as a business group. So he scaled the activity back and more or less put it to sleep.

The only major company that continued to pursue custom design was American Microsystems Inc., which had lots of experience in calculator chips. AMI engineers knew how to handcraft custom designs.

In 1979, Gould made a hostile takeover bid for Fairchild. Corrigan sold the company to the white knight, Schlumberger, and made a handsome profit for Fairchild stockholders. Corrigan told Schlumberger that he would stay around during the transition and he agreed not to compete for a year.

A rare approach

Though the agreement banned competing, it said nothing about preparing to compete—which is what he did. If it weren't for that agreement, Corrigan admits, he would have started a semiconductor company right away because that's all he knew about. He would have figured out later what the company was to do. But because he had all this time, he approached the startup idea logically, a rare approach then — and still.

He visited his old customers, said he wanted to start a semi company and asked for their advice. In general, they were discouraging. The game was over, they said. Only the big guys would be left. Then they told him what they didn't need. They didn't need another memory company, another microprocessor company or another company making ECL or TTL.

But what did they need that they weren't getting, Corrigan persisted. And they said they needed custom circuits. He should know that, they pointed out, because every time they asked for custom circuits when he was at Fairchild he declined to supply them.

By the time Corrigan heard this tale from seven or eight customers, he realized that customers really needed custom circuits and it was a business that nobody was addressing in a systematic way. The semi companies didn't want to design custom circuits unless there was large volume; they made money only on the chips. Everybody knew that.

So Corrigan offered a ridiculous idea. "Suppose I charge you for the nonrecurring engineering. Then I won't care how much volume you have." And the customers said that would be OK.

So now Corrigan had to create a company. He called Rob Walker, then with Intel, and one of the engineers he remembered from his Fairchild days. He asked Walker if he could assemble the old custom-circuit group for a meeting. These people had flown in many directions and were doing many different things.

When they finally met at Corrigan's home and he explained that he wanted to start an ASIC company, the response was not surprising. "You were the guy who shut us down. And now you want to start a company to do what we did 10 years ago?"

Corrigan presented his arguments: The markets had changed, there were many more potential customers now. Ten years ago, you could get 300 gates on a chip; today it's 5,000. Ten years ago, CMOS was laughed at; today it is a respected technology. Further, he argued, Intel has created a microprocessor that's surrounded by a bunch of poorly designed peripheral circuits. Every customer who uses the Intel processor says he'd like to design his own peripheral circuits. Put all those things together, Corrigan said, and if you can produce a state-of-the-art CAD system, we'll have a real market.

These people were now in their 40s, which is typically rather old for starting a new company. But they felt that this was unfinished business. They wanted to prove that they were right a decade earlier. They quit their jobs and, starting with Rob Walker, Jim Koford, Bill O'Meara and Mick Bohn, they created LSI Logic Corp. in 1981.

Customers provided the netlist and the test vectors and the company automated the process. LSI bought gate-array wafers from Toshiba, so all it had to do in production was the metallization that customized the circuit. LSI's founders may not have thought about it at the time, but they had invented the fabless semiconductor business.

The major breakthrough came in disk drives. Manufacturers had graduated from 8-inch floppy disks to 5-1/4-inch disks with a 10 x 10-inch circuit board. When LSI could put that board on one chip, the ASIC business finally began its rapid ascent.

Today, the ASIC business, which nobody wanted 30 years ago, is a great business for LSI and quite a few competitors around the world. It has moved ahead significantly from the days of 300 gates on a chip to more than 20 million, allowing multiple systems on a single chip rather than a single circuit.

The Century of the Engineer: Misunderstood Milestones