a warehouse, useless because of the absence of a single small component worth only a couple of bucks. It was in order to avoid this nightmare that computer companies preferred to buy parts where there were at least two makers in business. For an electronics company with its own technology, a second source was a necessary evil: it took the cream off margins, but it was the only way to get customers to buy the part.
There were two ways to become a second source. The formal way was to sign a licensing agreement with the company that invented the product, paying an up-front fee and a royalty. In return, you got a set of the masks containing the master layouts of the circuits. Sometimes, if you were lucky, some engineering visits to help you to get the line working properly so that the process would deliver reasonable yields.
The informal way to do it was to buy a few sample parts from a distributor as soon as the part came out, rush them hack to the lab, pop the top and take blowups of the circuits inside. You then assigned a team of people to perform traces on the part, carrying out a kind of electrical audit to try to deduce from the signals going in and out what was happening inside. This was slower and more difficult than official second-sourcing, but cheaper. And although it was frowned on, the law of trademarks and patents, secrets and copyrights was murky enough to make it possible without actually breaking the law.
Second-sourcing, then, was the strategy that Sanders proposed to use to bootstrap himself into the semiconductor business. For this purpose, the key technical member of his team was Sven Simonsen, a Danish engineer who had come to the United States after working for Westinghouse Electric in Copenhagen. Simonsen had been a circuit designer at Fairchild, where he had conceived ten different logic devices, all in medium-scale integration, and had then as head of department presided over the development of fifteen more. By 1970 the devices were no longer state-of-the-art; they contained far fewer components than the 4,000-transistor monsters that Intel was working on. But they were good, solid sellers, and Simonsen knew the product range so well that he did not need to take so much as a scrap of paper from Fairchild with him when he left to join AMD.
The great advantage of trying to redesign a chip that you had already designed once before was that it gave you a chance to correct all the mistakes you had made the first time round, without having the manufacturing guys on your back screaming about the production that would be lost every time you stopped the line to make a change. But it was risky to improve things too much. The object of the exercise was to produce a part that was ‘pin compatible’ with the market leader – meaning that it could be slotted into the circuit board using the same number and placing of metal connectors, or ‘pins’, as the other company’s part. If the redesigned circuit used only half as much power and ran three times as fast, the computer company that was buying it would have to make design changes before it could start using your part. Far better just to aim for a 15% speed improvement, and make it easier for the customer to switch.
There was just one problem. If the competition found out that its part was being second-sourced without authorization, it could always fall back on an old-fashioned strong-arm tactic to squeeze out the interloper. ‘OK,’ it could say to its biggest customers. ‘You’ve received a bid from the guys down the street on three of our parts that undercuts us by 20%. But you’re still buying six other parts from us where there’s no second source. You’re gonna have to choose: do you want those six parts or not? Because there’s no way we’ll sell them to you if you’re buying the other three from the competition.’ Alternatively, it could simply refuse to sell the different parts individually.
This form of defence was called ‘packaging out’, and it raised the stakes dramatically for a company that was thinking of becoming a second source. Copying just a handful of products from the range wasn’t enough. To get into the chip business seriously, you might need to clone ten different parts or more in order to protect your customers against the risk of retaliation.
It was a measure of the energy of the AMD engineers that even with the upheaval of changing jobs, moving into the company’s site in Sunnyvale, and scratching their heads to think back over a number of past projects, they still managed to bring no fewer than sixteen finished circuits to market within nine months of starting work.
At first, AMD’s parts were hard to sell. Purchasing managers loved to cut their costs, but they knew that switching a contract to an untried new company that failed to deliver might cost them their jobs. So it would take more than just a few nickels off the price before they would commit themselves. It would also take credibility and good, old-fashioned salesmanship. After several years of sitting in Fairchild’s LA office telling a fleet of salesmen what to do, Ed Turney suddenly found himself pounding the pavements for AMD, visiting clients. Because credibility was crucial, he would often take Simonsen or another engineer with him. In particularly difficult cases even the master salesman himself -Jerry Sanders – would occasionally be drafted in to help.
Just when pessimism was beginning to set in, the fledgling company received an unexpected bonus. The new management at Fairchild Semiconductor sent a memo around to its salesmen, attacking AMD and demolishing its product line point by point. When a friendly client sent a copy to Turney, he immediately realized its value: the memo proved that Fairchild took the threat from AMD seriously. Since credibility was the company’s greatest problem, this was a positive rather than a negative thing. Turney promptly ran off a few hundred copies of the memo, and circulated them to all the purchasing managers who were still humming and hawing about whether they were going to buy from AMD.
Once the customers began to look at AMD’s product line more closely, they saw another attraction. Traditionally, the electronics market had been split into two: there were the standard components for the computer industry, which would work in a temperature range between zero and 75 degrees Celsius; and there were the parts sold to the military, which were guaranteed to work anywhere in a much wider range from –55 to +125. The two sets of components were often run off exactly the same production line; the only difference was that the parts which still worked during the extreme temperature tests would get sold to the military for three or four times the normal price so they could be frozen or boiled up in the sky or under the sea.
Sanders saw an opportunity in this. Because AMD was making design changes to the parts it was second sourcing, and was fabricating them on a new production line with clean equipment, the yields were often higher than in the Fairchild plant where the original part was still being made. As a result, AMD found that an unexpectedly high proportion of its parts met the more rigorous military specification. Most civilian customers had no practical need for ‘Mil Spec’ parts, since they weren’t building fighter planes or missiles. But the military standard of quality sounded good, and helped the parts to sell.
Shortly after AMD’s first advertisements appeared in the trade press, Simonsen took a call from an executive at Westinghouse Electric, the company he had worked for before joining Fairchild.
‘Sven, you’ve been advertising a four-bit adder,’ said the voice. ‘Do you really make the part, or is it just a marketing ploy?’
The question was a reasonable one; plenty of companies in the industry saw nothing wrong with advertising parts that weren’t yet in production. But Simonsen protested that AMD was innocent. Not only was the company making the part, a small logic chip that could add together whole numbers between 0 and 15. AMD could even provide 1,000 units the next day if his former colleague wanted them, he boasted.
‘Not quite,’ he was told. The Westinghouse man explained that he was in charge of a project to develop a radar for the US air force. The air force’s standard practice was to ‘fly before buy’; it would happily issue development contracts, but to get a real production order, a company would have to demonstrate an aircraft that actually flew – or in this case a radar device that worked. Westinghouse’s problem was that it was ninety days away from a crucial test of its new radar system, and it had just been told by Texas Instruments, the most powerful company in the chip industry, that a four-bit adder crucial
