Inside Intel. Tim JacksonЧитать онлайн книгу.
a six-month visa that effectively allowed him unlimited crossings. A bigger issue was how to get the finished silicon wafers with memory circuits laid out on them across to Mexico in good time for packaging. In theory, MacKay was supposed to clear them through customs, something that could only be done through a broker every Tuesday. In practice, he carried them in a leather carpet bag on the floor of his car, immediately behind the driver’s seat. Because a two-inch wafer could carry as many as 200 circuits, depending on the chip’s die size, he could carry thousands of chips at a time. After a while MacKay began to mark the bag every time he carried a shipment, in the same way that wartime fighter pilots used to mark the fuselage of their aircraft after shooting down an enemy plane. It was not always a clear run. On one occasion MacKay was turned back at the border because his hair was too long. Undaunted, he turned the car back to the US, changed places with his colleague in the passenger seat a mile up the road, tucked his pony-tail under his baseball cap, opened his newspaper to the sport page, and lay back sleepily as he was driven into Mexico through a different customs lane.
With Intel’s head office almost a day’s drive to the north, communications were of paramount importance. Telephone service through the local Mexican phone company was out of the question, not merely because the line quality was so poor but also because of the delay. So MacKay struck a private deal with a telephone engineer working at the local phone company in San Diego, and arranged for a five-mile cable to be run across the border so the assembly plant could be connected directly into the American phone system.
Another creative engineer at Intel was John Reed, who bumped into Andy Grove at a party while still working in his first job out of graduate school. Grove asked him what he thought of Intel’s product line. Pouring himself a drink, the young Reed replied breezily that the idea of using silicon gate technology in a memory chip was ‘kinda neat’, but added that he thought the rest of the company’s circuits ‘lacked creativity’. It was an off-the-cuff, ill-considered response – but a week later Reed received a call from Les Vadasz, inviting him to come and talk to people at Intel about whether he might like to put a little more creativity into their product line. Ever the joker, Reed turned down the first proposed date on the grounds that his wife was going to give birth that day. (Her doctor had warned her that the baby would need to be delivered by Caesarean section.) But Reed agreed to an interview the next day, 5 May 1971 – and the result was that he appeared at Intel’s Mountain View facility a month later for his first day at work.
Reed soon discovered that Intel had been a great deal more creative than he realized. Ted Hoff, the young Stanford researcher who had asked Bob Noyce point-blank whether the world needed another semiconductor company, had drawn up a concept for an entirely new memory cell, which needed only three transistors compared with the conventional four, and fewer interconnections. The concept had only one drawback: it would not produce stable storage when the computer was switched off. To maintain the information in the cell, the circuit had to be ‘refreshed’ every thousandth of a second. This added a substantial overhead to the memory system as a whole – but it offered the promise of packing cells together three to four times as densely as any existing product. If Intel could build it, this ‘dynamic random-access memory’ or DRAM would store 1,024 bits, four times as much information as the highest-capacity semiconductor memory device currently available in the world.
Karp and Vadasz had been assigned to turn Hoff’s one-cell circuit diagram into a working part. They had already produced one chip design – a part called the 1102, which was sold as a custom project to Honeywell, the customer for Intel’s original bipolar chip. It was their second version, the 1103 chip, that Reed was soon told to work on. Partly because of the complexity of the chip’s design, the first prototype wafers that came off the line did not contain a single working circuit. Reed spent a number of months on the project, fine-tuning the design and the manufacturing process before finally coming up with something that he believed would be manufacturable.
When Reed arrived in the fab area late one night to see the first prototypes of his new design roll off the line, Gordon Moore was waiting for him.
‘Reed, you screwed up,’ he joked. ‘Only seventy-five working dice per wafer.’
Achieving a yield of nearly 50% was thrilling news. Noyce and Moore were totally confident that, with further improvements to the design and the process, the manufacturing cost of the 1103 chip could soon fall to a point where Intel could sell it for $10.24 – a penny a bit. Computer makers would complain that this new dynamic semiconductor memory was more complex to work with than old-fashioned core memories. But it was smaller, much faster, and used much less power. And if its price per bit was lower than that of core memory, DRAM would conquer the world.
A delegation was hastily sent over to Massachusetts to persuade Honeywell to abandon the 1102 project and to put its backing behind the 1103. Three people went: John Reed, someone from sales, and Bob Noyce himself. When the presentations began, the Honeywell engineers were highly dubious. They had been cooperating with Intel for many months on the 1102, and the part was designed precisely in order to meet the requirements of Honeywell computers. Why should they throw away this work and switch to a product that Intel would soon be offering to all comers?
Essentially, Intel had two answers. The engineering answer was that the 1103 design, for all its additional complications, was proven to be easier to manufacture. This meant not only that it was more reliable now, but also that it offered better technical improvements in the future. The marketing answer was based on cost. Because of the manufacturability advantage, the 1103 was already considerably cheaper to build than the 1102 – and the gap would only widen. But the Honeywell engineers remained unconvinced.
It was only when Noyce, the father of the integrated circuit, began to speak that they sat up in their seats. Talking without notes or preparation, Noyce repeated some of the points that his colleagues had already made. What he said was in itself not new. But Noyce’s presence, his aura of authority, his seductively deep voice, won the day. Honeywell signed up to the 1103, and the three men returned triumphant to Mountain View in the knowledge that Intel could now focus all its efforts on developing a single memory product.
The launch of the new chip, in October 1970, proved to be a turning point in the history of the computer industry. As we’ll see, it was by no means a perfect product. But by undercutting the price of core memories, the 1103 established semiconductor memory as the technology of choice for computer makers from 1970 onwards – and set the industry on a familiar path of falling costs, rising performance and diminishing size. As the price of semiconductor memory fell over the coming decades, it would become cost-effective to build memory into lots of other devices too. Demand for memory products would become a multi-billion-dollar market – a market that Intel could legitimately claim to have created single-handed.
Noyce’s success at Honeywell was typical of his talents, and proof that he was admirably qualified to be Intel’s first CEO. He combined formidable technical expertise with a will to win and a sense of fun. In the passenger seat of his Mercury Cougar on the way to San Francisco airport, Reed discovered that his boss was a terrifying driver – weaving back and forth at high speed between lanes on Highway 101 in a playful attempt to show that he could get to the airport before any other car on the road. When Noyce came back from a weekend’s skiing with a broken leg freshly set in plaster, he would challenge the first comer to a wheelchair race down the corridor, laughing uproariously as he spun the wheels of his chair faster and faster, clattering against the wall as the wheelchair veered out of control.
The other role models inside the company were Andy Grove and Les Vadasz. Both of them were as determined to win as Noyce was. But they took life, and work, far more seriously. Grove told his old Fairchild colleagues that he believed their old company had been run far too much like a holiday camp. Vadasz, meanwhile, wanted tidy desks, proper filing systems, accurate lab notes, regular performance reviews.
The attempt to impose these qualities on the company’s engineering staff soon began to lead to friction. In March 1971, three months after the 1103 was introduced in commercial quantities, Vadasz wrote a performance review for John Reed which said as much about his own personality as about the engineer whose performance he was supposed to be rating.
‘One of the clichés one uses in reviewing’, Vadasz