Inside Intel

скачать книгу бесплатно


Into the Potato Patch (#ulink_fe38bd04-6c07-5eb0-aa70-609cbef19ac9)

SOME TIME IN THE FALL of 1968, when Intel’s research into semiconductor memory was already in full flow, one of the company’s first hires grabbed Bob Noyce in a corridor. ‘Bob,’ he said, ‘there’s something I need to ask you. I’ve been here for three days already, and I’m not really clear on the reporting structure of this outfit. Can you just draw me a quick organization chart?’

Noyce smiled, and turned into an open doorway. Walking to the blackboard, he picked up a piece of chalk and drew a small X. Around it, he swept a circle, and along the circle he added six or seven more Xs. Then he drew a spoke connecting each of the Xs outside the circle to the X in the centre.

He pointed to the X in the centre. ‘That’s you,’ he said. ‘And these’ – he tapped the outside Xs one by one, the clack of the chalk echoing against the linoleum floor – ‘these are me, Gordon, Andy, Les, Bob, Gene, and the other people you’ll be dealing with. That’s what our organization chart looks like.’

Noyce’s point was more than mere rhetoric. At Fairchild Semiconductor, the East Coast owners of the business had been very strong on hierarchies and on reporting. They believed in ‘clear lines of command’. They thought employees should be like officers in an army, communicating only with their immediate superiors and immediate inferiors. And like soldiers, they considered it the height of disloyalty for an employee to raise an issue with someone higher up if he was dissatisfied with the response he had been given by his immediate superior.

Visionaries before their time, Noyce and Moore saw that in a fast-moving industry where speed of response to change was all-important, and where information had to flow as swiftly as possible if the company was to make the right decision, this approach did not make sense. Instead, they wanted to encourage anyone who had a good idea to speak up, anyone who had a question to ask it. Staff meetings were to be open to anyone who thought they could contribute something by attending; no manager, no matter how senior, should refuse a request for help or information from another employee.

The prestige of Intel’s stars was never in doubt – the PA system kept drumming it in, day after day, as announcements went out every few minutes for Doctor Noyce, Doctor Moore, Doctor Grove. But they were very visible, mucking in where needed. If a circuit layout needed to be checked, Bob Noyce would be ready to lend a hand. If a process designer needed to know something about the behaviour of transistors under specific temperature ranges, Andy Grove would be ready to pull down from the shelf the textbook he had written on semiconductor physics, identify the key equations that predicted how a substance would behave, and help to turn the equations into statements in the FORTRAN programming language. If a piece of complex machinery didn’t work, the man who unscrewed the casing and took a look inside might well be Gordon Moore.

Inevitably, this led to some tensions within a matter of months. Gene Flath, who had been hired to run the company’s first fabrication line, began to find that engineers he had asked to go off and deal with a process problem would come back and announce that they had done something completely different. When he asked them why, the answer would always be the same: ‘Grove told me to do it’. When Flath confronted Grove – and confrontations inside Intel would often be with both parties screaming at the top of their lungs, in front of other people – Grove would deny that he’d given the errant engineers any direct orders. It took some time for Flath to convey to Grove that, as director of operations, he had more power than he realized. To the average engineer, a ‘suggestion’ from Grove was something to be acted on immediately. If Flath was expected to deliver on commitments he had made, this would have to stop. The philosophy inside the company that anyone could talk to anyone else was fine, Flath believed – but it had its drawbacks: ‘It’s very desirable, because you get a lot of good ideas. But it’s not OK to change the batting order without anybody decreeing that this is what would happen’.

The company’s first year was punctuated by intermittent stand-up screaming matches while issues such as this one were gradually sorted out. But broadly, Intel people felt a refreshing sense of freedom. Instead of having to fight the bureaucracy of a purchasing department, every engineer had the authority to sign on behalf of the company for equipment costing up to $100,000. The company parking lot had no pre-assigned spaces for senior management; instead, the rule was simply that those who arrived earliest for work got the spaces closest to the front door. And as the research continued, there was a feeling that you were part of an elite corps, an assembly of the brightest, hardest-working people, a world-beating team. Of course there was a risk of failure. But everyone knew that a talented engineer could easily find work elsewhere if things didn’t work out. There was no fear of long-term unemployment to discourage risk-taking.

‘They offered me the job at the end of breakfast,’ said one of Intel’s very first engineering hires. ‘I called my wife, and told her that I’d just accepted a job with a pay cut of one-third, working for an unknown startup. The good news, I said, was that there were some big names running it. And if it proved a mistake, I could always go and pump gas some place.’

Of the three memory technologies that Moore wanted to investigate, one swiftly ruled itself out. An early look at multichip memory modules suggested that it was too far from becoming a commercial product to be worth devoting effort to. That left the bipolar route and the MOS route. By the fall, Intel’s engineering effort was clearly organized into two terms, each led by a former Fairchild research engineer, to follow these up separately. The MOS team was led by Les Vadasz, a balding, short-tempered engineer who shared Andy Grove’s Hungarian background; the bipolar team came under the control of a brilliant but equally short-tempered engineer called Dick Bohn.

From the very first there was friendly rivalry between the two teams to see who could deliver a manufacturable product and a stable process first. The bipolar team had an early psychological boost when Phil Spiegel arrived from Honeywell, an East Coast computer company that was one of the ‘seven dwarfs’ competing against IBM to sell mainframe computers. Spiegel explained that Honeywell wanted to steal a march on its competitors by being the first computer manufacturer to build a machine that used semiconductor memory instead of magnetic core. He knew that Fairchild had come very close to developing a bipolar memory circuit with sixteen cells, or bits. Yet Fairchild had never quite managed it, because somehow one of the cells was always a dud. Insiders used to joke that Fairchild’s R&D people had built a great 15-bit memory chip.

Spiegel explained to the people at Intel that Honeywell wanted to be able to ship a new line of computers in 1969 or 1970 that would contain a 64-bit scratchpad memory, big enough to store an eight-letter English word. The company was inviting a number of firms in the electronics business to try to build some working prototypes to this testing specification. Intel was new and untried, but Noyce and Moore’s pedigrees were impressive. If Honeywell could give Intel a downpayment of $10,000 to help fund its research work, was the company interested in trying to beat the other six companies that had already started on the problem?

The offer was less crazy than it sounded. Whichever company came up with a mass-produced chip first was likely to win an order from Honeywell for 10,000 units at $100 apiece. So the up-front fee, although crucially important for a startup that had to keep an eye on its bank balance, represented a bet that only represented 1% of what Honeywell was expecting the finished chips to cost. Since it was by no means clear that anyone could build a 64-bit semiconductor memory circuit, the $10,000 was a small price to pay for adding one more talented team to the field already competing for the prize. Anyway, one of Intel’s engineers had impressed Spiegel and his colleagues as particularly committed and reassuring: H. T. Chua, a Stanford graduate of Chinese ancestry who had immigrated to the US from Singapore. Chua was a quiet, thoughtful man, but he had a air of unmistakable quiet confidence about him which seemed to exude the message: We’ll build your chip for you, and we’ll beat everyone else too.

Chua kept his promise. When Spiegel returned to California in the spring of 1969, Chua met him at the factory door with a sample chip in his hand. The chip designed for Honeywell therefore became the new company’s first commercial product. It was a symptom of Intel’s target market that the new chip wasn’t even given a name. Instead, it was referred to only by a part number, 3101. Intel’s potential customers were engineers inside computer companies, who thought of themselves as rational decision-makers choosing between one part and another strictly on technical merit, quality and price. A catchy name wouldn’t have increased sales; on the contrary, it might have excited suspicion that there was a shortage of engineering talent to cover up for. A simple part number, preferably a number that meant something, was what Intel needed to go for.

Bob Graham, Intel’s marketing chief, realized the success of the 3101 could be of enormous value to Intel. The industry was littered with companies that made grand promises which they failed to fulfil. Cynics used to joke that National Semiconductor, in particular, used to send around a specification for a new chip to customers, and then wait to see what reaction it got before deciding whether to start designing it. Graham wanted Intel to earn the opposite reputation, so he coined the slogan INTEL DELIVERS. It became almost an unbreakable rule inside Intel never to announce a product in advance, just in case something went wrong ‘twixt cup and lip. Instead, he resolved that the company would wait until chips were already on distributors’ shelves before going out to customers to spread the word about a new device.

This early triumph from the bipolar team cranked up the pressure on the competing MOS team to deliver. At one point Vadasz and his colleagues became convinced that they were almost there. The test production line they had set up yielded one device that worked perfectly. The MOS team immediately toasted its arrival with champagne in the cramped company cafeteria – but it was to be many months before they were able to build a second working MOS circuit.

Part of the trouble was that the manufacturing process itself was so rudimentary. Everyone understood that particles in the air could contaminate a semiconductor production line; the defence industry did their most sensitive assembly work inside giant sealed ‘clean rooms’, where the air was filtered to remove the tiny specks that could spoil a circuit. But Intel had no such luxury. Its fab area, recalled Andy Grove, ‘looked like Willy Wonka’s factory, with hoses and wires and contraptions chugging along … It was state-of-the-art manufacturing at the time, but by today’s standards it was unbelievably crude’.

No matter how hard they tried to clean up the fab area, the MOS engineers still couldn’t make circuits that worked. At one late-night meeting, almost in desperation, Andy Grove finally lost patience. Why was the company bothering with silicon-gate MOS technology at all? he asked. Why not go back to a similar, less tricky metal-gate technology where this problem would not arise?

It was Gordon Moore – quiet, thoughtful Gordon – who broke in on Grove’s tirade. ‘I want to see every wafer that comes off the line for the next thirty days,’ he said slowly. ‘Then we’ll make a decision on what to do.’

Over the coming days and weeks the engineers of the group brought the faulty wafers to Moore one by one. They watched him check the devices under a microscope, and test them with the makeshift equipment they had developed. Before the thirty days were up Moore told them what he thought the problem was. When a memory chip was being built, he reminded them, it had to be repeatedly heated up and cooled during different stages of the production process. The temperature change was not abnormal for the electronics industry, but this circuit was particularly sensitive. Because the design had sharp comers where the metal oxide and the silicon met, one would expand more quickly than the other, and a crack would appear that broke the circuit and rendered it useless.

Moore came up with a solution to the problem that was brilliant in its simplicity. He told the engineers to ‘dope’ the oxide with impurities so that its melting-point would fall. This would reduce the brittleness of the chip’s edges, and allow the oxide to flow evenly around the rough comers like melting ice cream. To their astonishment, the MOS team soon discovered that Moore was right. Working almost as an armchair engineer, he had solved the problem that had eluded them for months.

There was a long debate inside Intel as to whether the company should patent the ‘reflow’ process that Moore had invented. The issue was not whether the process could be patented; it clearly satisfied all the legal requirements for a patent. The bigger question was whether the patent would be self-defeating, because the information Intel would have to publish would set competitors on the right track towards similar solutions. In the end the team chose a halfway house. The process was patented in Moore’s name (and a framed copy of the patent was hung in his office as a reward); but once the chip was in production, the exact nature of the reflow process was kept secret from the hourly workers who had been hired to carry out the chip fabrication and packaging. On the long list of processes that the silicon wafer had to undergo before it could be scored and sliced up into dozens of finished memory chips, the reflow process was referred to only as ‘anneal’, a word used to mean heating glass and cooling it slowly. That way, the risk was reduced that a line worker would be offered a dollar more per hour by a competitor, and walk out of Intel’s door with the fledgling company’s most valuable trade secret.

Until there were products to sell, Bob Graham had little to do. The company did not have enough spare cash to hire salesmen to sit around and wait for the engineers to do their job. So Graham identified a candidate who could serve as his second-in-command in the sales and marketing division when the time came – and in the meantime, he amused himself fishing.

Setting his alarm clock to wake him several hours before dawn, Graham would tiptoe out of his modest house in Saratoga so as to avoid waking his wife. Then he would climb into the old Ford Mustang that he had bought from Bob Noyce, and roar up a deserted Highway 101 all the way to San Francisco, where he would park as close as possible to the Golden Gate Bridge. There, waiting under a streetlamp, he would find Gordon Moore in his overalls and work boots. With the motor engine burbling softly beneath the dark wash, the two men would ease Moore’s fishing boat out under the bridge and towards the grounds beyond the bay where salmon were plentiful.

For a scientist, Moore seemed to show scant interest in the state of repair of his craft. Sometimes he had to scrape the rust off the spark-plugs with his pocket knife. Sometimes, the boat’s rudimentary radio would cut out, leaving the fishermen cut off from the outside world. But on one occasion a more serious problem arose. The part of the expedition that required the most skill was traversing what the local fishermen called the ‘Potato Patch’, a narrow channel bordered with rocks just beyond the bridge that led to the fishing grounds beyond. One morning, just as they were halfway through the Potato Patch, Graham noticed water slopping around in the bilges of the boat.

‘Oh,’ said Moore absent-mindedly. ‘I must have forgotten to switch the bilge-pump on.’ He disappeared for a few seconds, and Graham began to hear the sound of the electric pump groaning into action. Thirty seconds later, however, the water level was still rising.

‘The pump!’ yelled Graham. ‘It’s not working!’ Frantically, he and Moore grabbed whatever receptacles were closest to hand, and began to throw bucketfuls of water overboard. Yet as fast as they bailed water out, more seemed to come in. While Graham continued to empty the buckets as fast as he could – splash, splash, splash, splash, splash – Moore went to inspect the drain fittings.

Ten minutes later the problem was solved. The hole in the boat that Moore had discovered was plugged with an old oily rag, and the two friends lay back, exhausted by their efforts. As the sun rose over the city behind them, they celebrated their survival into a new day with an early-morning beer.