The Santa Clara, Calif.-based company is moving into full-scale production on more than one factory using its 32-nanometer manufacturing process. Since a shift like this (away from its established 45-nanometer process) involves billions of dollars of spending and is fraught with risk, Intel is reassuring its stakeholders today — such as partners and investors — by offering an update that shows just how much progress it has made.
Whenever Intel makes such transitions, it’s easy to see why size matters in the chip industry and how much manufacturing muscle the world’s biggest chip maker really has. With each shift, Intel can put greater distance between itself and rivals, particularly those who don’t invest as much as it does. The company makes these shifts on a routine basis every couple of years. It happens so often that we’re getting spoiled by the new gadgets (see right) that these manufacturing advances make possible.
To get a sense of the scale of this shift, Intel plans to spend $7 billion to bring the 32-nanometer manufacturing process into multiple chip factories. The company estimates that it is one to two years ahead of its rivals in the shift to each new generation, said Sanjay Natarajan, 32-nanometer program manager. To keep pace, rivals have had to team up. Advanced Micro Devices licenses its manufacturing process from IBM; others such as Nvidia are dependent upon contract chip manufacturers such as Taiwan Semiconductor Manufacturing Co.
AMD plans to ship 32-nanometer chips in 2011. The cost of investments in chip manufacturing nearly drove AMD out of business. That’s why it split the company in two last year and sought outside investors for its manufacturing arm. The government of Abu Dhabi invested in the manufacturing operation, dubbed Globalfoundries, as part of a $10 billion gamble on chip manufacturing.
The shift from 45-nanometer production to 32-nanometers is the tiniest of transitions. A nanometer is a billionth of a meter and the dimensions of the manufacturing are now on an atomic scale. With this process, Intel can now put billions of transistors on a chip; back in 1971, it could fit 2,300. The improvements in moving to the new process include 22 percent better chip performance at a given power level, or, at a fixed performance level, a 10-fold reduction in chip power consumption.
“This means that the devices you use will be able to do more and have much longer battery life,” Natarajan said.
Precise control of the production process is a must. It takes about three months to process a chip from start to finish. If the production isn’t perfect, you may have three months of duds on your hands. If they work properly, then Intel can sell them for hundreds of dollars a piece. That’s why it can take months or even years of running test batches through a production line before a company will initiate full-scale production. The milestones include making test chips, fixing the recipe for making chips, launching production in a factory, and duplicating the working process in multiple factories. That’s why Intel offers repeated updates on its manufacturing transitions.
Although the transitions cost billions of dollars, Intel is sticking to its plan of introducing a new manufacturing process every two years — matching the cadence of Moore’s Law, which dictates that the capacity on a chip doubles every two years. The 45-nanometer process that it introduced two years ago has already been used to make more than 200 million microprocessors, the brains of computers. The 45-nanometer process has the highest yield, a measure of good chips out of the entire batch of chips, that Intel has ever had, Natarajan said.
Intel will disclose more details about its 32-nanometer launch at its Intel Developer Forum conference in San Francisco from Sept. 22 to Sept 24. The company’s researchers also plan to describe the innovations in its 32-nanometer process at the International Electron Devices Meeting technical conference in December.
