Even Intel can’t move fast enough
1 min read
The ideal lifetime for a process technology is just 18months, according to Anand Chandrasekher, general manager of Intel's ultra mobility group.
Providing the keynote address to the National Microelectronics Institute's annual dinner, Chandrasekher said: "But we can't do this because of human limitations. Even making something happen on a two yearly basis takes 10 years of investment to get the pipeline in place."
Currently, Intel moves to the next technology node every two years. "That's optimal, given that the engineering investment is small relative to the capital investment," he noted. "But every six quarters would be best."
Leaving longer gaps is not a good idea either, he continued. "If we updated every 2.5years, it would cost Intel an extra $10billion. If we waited for three years, that cost rises to $30bn."
Explaining the technology pipeline, he told the audience of UK microelectronics executives that research into high K metal gate – which has just entered production at Intel – started 10 years ago. "It was a small investment, but it carried high risk. At the same time, we started investing in strained silicon." Each area of research comes to fruition at different times. "We need a range of technologies," he claimed, "so we can time their insertion into manufacturing."
Chandrasekher also explained the financial benefits of die shrinks. "Producing a chip with a die size of 120mm2 requires an investment of $19bn in manufacturing and tooling," he noted. "If we didn't invest in process technology that allowed scaling, we would need to spend $39bn over 10 years. That $20bn difference is worth $10bn at net present value."
That holds if the transistor count remains the same. If die size is maintained, the figures become more interesting. "If you don't invest in the latest process technology," Chandrasekher highlighted, "then the cost over 10 years is $264bn, or $85bn at net present value. Clearly, investment in process technology makes sense because the payback is real."
Asking how long Moore's Law might continue, Chandrasekher concluded: "There are challenges finding applications that extend Moore's Law, but we have been delaying the inevitable."
