28 May 2013
TSMC opens up about Altera's move to Intel
Earlier in 2013, the news broke that Altera was to use Intel to build fpgas at the 14nm node after dealing exclusively with Taiwanese foundry TSMC for 20 years.
The move shocked TSMC. The company's chief executive Dr Morris Chang noted during an analysts' call with New Electronics: "I very much regret Altera's decision to work on 14nm with Intel."
Elizabeth Sun, pictured, director of TSMC's corporate communications division, explained his comment. "It was a big deal; not in terms of the size of the business, but in terms of emotion. Altera has been a customer for 20 years and has always single sourced at TSMC. Its move to Intel for 14nm – even though the rest of its products remain at TSMC – is hard for us."
In his analyst call, Dr Chang noted the move had forced an internal investigation. While Sun wouldn't say what lessons had been learned, she inferred that Altera had made its call on the basis of one parameter – transistor performance.
How, then, can TSMC compete in the future at the leading edge? "Semiconductor technology has many dimensions," she claimed, "and transistor performance is only one – there's also interconnect, density, the overall architecture, power efficiency, cost, yield, reliability, infrastructure support and design enablement. TSMC is ahead in every category, with the exception of transistor performance – and that's because Intel has optimised its 14nm transistor performance for cpus.
"It's TSMC's job to offer standard technology and capacity. Our technology should be the one all customers can use and rely upon, so they can compete in their markets. We have that ability and will continue to offer it."
Altera's move comes at a time of significant technological change at TSMC – not only is it continuing to move down the process node curve, it's also looking at manufacturing on 450mm wafers and considering the use of extreme ultraviolet (euv) lithography.
Addressing 450mm manufacturing, Sun said the industry is moving forward in terms of developing tools. "The level of consensus is becoming higher and the prototype process has demonstrated uniformity comparable to 300mm tools."
TSMC now thinks 450mm could be in pilot production in 2016 or 2017. "Additional efforts will be needed with vendors," she said, "but the tools are capable of mass production. However, we will have to look at productivity and our financial targets."
One big problem, when it comes to 450mm productivity, is the euv light source. For euv to be viable, say industry experts, the light source needs to generate 100W – so far, nothing like that has been achieved. "Development has been slow," Sun reflected. "But, in the last two months, ASML has demonstrated an exciting breakthrough, particularly in terms of source power. At 40W, it is now at the level where we can begin to think about euv in a more realistic way – in the past, we never knew when we might rely on euv to do anything. We are still quite a distance from the real productivity we want, but there's now hope."
Despite the view that semiconductor manufacture beyond the 10nm node will be more than a challenge without euv, Sun said it will be possible. "It's not an issue whether we can go beyond 10nm without euv, but it's about the economics. TSMC has been working on extending existing lithography technology. "We've been preparing for 10nm without euv by using multiple patterning," Sun noted. "But we need innovation and if euv doesn't come through, we have to work out other ways of solving the problems. However, while euv doesn't determine the future beyond 10nm, the economics will be better."