SAN JOSE, Calif.—Analysts differ on who has the best semiconductor process technology. Their diversity of views is understandable given the complexity of the subject and sometimes confusing messages from chip makers themselves.
Intel will be superior to TSMC and Samsung at 10nm as it was at 14nm, says Linley Gwennap of the Linley Group. TSMC’s emerging 10nm process leapfrogs Intel’s 14nm node, and TSMC is now moving at a faster pace than Intel, says G. Dan Hutcheson of VLSI Research. Handel Jones of International Business Strategies believes at 10nm it will be pretty much a tie between Intel and TSMC.
All sides agree multiple variables in how devices are made affect different kinds of chip differently. And they all lay some blame at the feet of marketing departments that often do more to obscure than clarify the situation.
“There is no single measurement that determines the performance, power consumption and transistor density of a technology,” said Jones. “M1 pitch is important but local interconnect can also have an impact on wired gate density and performance; gate pitch is important for gate density but fin height has a big impact on performance,” he said.
“Interconnect delays are becoming the big challenge with 80 percent of performance at 10nm determined by the interconnect delays,” he added.
Height and width of FinFETs is a good metric of a node’s and a chip maker’s prowess, Hutcheson agreed. SRAM cell size also is worth considering, he said.
But “I believe the ultimate measure of technical progress is the ability to double density with each node for the same design intent,” Hutcheson said. “Intel has been doing this so far with each node,” he said.
That said, TSMC has achieved metal-layer 1 pitches at 10nm that are “a full shrink (~70%) ahead of Intel’s 14nm,” Hutcheson said, noting Intel’s 14nm node is two years old now. With the recent disclosure of its 10 and 7nm plans, “TSMC has proven not only that they have their mojo back [after initially struggling with 28 and 16nm nodes], but that they also have a faster cadence than anyone,” he added.
Meanwhile, the Taiwan foundry’s current 16nm node “has blown past 28nm in the same time frame on revenues and yields” despite low expectations among some pundits for the 16/14 nm node as recently as January, he said.
Next page: A node by any other name