Eventually, the
conventional ways of manufacturing microprocessors, graphics chips, and other
silicon components will run out of steam. According to Intel researchers
speaking at the ISSCC conference this week, however, we still have headroom for
a few more years.
Intel plans to
present several papers this week at the International Solid-State Circuits
Conference in San Francisco, one of the key academic conferences for papers on
chip design. Intel senior fellow Mark Bohr will also appear on a panel Monday
night to discuss the challenges of moving from today's 14nm chips to the 10nm
manufacturing node and beyond.
In a conference
call with reporters, Bohr said that Intel believes that the current pace of
semiconductor technology can continue beyond 10nm technology (which we would
expect in 2016) or so, and that 7nm manufacturing (in 2018) can be done without
moving to expensive, esoteric manufacturing methods like extreme ultraviolet
lasers.
Why this
matters: The discussion
is anything but academic. This year marks the 50th anniversary of Moore’s Law,
Intel founder Gordon Moore’s axiom that transistor density doubles about every
eighteen months. In the real world, that’s meant that the silicon chips that
power PCs, phones, servers and more can run faster and consume less power as
they move from generation to generation every two years or so.
The process to
make silicon chips is complex—an Intel primer
on the subject details some of the steps—but the gating factor is light itself.
Chips are etched out of silicon using light, and chip makers have to wrestle
with the wavelengths of light itself to continue to eke out new improvements.
If the industry collectively fails to do so—or fails to do so
cost-effectively—chip improvements will halt.
Intel's diagram
marks the improvements in cost and transistor size over the last few process
generations.
Intel: 14nm
"Broadwell" Technology Is Back On Track
Intel is on the
cutting edge of silicon manufacturing, however, and Bohr’s role as its senior
fellow of logic technology development carries weight. Intel will present five
papers at ISSCC, three of them covering the current 14nm technology. It will
also participate in the 10nm panel, where Bohr said he expects “spirited debate
and discussion” on what the industry needs to do to get there.
Intel was
already forced to delay its 14nm “Broadwell” chips by several months due to
manufacturing issues, and hopes to avoid that during the 10nm generation.
“I think we may
have underestimated the learning rate—when you have a technology that adds many
more masks, as 14[nm] did...it takes longer to execute experiments in the fab
and get information turned, as we describe it," Bohr said, when asked what
went wrong. "That slowed us down more than we expected and thus it took
longer to fix the yields. But we’re into high yields now, and in production on
more than one product, with many more to come later this year.”
Bohr said that
Intel’s pilot 10nm manufacturing line is running 50 percent faster than the
14nm line in terms of major steps per day, which will keep Intel’s 10nm
development on track.
That's good news for the majority of the PC market, which are powered by
Intel's chips. But if the chip industry as a whole can make out a few more years
without radical changes in its manufacturing technology, that's even better.
