The process of designing a chip involves deciding where to
place tens of billions of tiny on-off switches called transistors on a piece of
silicon to create working chips. The exact placement of those transistors has a
big impact on the chip's cost, speed and power consumption.
Chip design engineers use complex design software from firms
like Synopsys and Cadence Design Systems to help them optimise the placement of
those transistors.
On Monday, Nvidia released a paper showing that it could use
a combination of artificial intelligence techniques to find better ways to
place big groups of transistors. The paper aimed to improve on a 2021 paper by
Alphabet's Google, whose findings later became the subject of controversy.
The Nvidia research took an existing effort developed by
University of Texas researchers using what is called reinforcement learning and
added a second layer of artificial intelligence on top of it to get even better
results.
Nvidia chief scientist Bill Dally said the work is important
because chip manufacturing improvements are slowing with per-transistor costs
in new generations of chip manufacturing technology now higher than previous
generations.
That goes against the famous prediction by Intel co-founder
Gordon Moore that chips would always get cheaper and faster.
"You're no longer actually getting an economy from that
scaling," Dally said. "To continue to move forward and to deliver
more value to customers, we can't get it from cheaper transistors. We have to
get it by being more clever on the design." © Reuters