The Embedded Systems Conference could hardly have found a
more interesting keynote speaker than Dr. Michio
Kaku, a well-known author and theoretical physicist who recently wrote a
book entitled "Physics
of the Impossible." As a frequent watcher of his programs on the Science Channel, I wasn't surprised
when he talked about some amazing technology that's just a decade or two away.
I was somewhat surprised when he directly addressed Moore's Law, and
suggested that quantum effects may halt its advance by 2020. "Unless we prepare
for this," he warned, "Silicon Valley may
become a rust belt."
Looking a Decade Out
First, the good news. Setting aside (for now) the really
far-out stuff like interstellar travel, Dr. Kaku talked about some exciting
technology that's on the near horizon. For example:
But What About
But how are we going to build all this nifty stuff? The bad
news, for those of us connected to the semiconductor industry, is that it may
not be with silicon as we know it today.
Law tells us that compute power will double every 18 months as we move to lower
process nodes. By 2020, Dr. Kaku believes, chips will cost a few pennies and
will be everywhere - but Moore's
Law will be in serious trouble. One reason is that heat generation will be so
intense that "you could fry an egg on a chip." Stacking die into cubes won't
really help, he said, because heat generation increases exponentially as we go
down the process curve.
The second problem is quantum tunneling,
which refers to a particle's ability to penetrate energy barriers. When feature
sizes are merely 5 atoms across, Dr. Kaku said, "electrons leak out and you
can't predict where they are." This is because of the Heisenberg uncertainty principle,
which tells us that the location and velocity of a particle cannot be
Dr. Kaku noted that work is ongoing on a number of silicon
alternatives, including molecular computers, quantum computers, optical
computers, and graphene
sheets. But none of these are "ready for prime time." So the question remains -
what will the new technologies he describes be built from?
A Reality Check
Long before quantum effects relegate further scaling to the
"physics of the impossible," moving to the latest process node will be too
expensive for many companies. SoC development costs at 32 nm are now estimated
at $100 million, and 22 nm, with its requirement for double patterning, will be
even more expensive. Yet we need billions of microscopic chips that cost
pennies to bring about the inventions Dr. Kaku described.
How do we solve this problem? By easing hardware and
software IP integration and lowering development and unit costs, the EDA360 vision articulated by Cadence
earlier this week can help. EDA360 isn't looking at quantum or molecular computing
- yet - but it may help us extend the life of silicon, and thus support the kind of
innovation described in Dr. Kaku's fascinating keynote.