I once wrote an article about how a new technology called
electron-beam (e-beam) lithography was showing great promise for semiconductor
manufacturing. I will be dating myself if I tell you when it was, but here goes
-- it was 1975, I was just out of college, and I was writing for a long-gone
publication called Northern California Electronics News.
E-beam was attractive because it directly writes patterns on
the surface of a wafer, and does not require a mask. Ironically, perhaps, it
can also be used to make photomasks. But outside of mask-making and
prototyping, e-beam lithography didn't find much adoption. That's because of
one fatal flaw - slow throughput. It could literally take hours to write a
single wafer, an unacceptable situation in a real production environment.
In recent years, however, interest in e-beam has been
resurfacing. That's because masks have gotten really expensive, and now require
optical proximity correction (OPC) software in order to print correctly. Cadence
started a mask cost reduction R&D project in 2002, and launched an
"incubation seed" project in 2004 to research e-beam direct write technology. In
2007 an independent company, D2S
(Direct to Silicon), spun out of this project, with former Cadence CTO Aki
Fujimura as the CEO.
Reducing shot count
D2S is building design for e-beam (DFEB) design kits that will
overlay existing IC implementation tools. The kits include standard cell libraries
optimized for e-beam. The basic idea is simple - reducing the number of "shots"
that the e-beam machine must take to write a pattern on the wafer. For example,
you could write the letter "E" in three horizontal shots and one vertical shot.
But if you could write the whole letter in a single shot, you'd have a theoretical
four-fold speedup.
Thus, DFEB seeks to take the most commonly occuring patterns
on the wafer, and turn them into templates. This results in the creation of
stencils through which characters are projected, hence the term "character
projection." When I interviewed Aki for SCDsource in 2007, he said
DFEB could potentially reduce shot count by an order of magnitude, making it
economically feasible for a 100 wafer production line.
New incentive at 22
nm
A recent announcement appears to be taking DFEB in a new
direction. D2S is working with other members of the eBeam Initiative, which D2S helped launch
in 2009, to develop DFEB technology for writing masks
at the 22 nm node.
I asked Aki why D2S is taking aim at mask writing. At 22 nm
and below, he explained, the ability to draw circles and curvilinear shapes on
a mask will become essential to enable OPC. That's very difficult today because
mask writing equipment draws rectangles and triangles. Using existing e-beam
mask writers, D2S DFEB technology promises to use character projection to draw
circular shapes, greatly reducing turnaround times for masks containing these
shapes.
"We believe all 22 nm
masks should be written at least partially this way," Aki said. Portions of any
22 nm mask will be dense enough to require it, he noted.
Meanwhile, Aki believes that direct-write for wafers will
become more attractive at 22 nm as well, due to "explosive" mask costs in the
$8 million range and the need for extra masks for double patterning. He's still
talking in terms of 100 wafers or less, but for 8 mm x 8 mm die, that's 100,000
chips. There are, as he noted, a number of low-volume, high-value applications
that might be able to use 22 nm technology if DFEB is available, and DFEB can
also make prototypes affordable.
Of course, an ecosystem is needed to bring e-beam technology
into production manufacturing. While e-beam machines exist today, Aki noted
that new capabilities will be needed for 22 nm. He said that silicon IP
optimized for low shot counts will also be useful. The E-Beam Initiative, which
he described as an "open collaborative forum," is working to build the
ecosystem. As of late February, it had 27 members,
including foundries, equipment makers, semiconductor companies, and EDA
providers. Cadence is one of the members.
In 1975, e-beam lithography was a good idea that was way
ahead of its time. As we head into the murky depths of 22 nm and below, it
appears that time may have finally come.
Richard Goering