IEEE Solid-State Circuits Magazine - Fall 2017 - 68

"In the end, it is founders that make
companies succeed, not any of us: we are
here to support and enable."

state-of-the art chip (~US$80-100 million), combined with the lack of significant exits, made semiconductor
investments very unattractive [5].
Venture investment hit a low point in
2012-2013 [6], when fewer than five
Additionally, there are many more
Barefoot networks, a company procompanies were raising series A per
angel investors than there have ever
ducing programmable net work
year. Luckily, there has been some
been. Figure 2 shows the growth of
switches, raised US$1.35 million seed
modest growth since 2013. What has
angel investors in the popular fundbefore raising a US$24 million series
pushed the recovery?
raising platform Angel List.
A. In other cases, companies de-risk
It has not been better perforPear VC is a seed VC that is leadthe initial venture investment by
mance on old applications but rather
ing initial companies' investments
raising nondilutive capital (e.g.,
the rise of new applications. One key
across disciplines. We typically write
SBIRs [funding from the U.S. governapplication is the explosion of artifithe first institutional check. We are
ment's Small Business Innovation
cial intelligence (AI) applications [5].
focused on finding great founders
Research program]) and/or corpoSince 2012, when Alex Krizhevsky
with the ambition to build large
rate financing.
used two NVIDIA graphics processing
companies. We are not looking for
units (GPUs) to win the 2012 ImageNet
specific operational traction but
Q: Is there a future for start-ups
computer image recognition comrather an understanding of how it
that include ICs as part of their
petition, several major corporations
could be possible for building and
core technology?
and a large number of start-ups have
defining a company.
Mar: Absolutely. After a decade of
been racing to build AI applications
In the case of semiconductors,
drought, we are seeing investments
(see Figure 3 for a graph showing the
one could argue that seed financing
in the semiconductor industry from
increase in the number of start-ups
is not viable. Semiconductor compatop-tier VC firms. Of course, it is not
with "AI" in their company descripnies still require large amounts of
the panacea of the late 1990s and
tions). In 2016, Google announced
capital, so small seed financing is not
early 2000s, when early-stage fabthe first-generation tensor flow propossible. While it is true that more
less semiconductor start-ups were getcessing units (TPUs), which are cuscapital is needed to truly proof the
ting bought by public companies for
tom ASICs designed to run Google's
business, it is possible to raise seed
hundreds of millions of dollars, even
tensor flow machine-learning framefunding from the more widely availbillions. Remember Broadcom purwork more efficiently. In 2017, the
able sources and focus on de-risking
chasing MIPS Technology's core procompany announced the second
some of the investment hypothesis.
vider SiByte in a US$2 billion stock
generation of TPUs, which boast
How are companies doing this? In
deal in 2000 or purchasing RF trans15-30 times performance improvesome cases, semiconductor comceiver provider Innovent in a US$440
ments versus general purpose CPUs
panies can de-risk the investment
million stock deal, also in 2000?
[central processing units] and GPUs
early on by building an initial proUnfortunately, starting in the early to
[6]. The success of TPUs has pushed
totype on a field-programmable
mid-2000s, venture investments in
a new breed of companies to develop
gate array, plus basic software, and
semiconductors declined dramatically.
custom ASICs for specific data applishipping that to early customers.
The increased cost of developing a
cations. One example is Nervana,
Initial customer validation,
which was sold to Intel for
makes it much easier for
US$350 million in 2016
them to raise a large series
after two-and-a-half years
Investors Registered on Angel List
A that will fund their custom
in business and raising less
application-specific IC (ASIC)
than US$25 million. There
30,057
design. There are several
is another suite of recent
examples of seed rounds
start-ups specifically pow18,000
for current successful semiconering higher-performance
ductor companies: Nervana,
data centers: for example.
a compa ny developing
Innovium (founded in 2014,
2,500
custom ASICs for machineraised US$88 million) and
50
learning applications, raised
Barefoot Networks (founded
a US$600,000 seed round
in 2013, raised US$153 mil02/2010
12/2011
06/2013
11/2016
followed up by a US$3.3 million) [7].
lion series A and a larger FIGURE 2: The number of active angel investors on Angel List.
AI and data center apUS$20 million series A-II; (Sources: Angel List, Techcrunch, and Harvard Business Review.)
plications are not the only

FA L L 2 0 17

IEEE SOLID-STATE CIRCUITS MAGAZINE

Table of Contents for the Digital Edition of IEEE Solid-State Circuits Magazine - Fall 2017