[Freedombox-discuss] [wolfgang at sharism.cc: How to Blow $100 Million]

Sat Sep 4 15:30:16 UTC 2010

----- Forwarded message from Wolfgang Spraul <wolfgang at sharism.cc> -----

Date: Sat, 4 Sep 2010 05:34:17 +0000
From: Wolfgang Spraul <wolfgang at sharism.cc>
To: discussion at lists.en.qi-hardware.com
Subject: How to Blow $100 Million
User-Agent: Mutt/1.5.20 (2009-06-14)

Hi,
read a nice editorial about the challenges of IC design and
manufacturing, something we are gearing up to with Milkymist

http://www.mdronline.com/editorial/edit24_34.html

For those who don't know - Milkymist [1] is a new, free, GPL licensed
SoC architecture. It is built around an openly licensed [2] Mico32
processing core, which gives us a headstart in the instruction set
design and toolchain support. The Mico32 part makes up about 25%
of the entire Milkymist SoC design (just counting lines of code [3]),
so I think it's fair to call the entire work a 'Milkymist SoC
architecture'. It would be great if Mico32 would see more development
in the openly licensed part (there are some closed parts too AFAIK),
but right now it seems rather static (it was developed in 2006), so
the Milkymist part will probably grow in percentage overall, especially
if someone can start working on the ambitious roadmap items
http://www.milkymist.org/wiki/index.php?title=Roadmap

The first hardware product to be tailor-made around Milkymist is
the Milkymist One VJ station [4]. Others are in the pipeline, such as
the Xué digital camera [5], and Werner is even thinking about a Milkymist
NanoNote already...

Quoting from the editorial
"
... Our rule of thumb is that a leading-edge fabless chip company
requires about $100 million to survive ... Startups today, unlike
those of yesteryear, are taking on massive system-on-a-chip development
projects. These aren't "simple" microprocessors, DSPs, video
controllers, interface chips, or other easily understood
single-function chips; they're complex and heavily integrated,
with hundreds of millions of transistors fabricated in 40nm or
better technology.
...
One or two spins of the silicon should be factored in, including
tapeout fees, manufacturing charges, testing, and more. Nothing
this complex is designed from scratch, so large chunks of the
design will need to be licensed from third-party IP vendors.
...
We figure two years from first serious design effort to working
silicon is typical. That's two years of toil, expense, risk, and
revision before the product sees the light of day.
...
But that's only half the expense. The other half comes in the waiting.
...
"

I found the editorial very inspiring. Gives you the famous 'Bring
them on' feeling to ignore reality and just be happy about yourself :-)

We don't have 100 million USD, and while I wouldn't mind talking with
someone to invest, I don't think it will happen and I don't actively
spend time on it. We work with a few 10k USD instead, our savings.
Attempts at ASIC manufacturing are postponed until we get everything
rolling in FPGA land. And maybe the first steps into ASIC will be the
'simple' microprocessors the article is talking about, for example
if we find something where we can replace an expensive or interesting
proprietary IC that implements some isolated function.
Our ICs, even Milkymist, also will not have 'hundreds of millions of
transistors', and while the Spartan-6 FPGA is manufactured in a 45nm
process, and the upcoming Spartan-7 (give it at least another 2 years)
will be manufactured in a 28nm process, our own steps into ASIC will
more likely start with 500nm or larger processes.

In general we cannot approach Milkymist in the same aggressive way
someone with 100 million USD on hand would, but the path is the same,
only we walk slower. The key will be to innovate on freedoms,
not to fall for proprietary shortcuts, not to license 'large chunks
of the design from third-party IP vendors'.

Roll Over ARM, and tell Intel, MIPS and all the others the news. :-)
Wolfgang

[1] http://www.milkymist.org
[2] http://github.com/lekernel/milkymist/blob/master/LICENSE.LATTICE
[3] http://github.com/lekernel/milkymist/
[4] http://en.qi-hardware.com/wiki/Milkymist_One
[5] http://en.qi-hardware.com/wiki/Xue

_______________________________________________
Qi Hardware Discussion List
Mail to list (members only): discussion at lists.en.qi-hardware.com
Subscribe or Unsubscribe: http://lists.en.qi-hardware.com/mailman/listinfo/discussion

----- End forwarded message -----
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.alioth.debian.org/pipermail/freedombox-discuss/attachments/20100904/fcfe851f/attachment.pgp>