Snoop-proof DSP: possible?

Please excuse me if this is a stupid question, but DSP hardware
isn't my field of expertise.

A business partner has an algorithm applicable to schocastic data,
for which it isn't possible to reverse engineer by studying the
outputs and inputs.  Is it possible to fabricate a custom DSP
chip in such a way to prevent the underlying algorithm from being
extracted, so it can remain a trade secret?

-A

axlq wrote:

> Please excuse me if this is a stupid question, but DSP hardware
> isn't my field of expertise.
> 
> A business partner has an algorithm applicable to schocastic data,
> for which it isn't possible to reverse engineer by studying the
> outputs and inputs.  Is it possible to fabricate a custom DSP
> chip in such a way to prevent the underlying algorithm from being
> extracted, so it can remain a trade secret?

It is possible to make a custom or semi-custom chip, however you need to 
order large quantities to be commertially viable. You can use DSP with 
security features, like ADI BlackFin. That protection is good enough to 
stop an individual hacker. Any chip can be reverse engineered, this is a 
matter of money and capabilities of the attacker.

Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On Sat, 26 Sep 2009 16:59:44 +0000 (UTC), axlq@spamcop.net (axlq)
wrote:

>Please excuse me if this is a stupid question, but DSP hardware
>isn't my field of expertise.
>
>A business partner has an algorithm applicable to schocastic data,
>for which it isn't possible to reverse engineer by studying the
>outputs and inputs.  Is it possible to fabricate a custom DSP
>chip in such a way to prevent the underlying algorithm from being
>extracted, so it can remain a trade secret?
>
>-A

There  is no %100 protection but there are several levels of
protection which can be put into a design to make the cost of reverse
engineering increasingly higher. Most protection mechanisms involve a
battery backed volatile storage for your design content/encryption
keys. These features are available in a number of DSP
processors/FPGAs. Solutions which have non-volatile storage of the
design are usually much easier to duplicate than otherwise.
-- 
Muzaffer Kal

DSPIA INC.
ASIC/FPGA Design Services

http://www.dspia.com

On Sat, 26 Sep 2009 16:59:44 +0000, axlq wrote:

> Please excuse me if this is a stupid question, but DSP hardware isn't my
> field of expertise.
> 
> A business partner has an algorithm applicable to schocastic data, for
> which it isn't possible to reverse engineer by studying the outputs and
> inputs.  Is it possible to fabricate a custom DSP chip in such a way to
> prevent the underlying algorithm from being extracted, so it can remain
> a trade secret?
> 
> -A

And you can put the World's Best Lock on your door -- but sooner or later 
you'll come home to find a hole cut out of your wall.

Nothing will keep a well-funded potential competitor from grinding the 
top off the chip and reverse-engineering (or just outright copying) the 
code and/or design.

But as others have said, there are chips with security features that'll 
make it harder.

All you can ever do is make it hard enough to be unprofitable; you'll 
never make it impossible.

-- 
www.wescottdesign.com

In article <E_adneyngKqVlCLXnZ2dnUVZ_qydnZ2d@web-ster.com>,
Tim Wescott  <tim@seemywebsite.com> wrote:
>On Sat, 26 Sep 2009 16:59:44 +0000, axlq wrote:
>> A business partner has an algorithm applicable to schocastic data, for
>> which it isn't possible to reverse engineer by studying the outputs and
>> inputs.  Is it possible to fabricate a custom DSP chip in such a way to
>> prevent the underlying algorithm from being extracted, so it can remain
>> a trade secret?

>Nothing will keep a well-funded potential competitor from grinding the 
>top off the chip and reverse-engineering (or just outright copying) the 
>code and/or design.
>
>But as others have said, there are chips with security features that'll 
>make it harder.
>
>All you can ever do is make it hard enough to be unprofitable; you'll 
>never make it impossible.

That's what I figured.  I appreciate all the responses.

I would have thought these days the chip circuitry might be
sufficiently 3-dimensional to make grinding off the top impractical
for copying purposes.

-A

On 9/27/2009 8:45 PM, axlq wrote:
> In article<E_adneyngKqVlCLXnZ2dnUVZ_qydnZ2d@web-ster.com>,
> Tim Wescott<tim@seemywebsite.com>  wrote:
>> On Sat, 26 Sep 2009 16:59:44 +0000, axlq wrote:
>>> A business partner has an algorithm applicable to schocastic data, for
>>> which it isn't possible to reverse engineer by studying the outputs and
>>> inputs.  Is it possible to fabricate a custom DSP chip in such a way to
>>> prevent the underlying algorithm from being extracted, so it can remain
>>> a trade secret?
>
>> Nothing will keep a well-funded potential competitor from grinding the
>> top off the chip and reverse-engineering (or just outright copying) the
>> code and/or design.
>>
>> But as others have said, there are chips with security features that'll
>> make it harder.
>>
>> All you can ever do is make it hard enough to be unprofitable; you'll
>> never make it impossible.
>
> That's what I figured.  I appreciate all the responses.
>
> I would have thought these days the chip circuitry might be
> sufficiently 3-dimensional to make grinding off the top impractical
> for copying purposes.
>
> -A

It's still expensive to de-cap a chip. You have to really want to do it 
to go that route, but it is possible.  And you're right, getting through 
all the layers is so expensive that it's rarely done, but, still, it's 
possible.


-- 
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.abineau.com

axlq <axlq@spamcop.net> wrote:
< In article <E_adneyngKqVlCLXnZ2dnUVZ_qydnZ2d@web-ster.com>,
< Tim Wescott  <tim@seemywebsite.com> wrote:
(snip)
 
<>Nothing will keep a well-funded potential competitor from grinding the 
<>top off the chip and reverse-engineering (or just outright copying) the 
<>code and/or design.
(snip)
<>All you can ever do is make it hard enough to be unprofitable; you'll 
<>never make it impossible.
 
< That's what I figured.  I appreciate all the responses.
 
< I would have thought these days the chip circuitry might be
< sufficiently 3-dimensional to make grinding off the top impractical
< for copying purposes.

That does make it harder, but I beleive that there is still
an etch photograph repeat process to slowly go through the
layers.

The usual processors used in systems such as smart cards are
produced in large quantities and for a low price each.  
The many metal layers of 64 bit processors are not needed
and not affordable.

http://www.cl.cam.ac.uk/~mgk25/tamper.pdf

-- glen

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
(big snip)
 
< It's still expensive to de-cap a chip. You have to really want to do it 
< to go that route, but it is possible.  And you're right, getting through 
< all the layers is so expensive that it's rarely done, but, still, it's 
< possible.

About 35 years ago someone showed my how to decap cerdip chips
by putting them in a vice such that one side pushes on the top
and the other on the bottom.  A little squeeze and the top
(usually) pops right off.  

Plastic is somewhat harder, but, again see:

http://www.cl.cam.ac.uk/~mgk25/tamper.pdf

where they show how to do it with inexpensive equipment.

-- glen

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>(big snip)
> 
>< It's still expensive to de-cap a chip. You have to really want to do it

>< to go that route, but it is possible.  And you're right, getting
through 
>< all the layers is so expensive that it's rarely done, but, still, it's

>< possible.
>
>About 35 years ago someone showed my how to decap cerdip chips
>by putting them in a vice such that one side pushes on the top
>and the other on the bottom.  A little squeeze and the top
>(usually) pops right off.  
>
>Plastic is somewhat harder, but, again see:
>
>http://www.cl.cam.ac.uk/~mgk25/tamper.pdf
>
>where they show how to do it with inexpensive equipment.

That paper is quite old. It seems a lot more is possible now, mostly
because people have had more years to think up devious strategies.

Beware of attacks by students. They tend to be in well equipped places for
attacking chips at low cost. :-)

Regards,
Steve

On Sep 26, 12:59�pm, a...@spamcop.net (axlq) wrote:
> Please excuse me if this is a stupid question, but DSP hardware
> isn't my field of expertise.
>
> A business partner has an algorithm applicable to schocastic data,
> for which it isn't possible to reverse engineer by studying the
> outputs and inputs. �Is it possible to fabricate a custom DSP
> chip in such a way to prevent the underlying algorithm from being
> extracted, so it can remain a trade secret?
>
> -A

You might want to follow different route: instead of selling
specialized DSP hardware (or software) you can sell a server-based
service.
This way nobody will have access to your servers so reverse-
engineering of software algorithms will not be possible.
That's what Google does for the most part
That's why they could compete against MShit while Netscape could not
Server-based vs client-based - makes all the difference
Tamper-proof, hacker-proof, trade secrets forever...