Re: Computer CPU chips with built-in crypto?

From: Martin Minow <minow@apple.com>
> I'm not sure if I can answer this but, at last week's SF cypherpunks
> meeting, an Intel engineer asked whether there might be any interest
> in a computer chip with some sort of encryption mechanism built
> into the chip. As I understand it, this chip would process an
> encrypted instruction stream. I.e., it could not execute a program
> unless the "key" for that program was first loaded into the chip.
>
> An interesting idea: does anyone have more information?

This sounds like something which might be used in a set-top-box or
"information appliance" application where pay-per-use programs would
be loaded from a CDROM or network connection.

People have been dreaming about pay per use software for many years.
It is a similar idea to the "mini application" concept which would
replace the monolithic super-apps, the Microsoft Words and the giant
do-everything web-browsers/newsreaders/mail-clients, with small, single
function utilities.  This is part of the idea behind Apple's OpenDoc
and Microsoft's OLE.  In the same way, instead of buying a big program
for hundreds of dollars, you'd just download and use the functionality
you needed for a small fee.

Yet in practice it is not clear whether either of these trends will have
any market success.  Monolithic applications seem to be doing very well,
with more integration being the trend, not less.  And the whole idea
of introducing metering to a market which is used to paying just once
for access is one which is bound to meet resistance.  Look at AOL which
is going to single-charge unlimited access to the net.

So in both cases the trend looks to be going in the opposite direction.

Another possible application for the built in encryption is software
piracy protection.  You'd unlock software for your CPU but it would
not run on anybody else's without a different key code.  Here again
there is not much benefit to the end user, unless software prices come
down dramatically when this device is used.  But otherwise the computer
manufacturers are selling computers which have features which will limit
the powers of the buyers, and having to sell them more expensively to
boot because of the special chip.  In these days of razor thin profit
margins in the PC business it is hard to see how this will sell.

Hal

Hal Finney writes:

>Another possible application for the built in encryption is software
>piracy protection.  ...  In these days of razor thin profit
>margins in the PC business it is hard to see how this will sell.
>

Let me offer a possible scenario: we're entering an era when
there is much more chip real-estate than "consumer-grade" PC's
can use. Adding an encryption engine to the instruction stream
could be as simple as adding a series of barrel shifters between
(or inside) the processor cache and the instruction decoder.
(Imagine blowfish or DES -- or something as simple as RC4).

If encryption is turned off, the chip would be bug-for-bug
compatible with the existing PC. The vendor would put the
chip into the ordinary production cycle and, in two to
three years, it would be on the target audience's desktops.
(Remember, it would run existing and new, non-encrypted,
software without change.)

Encryption would be turned on on a module-by-module basis by
operating system "loader" code that would detect a "key required"
cookie in the executable file (or the Open Doc file, or
the Java class file). Before starting the module, the
o.s. loader would lookup the cookie and load the decryption
key into the chip. The customer would purchase a key by
giving a magic number from the software and a magic number
(processor serial number) to the vendor. This could be
done automatically over the network.

Now, a software vendor could provide the latest software for
free from a public FTP site, and could offer a variety of
decryption keys (30 day free trial, one-time-use micropayment,
etc.) at a variety of prices. This could also be integrated
into multiple site-license managers such as KeyServer.
As with KeyServer, key management could be done "invisibly"
over the Internet.

Note that the chip does not offer end users any encryption
or decryption capabilities -- the decrypted instruction
stream cannot be directly examined by end users. On the
other hand, if the encryption key generator was available
to "anybody," it would be trivial to construct secret
messages by generating programs that, when run, constructed
the desired message. For that reason, I suspect that
keys will be limited to a length that "national interests"
are comfortable with.

Martin Minow
minow@apple.com