Cypherpunk Chip

[MIKEINGLE at delphi.com]

I'd like to propose a Cypherpunk chip to take the place of the
Clipper chip. This could go a long way toward bringing the
Cypherpunk vision to life, and it could also make someone a
fortune. But first, I'm going to flame a bit.
 
Even without key escrow and secret algorithms, Clipper is no good.
The Clipper chip uses a conventional single-key algorithm, so if
you want to use it with public-key, you have to do the RSA
operations in software. This makes it vulnerable to tampering and
key stealing. Clipper is essentially a beefed-up DES with a built-
in spy hole and a classified "trust us" algorithm.
 
The NSA seems to believe that only their classified algorithms are
unbreakable. This is not true. We don't need to trust the NSA to
give us an unbreakable single-key cipher. What we need are two
good, respected ciphers and a simple reorganization operation.
 
Choose, for instance, IDEA and triple-DES. Both are good
algorithms. They have keys which are long enough to rule out any
possibility of a brute-force attack. They are resistant to known
methods of cryptanalysis, including differential cryptanalysis.
They have good dispersion and produce pseudo-random ciphertext.
Whether the NSA could break either of these, using a classified
method, is pure speculation.
 
IDEA and triple-DES are very different algorithms. DES is based on
bit manipulation and permutation tables, whereas IDEA uses 16-bit
arithmetic operations. This is good, because it means that if
either algorithm has a flaw, the other one is not likely to have
the same flaw.
 
Take a 64-byte section of plaintext. Encrypt it using triple-DES.
Now, reorganize the ciphertext: take the first byte of each 8-byte
DES block (bytes 0,8,16, and so on) and make the first 8 bytes of
the reorganized section. Repeat with the second, third, etc. until
the entire 64 bytes are reorganized. So:
 
0       0       0       0       0       0       0       0       
ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789!@
becomes:
AIQYgow4BJRZhpx5...and so on.
 
Then, re-encrypt the section with IDEA, producing double-encrypted
ciphertext.
 
Suppose you now change one bit in the ciphertext and decrypt it.
When you decrypt IDEA, you get one 8-byte block of garbage (which
is indistinguishable from the rest of the still-encrypted data).
When you undo the reorganization, you get one bad byte in each DES
block. So when you DES decrypt, you get 64 bytes of garbage.
 
This compound cipher is effectively a 64-byte block cipher, in that
every bit in the 64 bytes depends on every other bit. This means
that you would have to attack a 64-byte section i.e. you would need
64 bytes of plaintext to attempt a plaintext attack. And even if
you had the 64 bytes, you probably couldn't do anything with it. If
you found a weakness in either cipher that allowed you to attack
it, the same weakness would not exist in the other cipher (since
they are based on different methods). So you could not attack the
compound cipher. For this reason, a combination of two ciphers,
especially with the reorganization, should be much more secure than
one cipher. If I had to choose a cryptosystem to bet my life on,
I'd choose this above Skipjack any day.
 
Of course, it would be slower than a single cipher, but there are
ways to make it usable. For example, the reorganization would be
performed by straight-line assembly language, with no loops. And
the encryption program would use a large disk buffer. Or, better
yet, use hardware. See below.
 
The Cypherpunk Chip
 
Why couldn't we make a chip of our own, with some venture capital?
This would be a public-key encryption chip, with all the necessary
hardware self-contained, which would make secure phones, faxes,
computers, and everything very easy. RSA Data Security could make
a fortune if they received only a small royalty on each one sold.
The design of the chip would be extremely public and readily
available. Any company could produce them if it was willing to pay
for the use of the RSA algorithm.
 
The chip would contain a hardware true-random-number generator;
facilities for executing RSA; MD5 or another message digest; and an
extremely secure conventional cipher such as the compound cipher
mentioned above. It would also contain nonvolatile (flash ROM)
registers to store its public key and secret key.
 
Hardware random numbers can be generated by several methods. One is
to measure the jitter in an unstable oscillator. Another is to
reverse bias a diode, right on the edge of zener breakdown, causing
it to produce white noise. These methods can be proven by quantum
mechanics to be inherently random.
 
The chip would be sold blank - no serial number or anything. The
user would instruct it to initialize itself by generating a key
pair. It would do this internally, producing and remembering a
public key and secret key. This might take a while, but you only
have to do it once. When this is finished, you can easily extract
the public key, but there would be no way to extract the secret key
from the chip. The chip would be designed to make it very difficult
and expensive to extract the secret key by physical surgery, thus
making key stealing hard. The secret key would be stored internally
in encrypted form, using a pass phrase much like PGP does, and the
chip would decrypt it before each operation.
 
In addition to the initialization, there would be five basic
operations. The chip could output its public key, encrypt, generate
signatures, decrypt, and check signatures.
To encrypt, you would send the chip one or more public keys for
people you want to send to. It would generate a random session key,
and output the session key RSA-encrypted with each of the provided
public keys. It would never reveal the actual session key. Then the
chip would accept plaintext, 64 bytes at a time, and output
ciphertext. When the encryption of a particular message was
finished, the chip would forget the session key.
 
To sign, you would send the text to the chip, along with the pass
phrase, and it would run the MD5 algorithm on it. When finished, it
would output the signature, the MD5 encrypted with its secret key.
 
To check a signature, you would send the text, signature, and
public key, and the chip would output good or bad.
 
To decrypt, you would send the ciphertext and pass phrase, and the
chip would output the plaintext.
 
This chip would do basically what PGP does, except that it would be
self-contained, very difficult to steal keys from, and easy to use
in any device. The chip would not need to be hard-wired into a
device. It could be built into a card or other plug-in module,
perhaps PCMCIA compatible. You could use this as an electronic
identity, while retaining the option to remain anonymous by getting
a second card and generating a new identity. The card might also
contain a memory for other people's public keys.
 
You could, for example, insert the card into a pay phone and dial
your pass phrase. This would secure the call, allow you and the
recipient of the call to verify each others' identities, and pay
for the call with digital cash.
 
If this chip existed, "crypto-anarchy" would be easy. Everyone
would have a motive to use it. There would be no more credit card
fraud, no more phone-code fraud, no more bad checks, no more
hacking, no more surveillance, etc. The chip could make it happen.
We cipherpunks could actually win.
 
The chip would be introduced in a low-key way. We would let the
market see its advantages and jump on it, before the bad guys
recognized the threat. For example, it could be introduced as an
option for computers - plug it into a PCMCIA card slot and use it
to secure E-mail, your hard drive, etc. Don't advertise what it
could become; just let it happen. When we get it started, it will
happen by itself.
 
                                     < mikeingle at delphi.com >