A 128 EPROM with encrypted read access

Abstract

A NONVOLATILE MEMORY which combines concepts from statistics and information theory with digital MOS circuits of the 128K EPROM t o provide a memory that can decide whether or not to allow access to its instructions, will be reported. The purpose of this development is to present a barrier to software pirates. This paper concentrates on the circuit design of the Pseudo-random Number Generator (PNG), the analysis of the randomness, and a theory of why a digital N-channel MOS device can produce unpredictable numbers.As an example, we can consider the case where two EPROMs are locked and their data cannot be read. A handshake between the two EPROMs is used to authenticate and unlock each part. One EPROM, the ORIGINATOR, sends a pseudo-random number to the other EPROM, the RECIPIENT. Both parts encrypt the number using a substitution-permutation algorithm that utilizes a 64b key stored in each part'. The encrypted number from the RECIPIENT is sent to the ORIGINATOR where both numbers are compared. If the encrypted numbers are equal, the ORIGINATOR can be unlocked to operate as a standard EPROM. For the two-way handshake, the two EPROMs can be made to swap roles to authorize read access t o the RECIPIENT. The PNG is used to make each handshake different. If the same number were encrypted each time, pirates could distribute ROMs that would reply correctly during the handshake.The following tests were performed on a sample of over 900,000 numbers to evaluate the randomness of the PNG': ( I ) A search was performed to determine the most corn mon pseudo-random number generated. The hexadecimal number 00000000 appeared 18 times. Eleven numbers appeared between 3 and 11 times and 117 numbers appeared twice, resulting in over 99.97%of the numbers generated appearing only once. The low percentage of repeat numbers generated hinders attacks by pirates monitoring the PNG output for multiple occurrences of the same pseudo-random number.( 2 ) The Collision Value was determined by counting the number of repeat occurrences of the same 32 bit pseudo-random number'. The ideal value was 40 to 70 for the sample size used, and the result was 110. -The value of 110, though not perfectly random, is sufficient for the intended operation. The correlation of two lists offset by one number was tested. The acceptable values are 0.0 to 0.2 and the calculated result was 0.00238, close to the best that can be expected3. The percentage of bits that are 1 bits was found to be 50.15%, also close to the ideal 50% value. A test was performed to find the bit following each 1 bit. It was found that 46y~ of the bits following a 1 bit were 1 bits. The close correspondence of this result to the ideal value of 50% indicates that there is little correlation between two consecutively generated bits. PNG performed excellentlv on all of the tests. illustrating the high degree of randomness present. Figure 1 is a schematic of the PNG. The design uses two high-frequency ring oscillators that are sampled by a slower ring oscillator. Randomness caused by frequency ...