Mathematically generated random number could be weak, while the randomness created physical elements often have insufficient entropy. This paper describes a XOR compiler having the potential to be a true random numbers generator (TRNG), leveraging physical unclonable functions (PUF) designed with memory products. The initial randomness is created by testing only the cells of the memory arrays that are naturally unstable under repetitive measurements, and can easily switch back and forth between a "1" to a "0", thereby generating a random data stream. The level of randomness of this data stream is then enhanced to generate the TRNG. In this paper we are presenting two complementary elements: i) how a fast XOR data compiler, while processing the data available from multiple ternary cells, can create an extremely high level of randomness; and ii) how a combinational probability model allows the quantification of the level of randomness of the TRNG. Deviations of absolute randomness of these TRNG in terms of probability to be non-random can be lower than 10-10 which is accepted as non-detectable for existing and computers of the foreseeable future.