It is possible to say that the success of Stochastic computing bases on
the efficient generation of random bit stream encoding real values.
Stochastic Numbers (SNs) generated must be random and
uncorrelated for the accuracy purpose. Fortunately, the exist works
introduced many approaches which can be generally devided into two main
groups: using comparator and without using comparator.
Binary
number /m c
/m
Figure 1.2: Digital to stochastic converter
For the first group, the basic idea for generating Stochastic
Number (SN) can be seen in Fig.1.2.
At each clock cycle, an m–bit random binary number denoted
y is generated by a random or a pseudo–random source and then
compared to an m–bit binary number denoted x. At the
output of the comparator, we obtain a 1 if y < x and a
0 otherwise. The received sequence is called a stochastic
sequence in which probability of observing a bit being logic 1 is equal
to x. This result can be mathematically explained as follows: A
random source generates a value y obeying a certain probability
density function f (y),
and a comparator compares the value y with a constant value
x, consequently, the output of this configuration is a sequence
of random bits that have probability of being logic 1 as