Approximate Reciprocal-based Divider

 The paper can be downloaded from the following link: https://eclectx.org/Publications/C70.pdf

First, it summarizes the state of the art in approximate computing with a focus on the division. It is easily understood without prior knowledge of the state of the art. 

As for the proposed algorithm, the designed divisor is only 8/4 bits, and I am not convinced it is possible to easily extend it to a larger number of bits. The idea is that the transform the division into multiplication by an approximated reciprocal value. More specifically, the approximate 1/B by the K/2^n. With this approximation, the division becomes the multiplication by K followed by the shift by n.

The issue is the selection of the number K. In the paper, the authors divide only by 4-bit numbers, while the K itself is an 8-bit number. As such, the K is represented by 8 independent logical functions. Each of the functions has 4 inputs. 

For using this approach with larger numbers, the K generation needs to be revisited.