Our second experiment (also shown in Table 1) shows the results of an area-oriented comparison of MINIMALIST and 3D.
The cost metric that best approximates area for technology-independent two-level logic is total literal count; hence, total literal count is used in this comparison.
Based on the above observation, we now indicate the settings of the various modes of MINIMALIST for this experiment. The vast majority of the MINIMALIST results in this set of runs use the fed-back output machine implementation style. Again, the table identifies the particular style chosen for each design. Throughout, MINIMALIST is directed to use the multi-output logic implementation style, and the literal count cost function, which best minimizes total literal count. Finally, the encoding step uses fixed-length constraint satisfaction mode.
These runs were obtained using a script identical to that of Figure 4, but using the multi-output logic implementation style. In particular, the single-output ('-s') flag was removed from the state encoding and logic minimization steps. Again, the cost function used was total literal count.
As shown in Table 1, MINIMALIST's term-sharing
across outputs and next-state provides for significant reductions in total area.
MINIMALIST's results for the area-targetted run show an average reduction
of 
in total literal count over 3D, the best being 
for
sc-control. For all designs, MINIMALIST achieved strictly
better results than 3D. Although these runs did not target product count directly,
they offer similarly dramatic reductions by that metric as well. An average
of 
improvement is observed, the best being 
for sc-control.
Again, MINIMALIST's results are strictly better than 3D in every case.
Unlike the performance-targetted runs, the code length used by MINIMALIST
rarely exceeded that of 3D (only 3 times out of 23 designs), and never by more
than 1 bit. In fact, MINIMALIST uses slightly fewer total
state bits over the entire benchmark suite than does 3D, by roughly 
.