After state minimization and encoding, MINIMALIST performs two-level hazard-free logic minimization. This step is normally performed using HFMIN, the first exact multi-output symbolic hazard-free two-level logic minimization tool.
HFMIN, as reported in [8], uses ESPRESSO to generate ordinary prime implicants, then refines them as needed to dynamic hazard-free (DHF) primes [20], and finally, performs a unate covering step using MINCOV [24], covering required cubes[20] in lieu of minterms. HFMIN's use of such highly-optimized algorithms for sub-steps allows it to readily handle most minimization problems we have encountered.
HFMIN has also been enhanced for MINIMALIST with the ability to produce output-disjoint5 and single-output covers, and the ability to minimize literal count. Output-disjoint and single-output covers are formed by generating a suitable set of prime implicants before DHF refinement takes place. The rest of the algorithm proceeds unchanged.
To minimize literal count, HFMIN performs a weighted unate covering step where prime implicants are assigned weights according to their literal count. In addition, HFMIN now supports a limited post-processing step that further reduces literal count. This step is similar in spirit to the make-sparse operation of ESPRESSO [24]. A single pass is made over each selected prime implicant, removing output literals as long as the result remains a valid (hazard-free) cover. The input portion is then expanded, if possible. Unlike make-sparse, our current method makes no guarantee that the resulting product is maximally expanded. Nonetheless, despite its simplicity, the operation often yields significant reductions in literal count.
HFMIN is now widely used in several other burst-mode CAD packages, including the 3D method [34] and ACK [12]. It has also been used as part of the asynchronous tool suite at Intel Corporation, where it has been applied in the design of a high-speed experimental asynchronous Instruction-Length Decoder (see [2]).