Advanced Ordering Search for Multi-level Approximate Logic Synthesis

Approximate computing, an emerging design paradigm for error-tolerant applications, has received considerable interest recently. Approximate logic synthesis (ALS) is an automatic process to generate approximate circuits. Many existing ALS methods are implemented in an iterative greedy way. For each iteration, they choose a local approximate change (LAC)with the highest score. Since greedy approaches have the shortcoming of easily getting into local minima, we apply two advanced search algorithms, beam search and Monte Carlo tree search, to determine a better ordering of LACs. Each algorithm achieves an order in which the area of approximate designs is improved.