Efficient Mux-Based Multiplier for MAC Unit

Multiply-Accumulate Units (MACs) are crucial for computer-intensive algorithms that are used in many application fields, such as machine learning, real-time embedded systems, etc. Therefore, the choice of the multiplier for MAC units has significant impact in the efficiency of these compute-intensive operations, particularly those executed on edge devices. The cost of fast full-tree or parallel-tree multipliers are high and hence limits applications to systems whose speed requirement is critical. The slow serial implementations are also usually mixed with the full tree to lower the cost to an acceptable range resulting in partial tree. In this paper, we propose a full tree of lower height with pre-computation of multiplicand multiples using multiplexer networks. It is demonstrated that for a radix-16 version of the multiplier, the partial product reduction network shrinks by a factor of two. The proposed design is more efficient consuming 35% less power and requires 13% less area than a standard Booth-recoded multiplier with comparable speed. Hence, the proposed multiplier design is suitable for integration in multiply-intensive applications such as DNNs (Deep Neural Networks) and other algorithms.