Dynamic operand transformation for low-power multiplier-accumulator design

The design of portable battery-operated devices requires low-power computation circuits. This paper presents a new multiplier-accumulator (MAC) design approach, which in contrast to existing methods exploits dynamic operand transformation to reduce power consumption. The key idea is to compare current values of input operands with previous values and depending on computed Hamming distance to use either original or two's complement form of the operands in order to decrease the transition activity of multiplication. Experiments show that such a formulation outperforms the related approaches minimizing the power dissipation of traditional MAC design almost by half with 31% area and 12% delay overhead. The circuit implementation is outlined.