Abstract: This letter presents novel approximate floating-point arithmetic units that leverage hardware-efficient rounding schemes to enhance computational accuracy while significantly improving ...