With the proliferation of floating point computing applications, the demand
for high performance, low power floating point hardware has increased. A n
ew architecture for low power floating point multiply-accumulate fusion is
presented. The proposed architecture minimises power consumption through tr
ansition activity scaling and data path simplifications. The switching acti
vity function of the proposed MAF is represented as a four-state FSM. Durin
g any given operation cycle, only a limited set of functional subunits are
active, during which time, the logic assertion status of the circuit nodes
of the unused functional units are maintained at their previous states. Cri
tical path delay and latency are reduced by incorporating data path simplif
ications and speculative rounding. The scheme offers a worst case power red
uction of more than 49%.