In multimedia Systems-on-Chips, the design of specialized IEEE-754-compliant floating point arithmetic units (FPU) is critical with respect to both operating speed and silicon area demand. Leading one anticipation is a well-known issue in the implementation of high speed FPUs. We investigated a novel leading one anticipation algorithm allowing us to significantly reduce the anticipation failure rate with respect to the state-of the art. We embedded our technique into a complete FPU and compared its performance against existing solutions, definitely showing both area savings and total latency reduction.
                        
                        
                        
                        
                            
                                Copyrights © 2012