Power Efficient Division and Square Root Unit

Performance/Power Space Exploration for Binary64 Division Units

“…The library is not the latest available, but it is the one used in [8] and this makes the comparisons straightforward. Moreover, it is a modern library with three types of cells for low power design (standard, high and low V t threshold voltages) which guarantees results are scalable to smaller features (e.g., 45 or 32 nm).…”

“…We compared the radix-8 a = 4 unit to a radix-4 unit similar to that of [10], and to the radix-16 unit of [8].…”

“…A comparison in terms of energy efficiency of the two approaches is done in [8] by comparing a radix-16 digit-recurrence unit (similar to the one of the Intel Core2) and a Fused Multiply-Add (FMA) floating-point unit (FP-unit) augmented with tables and by-passes to implement the Newton-Raphson division by reciprocal multiplication. The results show that the digit-recurrence approach is much more energy efficient.…”

“…1 shows the PpW trends for binary64 (doubleprecision) division units for different clock rates. The data are derived from [8] for a radix-4 division unit similar to that of the ARM processor [10] and a radix-16 unit similar to the one of Intel's Core processors. The clock period used to clock the unit is expressed in fanoutof-four (FO4) delay units.…”

Fast calculation of inverse square root with the use of magic constant – analytical approach

“…Many algorithms can be used to approximate inverse square root functions [12,13,14,15,16]. All of these algorithms require initial seed to approximate function.…”