An Asynchronous IEEE Floating-Point Arithmetic Unit

An asynchronous floating-point arithmetic unit is designed and tested at the transistor level using Cadence software. It uses CMOS (complementary metal oxide semiconductor) and DCVS (differential cascode voltage switch) logic in a 0.35 mu m process using a 3.3 V supply voltage, with dual-rail data and single-rail control signals using four-phase handshaking. Using 17,085 transistors, the unit handles single-precision (32-bit) addition/subtraction, multiplication, division, and remainder using the IEEE 754-1985 Standard for Binary Floating-Point Arithmetic, with rounding and other operations to be handled by separate hardware or software. Division and remainder are done using a restoring subtractive algorithm; multiplication uses an additive algorithm. Exceptions are noted by flags (and not trap handlers) and the output is in single-precision. Previous work on asynchronous floating-point arithmetic units have mostly focused on single operations such as division. This is the first work to the authors' knowledge that can perform floating-point addition, multiplication, division, and remainder using a common datapath.