Design of a high-speed arithmetic datapath

This paper presents the implementation of a 64/32 bit floating-point datapath circuit (WTL3*64). This high-speed CMOS circuit integrates a 2 Kbit six-port register file, an independent 64/32-bit alu, a multiplier, and a divide/square-root unit for IEEE binary floating-point format numbers. It is fabricated using a 1.2 micron two-layer-metal CMOS technology. Operating at 60 ns cycle time, the 165000 transistor/147600 square mil chip provides 33 double-precision Mflops of peak performance. The register file, together with special registers, a flexible input/output interface, features for high level language support, and 1.5 Watts of power facilitate ease of integration of the device into various computer systems. >