Power-efficient high-speed interface circuit techniques

Inter- and intra-chip connections have become the new challenge to enable the scaling of computing systems, ranging from mobile devices to high-end servers. Demand for aggregate I/O bandwidth has been driven by applications including high-speed ethernet, backplane micro-servers, memory, graphics, chip-to-chip and network onchip. I/O circuitry is becoming the major power consumer in SoC processors and memories as the increasing bandwidth demands larger per-pin data rate or larger I/O pin count per component. The aggregate I/O bandwidth has approximately doubled every three to four years across a diverse range of standards in different applications. However, in order to keep pace with these standards enabled in part by process-technology scaling, we will require more than just device scaling in the near future. New energy-efficient circuit techniques must be proposed to enable the next generations of handheld and high-performance computers, given the thermal and system-power limits they start facing. In this work, we are proposing circuit architectures that improve energy efficiency without decreasing speed performance for the most power hungry circuits in high speed interfaces. By the introduction of a new kind of logic operators in CMOS, called implication operators, we implemented a new family of high-speed frequency dividers/prescalers with reduced footprint and power consumption. New techniques and circuits for clock distribution, for pre-emphasis and for driver at the transmitter side of the I/O circuitry have been proposed and implemented. At the receiver side, new DFE architecture and CDR have been proposed and have been proven experimentally