UC San Diego

As semiconductor technology advances in the ultra deep sub -micron era, on-chip global interconnections have been an ever-greater barrier to achieving high-performance and low -power for the increasingly larger system-on-chip (SoC) designs. Various on-chip interconnection schemes are proposed to tackle the scaling issue of global wires by manipulating the wire operation regions, changing signaling methods, and applying different equalization techniques. Optimization frameworks are also proposed to aid the transmitter-wire-receiver co-design based on user- defined constraints. For the six representative global interconnection schemes, we investigate their performance metrics with technology scaling by performing optimizations using the proposed SQP-based framework. A set of simple models is also developed to enable early- stage system-level analysis. Performance of different interconnection schemes are predicted and compared over several technology nodes. We further perform studies on the pipelined $RC$ interconnection by exploring its performance metrics with voltage and technology scaling based on different design objectives. A performance evaluation flow is developed to generate the optimal designs for given objectives. Also, impacts of pipelining depth, voltage and technology scaling are illustrated. Finally, we propose an energy-efficient high-speed on-chip global interconnection by employing continuous-time active equalization. Modeling and design of transmitter and receiver circuits are discussed. Analytical formula of received eye-opening is derived for system-level design planning. We further perform transmitter-receiver co- design through an optimization framework and explore the design space to generate design based on best energy- throughput tradeoff