Dual Mode Logic (DML), which was recently introduced by our group, offers the possibility to operate digital gates either in the static mode to save energy, or in the dynamic mode to increase speed albeit with a higher delay or energy consumption, respectively. We showed that on-the-fly switching of critical paths between the static and dynamic modes enabled system self-adaptation to computational needs achieving both high speed and low energy consumption. In this paper, for the first time we show that the mixed mode operation of a DML based datapath can efficiently reduce design sensitivity to process variations at near threshold voltages. Specifically, the number of gates operating in the dynamic mode (when the datapath is switched to the high-performance mode) is selected as a function of the process corner. The number of dynamically operated gates can be adjusted during the post-silicon phase or at run-time with an architectural level solution. In a basic proof of concept, simulations of a chain of 20 NAND/NOR gates demonstrated that process variations were successfully alleviated by utilizing an optimal configuration of the chain. The DML design can meet CMOS TT performance requirements in the SS corner and save energy by 18% in the FF corner. A 64-bit ripple carry adder (RCA) confirmed the advantages of DML over CMOS for different optimization points.
Process variation-aware datapath employing dual mode logic
Taco Ramiro.;
2018-01-01
Abstract
Dual Mode Logic (DML), which was recently introduced by our group, offers the possibility to operate digital gates either in the static mode to save energy, or in the dynamic mode to increase speed albeit with a higher delay or energy consumption, respectively. We showed that on-the-fly switching of critical paths between the static and dynamic modes enabled system self-adaptation to computational needs achieving both high speed and low energy consumption. In this paper, for the first time we show that the mixed mode operation of a DML based datapath can efficiently reduce design sensitivity to process variations at near threshold voltages. Specifically, the number of gates operating in the dynamic mode (when the datapath is switched to the high-performance mode) is selected as a function of the process corner. The number of dynamically operated gates can be adjusted during the post-silicon phase or at run-time with an architectural level solution. In a basic proof of concept, simulations of a chain of 20 NAND/NOR gates demonstrated that process variations were successfully alleviated by utilizing an optimal configuration of the chain. The DML design can meet CMOS TT performance requirements in the SS corner and save energy by 18% in the FF corner. A 64-bit ripple carry adder (RCA) confirmed the advantages of DML over CMOS for different optimization points.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.