A sub-threshold cell library and methodology

Abstract

Sub-threshold operation is a compelling approach for energy-constrained applications where speed is of secondary concern, but increased sensitivity to process variation must be mitigated in this regime. With scaling of process technologies, random within-die variation has recently introduced another degree of complexity in circuit design. This thesis proposes approaches to mitigate process variation in sub-threshold circuits through device sizing, topology selection and fault-tolerant architecture. This thesis makes several contributions to a sub-threshold circuit design methodology. A formal analysis of device sizing trade-offs between delay, energy, and variability reveals that while minimum size devices provide lowest energy and delay in sub-threshold, their increased sensitivity to random dopant fluctuation may cause functional errors. A proposed variation-driven design approach enables consistent sizing of logic gates and registers for constant functional yield. A yield constraint imposes energy overhead at low power supply voltages and changes the minimum energy operating point of a circuit.

(cont.) The optimal supply and device sizing depend on the topology of the circuit and its energy versus VDD characteristic. The analysis resulted in a 56-cell library in 65nm CMOS, which is incorporated in a computer-aided design flow. A test chip synthesized from this library implements a fault-tolerant FIR filter. Algorithmic error detection enables correction of transient timing errors due to delay variability in sub-threshold, and also allows the system frequency to be set more aggressively for the average case instead of the worst case.