Evolving On-Chip Power Delivery through Particle Swarm Optimization

An evolving on-chip power delivery method is developed for the adaptive voltage assignment of a given voltage domain. The reference voltages of the on-chip voltage regulators (OCVRs) are determined and set through particle swarm optimization (PSO) to negate the effects of transistor aging, process variation, and power supply noise induced variation in the load circuit, OCVRs, and on-chip timing sensors. The on-line learning of the optimum voltages that evolve with time reduces the static voltage guard-band added during the design of the power delivery network for worst case process, temperature, and aging induced timing variation of a digital circuit. Applying the proposed PSO based on-chip power management technique ensures a minimal voltage assignment without incurring any timing violations on the critical paths, which also evolve with time. The run-time adaptive voltage delivery technique is applicable to any processor architecture as demonstrated through simulation of a four core multi-processor implemented in a 7 nm PTM FinFET technology. Results indicate an average reduction of 35% and 38% in, respectively, the dynamic power consumption and transistor aging.