Stochastic Gradient Descent Optimization for Low Power Nano-CMOS Thermal Sensor Design

  • Authors:
  • Oghenekarho Okobiah;Saraju P. Mohanty;Elias Kougianos;Oleg Garitselov;Geng Zheng

  • Affiliations:
  • -;-;-;-;-

  • Venue:
  • ISVLSI '12 Proceedings of the 2012 IEEE Computer Society Annual Symposium on VLSI
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

The drive for ultra efficient and low-cost portable devices continues to push the need for low power circuit designs. The increasing transistor density and complexity of IC designs aggravates the task of producing efficient low power and low cost design. The short time to market (TTM) also increases this burden on designers, as optimal designs have to be produced in an ever decreasing amount of time. This paper presents an optimization design flow methodology that optimizes the power (accounting leakage) consumption of integrated circuits (ICs). The design flow incorporates a stochastic gradient descent (SGD) based algorithm and is implemented using a 45 nm thermal sensor circuit as case study. Power-efficient high-sensitive thermal sensors are important to reduce the burden on the systems or circuits that they are implanted to sense. Experiments are performed to apply the proposed design flow methodology on the thermal sensor with the power consumption as the design objective while keeping the temperature resolution as a constraint. Experiments on full-blown (RCLK) netlist of sense amplifier show a reduction in power consumption by 38%.