Loading...

The Analysis of Low Power Embedded Systems through the Calculus of Variations

Jafari Nodoushan, Mostafa | 2020

359 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 53265 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Ejlali, Alireza
  7. Abstract:
  8. Embedded systems usually have severe limitations on power consumption. One of the reasons is that many embedded systems are battery operated. On the other hand, the battery is a non-linear source of energy, where the lifetime depends on how the system drains power from the battery, or in other words, the battery lifetime is dependent on battery discharge current curve. The nonlinear behavior of batteries is due to their inherent characteristics, where the rate capacity and the recovery effects are believed as the two most well-known and important factors. Also, embedded systems are usually fanless with limited cooling equipment, therefore temperature management is a prominent issue in these systems which must be done via controlling the power consumption.Most of the previous works on power, energy, and thermal management involve discrete time and discrete value factors when devising management techniques. For example, they assume there is a finite set of possible supply voltages that can be applied to the system. Even those techniques that consider continuous parameters usually use discrete time approach. A branch of mathematics is calculus of variations (CoV), where system parameters are considered as continuous functions of time. In this thesis we utilize CoV for battery and temperature management in embedded systems. Using CoV, we consider supply voltage and operational frequency as continuous functions of time, and we try to provide system level techniques to shape these functions (determine the right curve) to achieve objectives such as improved battery lifetime and minimize temperature variations. We believe because of the nature of the factors like supply voltage and operational frequency that are continuous functions of time, CoV must be very effective when devising system-level power management technique.Contrary to the research on power management of embedded systems, where little work has been done considering system parameters such as voltage and frequency as continuous functions of time, there is no work on computation and analysis of optimal battery discharge curve. In this thesis we try to calculate and analyze the optimal battery discharge current curve with the aim of maximizing battery lifetime. In other words, the lower bound of battery charge consumption in a DVS-enabled system is calculated. This is done for both tasks with known an unknown (with known probability distribution function of execution time) execution times. First, the calculations are done for single tasks, then it is extended to the system with periodic tasks. The results of our calculation for different tasks show that battery charge saving is dependent of the probability distribution function of task execution time and also the utilization of tasks. Also, the achievement of this approach for periodic tasks is not reasonable because of practical considerations.Another issue addressed in this thesis is the minimization of temporal thermal variation in a system aimed at reducing aging. Due to the impact of temporal thermal variations on chip aging, here, using the calculus of variations, the optimum task execution speed curve is computed to minimize these variations
  9. Keywords:
  10. Thermal Management ; Embedded System ; Variation Calculus ; Battery Lifetime Estimation ; Dynamic Voltage and Ferquency Scaling (DVFS) ; Battery Lifetime Management ; Power Consumption

 Digital Object List

 Bookmark

  • Signs.pdf
    • Sign Form
    • Declaration Form