Power management for increasingly complex microprocessor systems-on-chips (MPSoCs) is a significant challenge. Existing linear control models are ineffective where communications are via a network-on-chip (NoC), and these controls cannot adequately manage power dissipation. This can lead to inefficient communication and resource allocation in MPSoCs.
The authors propose a new power control approach based on fractal-state equations. They report significant runtime power savings for a variety of benchmark applications. The 20-page paper, while not for the faint of heart, should be of interest to those in the field.
NoCs are discussed in the introduction, and several recent papers are cited. Section 2 reviews the state of art regarding power and thermal management techniques. Section 3 considers the concept of fractional calculus and how it can be applied to the problem. Section 4 presents the fractional optimum control problem and summarizes the optimality conditions for the proposed power management algorithm. This is supported by several pages of mathematics. Section 5 covers the experimental setup, presents the experimental results in detail, and describes the results of the algorithm. The authors compare their work favorably to more classical power management approaches, cite limitations, and discuss extensions to the work. Section 6 concludes with a summary of the paper’s main contribution. The authors discuss the nature of future research in support of their insights. Approximately two-and-a-half pages of references conclude the paper.
While the paper is thought provoking, the mathematics described in section 4 are sketchy and formidable; anyone not comfortable with the math can accept the results or could consult the authors for further details. When I was working on my thesis, I would have called section 4 the “mathematical barrier.” A reader who questioned the thesis conclusions would be directed back to the impenetrable math in the barrier, because “anyone versed in the field would obviously agree with the conclusions,” thus deflecting any adverse reaction. (As a reviewer since 1967, I have always found mathematical barriers a concern, and this one is a great example.) Nevertheless, the proof will be in the results of any succeeding work.
Altogether, the preliminary results seem interesting, and follow-up results should either confirm or discredit the approach. One can hope for the former.