S-Space College of Engineering/Engineering Practice School (공과대학/대학원) Dept. of Electrical and Computer Engineering (전기·정보공학부) Theses (Ph.D. / Sc.D._전기·정보공학부)
Clock Tree and Flip-flop Co-optimization for Reducing Power Consumption and Power/Ground Noise of Integrated Circuits and Systems
- Joohan Kim
- 공과대학 전기·컴퓨터공학부
- Issue Date
- 서울대학교 대학원
- Clock tree synthesis; low power; post-placement optimization; simultaneous switching noise; peak current; power/ground noise
- 학위논문 (박사)-- 서울대학교 대학원 공과대학 전기·컴퓨터공학부, 2017. 8. 김태환.
- For very-large-scale integration (VLSI) circuits, the activation of all flip-flops that are used to store data is synchronized by clock signals delivered through clock networks. Due to very high frequency of clock signal switches, the dynamic power consumed on clock networks takes a considerable portion of the total power consumption of the circuits. In addition, the largest amount of power consumption in the clock networks comes from the flip-flops and the buffers that drive the flip-flops at the clock network boundary. In addition, the requirement
of simultaneously activating all flip-flops for synchronous circuits induces a high peak power/ground noise (i.e., voltage drop) at the clock boundary.
In this regards, this thesis addresses two new problems: the problem of reducing the clock power consumption at the clock network boundary, and the problem of reducing the peak current at the clock network boundary. Unlike the prior works which have considered the optimization of flip-flops and clock buffers separately, our approach takes into account the co-optimization of flip-flops and clock buffers. Precisely, we propose four different types of hardware component that can implement a set of flip-flops and their driving buffer as a single unit.
The key idea for the derivation of the four types of clock boundary component is that one of the inverters in the driving buffer and one of the inverters in each flip-flop can be combined and removed without changing the functionality of the flip-flops. Consequently, we have a more freedom to select (i.e., allocate) clock boundary components that is able to reduce the power consumption or peak current under timing constraint. We have implemented our approach of clock boundary optimization under bounded clock skew constraint and tested it with ISCAS 89 benchmark circuits. The experimental results confirm that our approach is able to reduce the clock power consumption by 7.9∼10.2% and power/ground noise by 27.7%∼30.9% on average.