1 \chapterauthor{Xuexin Liu, Sheldon Xiang-Dong Tan}{Dept. Electrical Engineering,
2 University of California, Riverside, CA 92521, USA}
3 %\chapterauthor{Sheldon Xiang-Dong Tan}{Dept. Electrical Engineering, University of California, Riverside, CA 92521}
4 \chapterauthor{Hai Wang}{Univ. of Electronics Science and Technology of China,
5 Chengdu, Sichuan, China}
6 \chapterauthor{Hao Yu}{School of Electrical \& Electronic Engineering, Nanyang Technological University, Singapore}
9 % This research was supported in part by NSF grants under
11 % No.~OISE-1051797, and
15 \newcommand{\ud}{\,\mathrm{d}}
17 %\renewcommand{\algorithmicrequire}{\textbf{Input:}}
18 %\renewcommand{\algorithmicensure}{\textbf{Output:}}
19 %\renewcommand{\algorithmiccomment}[1]{\% \textit{#1}}
21 \chapter[GPU-Accelerated Envelope-Following Method]{A GPU-Accelerated Envelope-Following Method for Switching Power Converter Simulation}
24 % % Power converters have seen a surge of new trends and novel
25 % % applications due to their widespread use in renewable energy
26 % % systems and emerging hybrid and purely-electric vehicles. More
27 % % efficient simulation techniques for power converters are urgently
28 % % needed to meet more design constraints.
29 % In this chapter, we propose a new envelope-following parallel transient analysis method for
30 % the general switching power converters. The new method first exploits
31 % the parallelisim in the envelope-following method
32 % and parallelize the Newton update solving part,
33 % which is the most computational expensive, in GPU platforms
34 % to boost the simulation performance.
35 % To further speed up the iterative GMRES
36 % solving for Newton update equation in the envelope-following
37 % method, we apply the matrix-free Krylov basis generation
38 % technique, which was previously used for RF simulation.
39 % Last, the new method also applies more robust
40 % Gear-2 integration to compute the sensitivity matrix instead of
41 % traditional integration methods.
42 % %Furthermore, the resulted Gear-2 and matrix-free GMRES have been
43 % Experimental results from several integrated on-chip power
44 % converters show that the proposed GPU envelope-following algorithm leads to
45 % about 10$\times$ speedup compared to its CPU counterpart,
46 % and 100$\times$ faster than the traditional envelop-following methods
47 % while still keeps the similar accuracy.
49 \input{Chapters/chapter16/intro.tex}
50 \input{Chapters/chapter16/ef.tex}
51 %\input bdf.tex % now inside gpu.tex now
52 \input{Chapters/chapter16/gpu.tex}
53 \input{Chapters/chapter16/exp.tex}
57 In this chapter, we present a new envelope-following method for
58 transient analysis of switching power converters. First, the
59 computationally expensive step, the solving of Newton update equation,
60 has been parallelized on CUDA-enabled GPU platforms with iterative
61 GMRES solver to boost performance of the analysis method. To further
62 speed up the GMRES solving for Newton update equation, we have
63 employed the matrix-free Krylov basis generation technique. The
64 proposed method also applies the more robust Gear-2 integration to
65 compute the sensitivity matrix. Experimental results from several
66 integrated on-chip power converters have shown that the proposed GPU
67 envelope-following algorithm can lead to about 10$\times$ speedup
68 compared to its CPU counterpart, and 100$\times$ faster than the
69 traditional envelope-following methods while still keeps the similar
75 \item[Envelope-Following] In transient simulation of switching power circuits,
76 nodal voltage waveforms in neighboring high frequency clock cycles are similar,
77 but not exactly the duplicates. Envelope-following technique approximates
78 the slowly changing transient trend over a lot of clock cycles
79 without calculating waveforms in all cycles.
82 \putbib[Chapters/chapter16/biblio16]
83 %\bibliography{./envelope,../../bib/interconnect,../../bib/architecture,../../bib/simulation}