Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Model reduction for multi-terminal RC circuits Roxana Ionut¸iu1
3
Industry supervisor: Joost Rommes2 Supervisor: Wil Schilders 2 3 Supervisor: Athanasios C. Antoulas 1 2 NXP
1 Jacobs University, Bremen, Germany Semiconductors, Eindhoven, The Netherlands 3 TU Eindhoven, the Netherlands
4th April 2010
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Smaller feature sizes but increasing complexity Parasitic interconnect dominates complexity of VLSI designs electromagnetic coupling effects circuit performance power
inadequate for simulations memory capacity, CPU time
Solution Interconnect modelling and reduction synthesis, optimization, verification
Intel 4004. First microprocessor (1971)
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
MOR for electrical circuits Original system
Σ C ,G , B , BT , D
Reduced system
Model Reduction
Large matrices n = 1M
Small matrices n = 1k
Sparse
Ideally Sparse
Chip
Network synthesis
Modified nodal analysis
Original circuit
Extraction
Layout
,G ,B ,B T , D Σ C
# R = 1k # C = 1k # L = 1k # nodes = 1k # terminals = 10k
# R = 1M # C = 1M # L = 1M # nodes = 1M # terminals = 10k ●
Reduced circuit
long CPU time
Full ● numerical hurdles Simulation
Simulation
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Modified nodal analysis (MNA) of RLC circuits From Kirchoff’s current and voltage laws, and the branch constitutive equations Dynamical system (differential algebraic) d v(t) C 0 G El v(t) Ev = iin (t) + 0 L dt iL (t) −El∗ 0 iL (t) 0 | {z } | {z } | {z } | {z } | {z } ˙ E −A B x(t) x(t) v(t) ∗ E∗v 0 = y(t) | {z } iL (t) | {z } C x(t)
x internal variables: node voltages and currents through Ls iin system inputs: currents injected into terminals y system outputs: voltage drops at terminals /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Model order reduction Given original system Σ: dimension n, p terminals ˙ (A, E) B E x(t) = A x(t) + B u(t) Σ= y(t) = C x(t) + D u(t) C D x ∈ Rn×1 , E ∈ Rn×n , A ∈ Rn×n , B ∈ Rn×m , C ∈ Rp×n , D ∈ Rp×m b dimension k , p terminals, p ≤ k Find reduced Σ: ) " ˆ ˆ ˆx ˆx ˆ u(t) ˆ˙ (t) = A ˆ(t) + B E ˆ = (A, E) Σ ˆ ˆx ˆ(t) = C ˆ(t) + D u(t) C y
n # ˆ B D
ˆ ∈ Rp×k ˆ ∈ Rk ×k , B ˆ ∈ Rk ×m , C ˆ ∈ Rk×1 , Eˆ ∈ Rk×k , A x Construct W, V to project Σ on a k -dimensional subspace ˆ = W∗ EV, E
ˆ = W∗ AV, A
ˆ = W∗ B, B
ˆ = CV, C
D=D
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Challenges in MOR
The usual targets achieving accuracy and small dimensionality preserving system properties: stability, passivity New requirements for multi-terminal circuits Critical for (re)simulation: preserving structure → convenient synthesized models preserving sparsity [few circuit elements] → memory and CPU requirements
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
How to meet MOR challenges ˆ = W∗ EV, E
ˆ = W∗ AV, A
ˆ = W∗ B, B
ˆ = CV, C
D=D
V, W ∈ Cn×k , k ≥ p, p = #terminals Existing methods: Krylov/SVD-/EVD-based V, W satisfy the usual targets: accuracy, stability, passivity ˆ A ˆ are dense → expensive to re-simulate! when p > 100, E, New methods for multi-terminal circuits ˆ A ˆ with preserved introduce graph-theoretical tools → E, structure & sparsity maintain the accuracy, stability, passivity requirements /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Sparse reduction of R networks [Rommes, Schilders: IEEE TCAD ’10] [Rommes ’09 - COMSON MOR School]
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Reducing large circuits with many terminals Candidates: circuits with O(103 ) nodes and O(102 ) terminals (and beyond!) Divide Partition network graph into smaller sub-networks identifies structure fewer #nodes and #terminals per sub-network reduction is simplified algorithmically and computationally Conquer Reduce sub-networks individually can use an MOR method of choice allows better control on sparsity /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Multi-terminal RC transmission line 1441 1406 1440
1442 1443
1439
1407 1444
1431
1438
1433
1410 1409
1408 1422 1411
1445 1402
1435
1446
1404 1401
1382 1384
1381
1426
1392 1383
1451 1479
1400 1385
1452 1379
1368
1454 1478
1399 1386
1453 1377
1376
1477
1464
1374
1396 1389
1371
1367
1465 1488
1398
1387
1456 1375
1455 1476
1485
1395 1390
1369
1462 1484
1394 1391
1378
1480 1463
1486
1393
1380
1448
1481 1482
1483 1417 1418
1403
1405
1447 1450
1449
1427
1425 1414
1419
1432
1430
1428 1429
1424 1415
1420
1416
1434
1436
1437
1423
1421
1412
1413
1397
1388 1457
1458
1475
1466
1373
1372
1785 1784
1370
1473
1487
1467 1474
1489
1492
1787
1543
1461 1460
1459
1366
1786 1783
1542
1472
1468
1490
1544
1517 1518
1471
1469
1788 1789
1499
1782
1541
1519 1516
1470
1491
1520
1618
1346
1605
1534
1533
1552 1364
1577
1631
1580
1856
1854
1863
1739
1736
1866
1735
1704
1869
1722
1724 1683
1712
1897
1902
1880
1884
1891 1885
1333 1318 1332
1248
1319
1924
1331 1320
1316 1306
1330
1315
1321 1307 1329
1314
1322 1308
1255
1313
1328 1323
1256 1309
1258 1222
1213
1223
1212
1312
1327 1324
1257
1310 1203
1259
1311
1326 1325
1260
1305 1204
1304
1292
1210
1231
1261
1211
1224
1925
1317
1243
1199 1252
1253 1201
1291
1290 1205
1303
1293
1263
1230
1209
1225
1928
1901
1882 1881
1890
1883
1892
39 1886
1251
1254 1215
1214 1202
34
1900
35 1904
1903 1893 1889
1247
1244 1250 1245 1249 1217
1216 1200
1926
1899
1896 1879 1894
1906
1888
1711
1699
1898 1895 1682
1877
1905
1246
1242
1221
1208
1226
1294
1289
1262 1206
1229
1927
1302
1296
1264
1295
1288
1265
1207 1301
1227
1298
1299
1266 1269
1287
1300 1193 1192
1182
1178
1198
1194
1282 1286
1270 1183
1179
38
1181
1195
1196
1271
1188
1280
1272 36
1273
1278
1274
40
1025
1159
1277
1275
1160
1026
1010 1020
42 1157
115843
1149
1027
1276
1008
1155 1154
1156
1150
1151
1153
997
1028
1007
1018 998
1029
1006
1017
1136
1092
1946 1140 1142
118
116
99
1016
981
97
982
983
90
96
94
985
84
1965
986
979 966 978
943 967
977
942
1049
1121
928 941
1051
976
927
3211
940
1053
975 970
931
1054
1107
67
65
71
3217
939 932
1056
974
971
925
64
1974
933
1058
1108
56
66
3216
3221
3214
3213
3215
938
1057 1117
1972
3220
3212
926
1055
1118 72
1973
3219
969 930
1052 1119 1106
68
70
3218
968 929
1050 1120 1105
69
74
73
1971
p = 22 terminals
980 965
75
79
80
1970
987 988
964
944
946
1104 82
81
1969 1968
989
963
945
1047 1048
78
76
83
1964 1966
990
962
947
1046 1122 1103
1967
991
1045
1123 1102 77
87
948
950
1044
88
86
85
1962
961
1043
1124 1101
95
1961 32
949
951
1042
89
93
1959
1963
992 984
1100 1960
960
1041
1125
1958
952
954
1040
92
953
955 1039
1126 1099
98
1957
993
1037 1038
91
101
1955
959
956
1036
1128 1127 1098
1956
994
957
1035 1129
1131
102
100
958
45 1130
103
115
117
1015
996
1034
104
1097 1953 1954
995
105
1132 1096
1952 1951
1014
1002
1033
106 1133
1095 114
112
1003
107
1134
1094 113
119
1001
1032
111
124
123 120
1004
110
1135
1093
122 121
1949
1000
1031
109
125
1141
1948
1950
1005
108
1143 1947
1943
999
1030
127
1139
1013
126
1144 1152
1012
44
1138 1137
1091
1145
1944
1011 1019
41 1148 1147
1941
1945
n = 3254 nodes
1009 1021
1161
1164 1146 1940
1942
1024 1022
1176 1162
1163
1171 1165
1939 33
1279
1174
1172
1170 1166
1938
1284 1023
1177
1175
1173 1187 1169 1168
1283
37
1184
1186
1190
1189
1935
1937
1281
1285
1180
1197
1185
1191
1936
1297
1268 1267
1228
1929 1930
1931
1932
1167
1677
1710 1723
1684
1876 1878
1907
1887
1218
1219
1220 1234
1233
1232
1934
1758
1824
1709
1713
1703
1686
1685
1875
1908
1241
1237
1236
1235
1919
1920
1923
1933
1823
1678
1714
1702 1700
1725
1240 1239
← RC network as a graph [shown are only R connections]
1818
1757
1708
1715 1701
1726 1719
1911
1238 1912
1915
1916
1707
1718 1687
1835
1706 1716 1721
1727
1688
1690
1909
1910
1913
1914
1917
1918
1921
1922
1836
1717 1689
1692
1760 1821
1837 1720 1705
1728
1874 1873
1834
1335
1729 1698
1691
1827
1833 1838
1845
1844
1822
1828
1829 1826 1832
1732 1839
1843 1730
1759
1825 1830 1831
1846
1840
1842
1731
1679
1754
1337 1753
1848 1748
1747
1733 1841
1737
1694
1756 1755
1850
1847
1746
1676
1675 1851 1674
1849 1752
1749
1745 1734
1334
1872
1338
1852 1673 1680
1751 1672 1750
1744
1741
1740
1761 1336
1339
1671
1742 1743
1864
1871
1817
1820
1762 1763
1670 1853
1861
1693
1814
1819
1764
1650
1651 1668 1855
1865
1697
1766
1765 1649
1669 1653
1857
1738
1767
1662 1663
1648 1667 1652
1858
1862
1681
1816
1664
1666
1654
1645
1859
1695
1815
1768 1661 1657
1665
1589 1647
1646 1644
1860
1867
1813
1769 1660 1659
1658 1590
1655
1641
1643
1696
1810
1812
1563
1361
1588 1656
1642
1809
1770 1562 1566
1592 1591
1594 1360
1868
1806
1564
1567
1587 1593
1359
1811
1561
1586
1595
1358
1640 1639
1870
1805
1771
1565
1357 1635 1637
1559
1568
1584
1596
1636
1808
1772 1560 1570 1583 1569 1585
1597
1638
1558
1571 1582
1598 1340 1356
1634
1807
1773 1555
1581
1572
1600 1354 1355
1800
1557 1573
1599
1353
1633
1801
1579
1574
1601
1352
1632 1629
1804
1774 1556
1576
1628
1630
1799
1803
1553 1575
1362
1802
1775 1554
1578
1363
1602 1351
1798
1776 1551
1532
1365
1603
1796
1797
1777
1535
1501
1604 1341 1350
1795
1794
1778
1550
1531
1530
1606 1348 1349
1627
1792
1793
1502
1624 1626
1549 1536
1503
1347 1623
1548 1537 1528
1529
1504 1500 1607
1625
1779
1527
1526
1506 1505
1608 1345 1622
1547 1538
1507
1609
1344
1620
1621
1780
1524
1508 1525 1610 1342
1343
1619
1791
1790
1509
1612 1613
1616
1617
1546 1539
1611
1614 1615
1523
1510 1522
1514
1497
1495
1781
1540
1511
1513
1498 1496
1545
1512 1521 1515
1493 1494
973
972
935
1059
3222
3210
937
1116 31
63
1975
57
55
62
934
1060
1109
936
910
1061
61
1978
54
58
60
916
921
1064
881 1088
1066
3209
905
3196 3205
3197
3202 903
884
3201
3206
3199
3198
134
3200
888 1071
135
874
137
1073
blue and green nodes are connected also to Cs
3207
889
136
138
3195
902
1086
1068 46
3208
906
887 133
883
10701072
1067 47
1987 1988
904
886
882
1087 1069
48
1985
3203
876 875
880
1090 129
49
3204 879
914 885
920
1065
50
1986
877
878 913
915 130
919
1112
131 1089
51 1983 1984
908
912
918 923
1111
1113 128 52
1982 1981
3253 907
922
1063
53
59
1979 1980
911
917
924
1062
1110
1114 1977
909
132
1115 1976
3194
873 901
1989
1085
869
1074 871
872
825
824
891
1075 866 838
840
1082
820
2014
3247 3245
3252
154
689
690 153 805
688
691 692 693
687
672
671
151
694
766 2024
673
670
150
798 2025728
3246
3251
3244
3243
3240
806
802 804
799 714
729
785
3250
3242
3238
177
176
746
738
152
784 2023 2018
3248
3249 3241
3239
800 801 765
731
730
2022
3230
174
175 155 807
803 764
852
853
2021
2019
3231
173
172 156 808
745
747 763
851
854
2016
2020
3232
3237
170
171 157 809
744
749
748 737
2017
3235
3236
169
168 158 810
743
750
736
762
850
166
167 159 811
742 751
735
761
849
855
3229
165 164 160 812 741
733
734
760
848
856
2013
2015
3223 162 163
161 813 740
732 752
759
847
857
2012
3224
815 814
754 755
753
758
846
858
2011
3225
896
1079 833
739
757
845
859
2008
2010
3228 897
894 816 831 895
1080
860
2007
2009
3227 893
817 830
829 1078 832
835
834
756
844
898
821
1077
836 1081
2004
2006
3226
892 827
828
837
841
842
843
861
899
822
1083 1076
865
864
863
862
2003
2005
3234
819
826
839 1995
1996
1999
2000
2002
900
823
1084
1992 1993
1994
1997
1998
2001
3233
818
867 868
3193
890
870
1991 1990
695 2027
767
686
2026
715 727 149
696 716
768
796
787
148
685 2029
726 675 717
668 769
697
724 2034
2035
719
794682
771
146
666
2033
2032
718
683 770
147
698 677 789
2030
2031
684 725
676
667
795
788
674 2028
669
797
786
723 699 772
145
790
2037
2036
720
793
678 681
722
665
700 792
701
776 783
2040
713 142
702
782
777
2041
139 704
661 656
703
143
781
778
657
2044 144
705
710
179 706
780
2043
2045
779 652
658
659
2042
711
178
660
2039
712
775 663
655
2038
773
774
141
664
662
140
721
791
679 680
653
708 180
651
2046
2086
642
465
440
196
432
195 433
194
193
422 191
325
327
300
337
417
328 420
301
320
329
224
318
319
297 225 296
304 223
309
313
312
311
306
308
228
244
281
248
283 259 282
284
258
260
288
249 257
280
253
254
279
278
277
3188
276
3189
3168
3187
3186
270
3185
3089
3048 3047
3050
3049
2307
3033
2297
3045 3046
3051
2298
3034
2306
3044 3052
2299
2305
3041 3035 3043
3037
3063
3055
3062 3064
3061
2279
2251
2278
2998
3066
3058
3060
2248
3072
3009 3010 2335
267
2992
2991
2989
2990
2988
2350 2987
2971
2344
2972
2338
2969
2339
2968
2986
2352
2984
2343
2973
2333
2985
2351
2337
2970 2334
2363
23542353 2983
2342
2974
2355
2981
2340
2332
2945
2323
2967
2362
2975
2324
2966 2330
2361
2953 2325
2943 2954 2955
2357
2965
2980 2979 2961
2977
2929
2359 2964
2930
2899
2962
2358
2327
2941
2960
2328
2940
2982
2976 2978
2360 2326
2942
2341 2356
2331
2952 2944
2956
2959 2934
2939 2933
2938 2937
2963
2889
2900
2936
2935
2932 2901
2902
2884
2865 2879
2905
2894
2908
2864 2880
2870
2878
2907
2906
2374
2869
2925
2924
2882 28672881
2868
2892 2895 2904
3161
2866
2883
2885 2373
2896 2903
2927
4
3175 3176
2890
2891
2893
2926
2886
2888
2928 2372 2931 2887 2897
2898
2957 2958
3157
2993
2994
2336 2345
2329
2364 2366
3158
3160
2349
3008 2346
3011
2365 2367
2951
3150
3151
3156
2995
2348
2347 3020
3022
2368
2947
2946
2996
3004
3021
3012
3023
3071 2949
3006
3007 3019
2310
2369
2948
3003
2316 2315
3013
3024 3073
3005
2317
2311
3025 2370
3070 3076
2314 3018
3026 3074
2318
2320
2313
3027 3014
2249
3069
3075
3078
2871
2371 3172 3180
3178
2875 2375
2909
2860
2922
3177 2923
2872
2687 2873
2686
2848
2688
2846
2857
2685
2844 2851
2853
2856
2845
2854
2912
2919
2876
2847
2850 2859
2911 2918
2843
2852
2803
2917 2855
2913
2689
2800
2684
2801
2802 2804
2842
2916 2836
2835
2678
2837
2805
2683 2841
2834
2675
2840
2839
2680 2691
2838
2682
2496
2497 2798 2809
2673
2796
2650
2696
2780
2635 2634
2630
2585 2560
2558 2561
2557
2552 2551
2553
2581
2562
2556
2563
2564
2394 2749
2748 2744
2742
2554
2565
2566
2392
2393
2747
2743
2391 2395
2398
2550
2389 2549
2396 2388
2399
2548
2543
2542
2397 2547
2545 2408
2384
2544 2546
2412
2385
2413
2411
2439
2555
2750
2540
2442
2567
2746
2745
2570
2751
2741
2541
2539 2426
2383 2415
2425 2441
2568
2569 2580
2579
2390 2573
2752 2740 2739
2400
2538 2438 2437
2444
2582
2583
2578
2754
2753 2738 2737
2537
2407 2436
2445
2571
2574
2755
2736 2735
2536
2534 2435 2434
2443
2559
2586
2577
2572
2597
2756
2734 2733
2535
2533 2433
2446 2448
2584 2587
2576
2575
2758
2757
2732 2731
2401
2532 2432 2431
2447 2449
2598
2760
2730 2729
2427 2430
2450 2452
2588
2594
2595
2625 2759
2761
2728
2531 2528
2429 2428
2762
2764 2763
2727
2530
2451 2453
2596 2624
2765
2726 2725 2766
2529
2454 2474
2593
2382
2381 2623
2724 2768 2767
2402
2473 2475
2592
2627
2626
2622 2723
2722 2769
2770
2527
2478
2589
2629
2628
2599
2621 2721
2720 2772
2525
2526 2455
2590
2632
2605
2604
2719
2771 2524
2406 2472
2591
2633
2606
2603 2620
2718 2773 2774
2522 2470
2471
2631 2636
2619 2717
2716
2523
2521 2469
2468
2477 2476
2607
2602
2715
2776 2775
2520 2467
2480
2601 2618
2714 2777
2403 2456 2466
2465 2481
2479
2638
2637
2600 2617
2713
2712
2778 2519
2464
2482
2608
2616 2711
2710
2779 2518
2457
2485 2484
2639
2609
2611
2709 2781
2483
2640
2610
2612 2655
2708
2782 2517
2463
2487
2615 2656
2707 2784
2783 2516
2458
2486
2642
2613 2654
2657 2705 2785 2706 2786 2515
2514 2461
2462
2641
2379 2658
2703 2788 2704
2787
2513 2460
2490 2488
2614
2659
2701 2789 2702
2404 2512
2459 2489
2644 2653
2670
2700
2511 2405
2493
2492
2645 2380
2669
2660
2699 2792
2790
2508
2494 2822
2491
2652 2671
2668
2663
2698 2791
2510
2507
2820
2821 2823
2643
2672 2665
2662
2697 2793
2506 2509
2818
2819
2825
2646 2651
2695
2794
2505
2504
2667
2664
2694
2797 2501
2503
2378
2377 2824
2647
2666
2795 2500 2502
2816 2817
2648
2692
2693
2498
2661
2814
2815
2812
2811 2828
2826
2649 2676 2681
2495 28072808
2499
2813 2831
2830 2810
2674 2677
2806
2833
2832
2829
2827
2376
2679
2690
2799
2914
2915
2874
2849
2858
2910 2921
2920
2863 2877
2862
2861
3171 3179
6 sub-networks are identified
2997
2312 3016 30293015
3068 2247
2242 2243
2319
2321 3030
3067
2250
3017 2246 2245
3080
3079
3002
2322
3031
3056
3028 3059 2253
2244
3081
3001
3032
3065 2277
3057
2241
3084
3083
3039 2999
2300
8
2280
2252
2171 2172
3085
3086
3000
2301 2302
7
2281
2275
2224 2276
2254 2173
2170 3088
3040 3038
3054 2225
2274
2255
3090
3036
2304
2226
2227
22562273
3042
3053 2303 2296
2223
2259
3159 3162
3
3173
268
3181
2258
2257 2272
2240
3152
3154
371
3163
2
3174
273
272
2309
2308
2294
2283 2234
2235 2260
2236 2271
2239 2169 2174
3092
2295
2284 2233
2232 2261
2237 2270
2238 2168 2167
2950
3147
378
206
3155
3164 3169
3170
266
271 3184
6
2292
2293
2282
2228 2179
2175 2166
604
3096
2290
2291
2286
2285
2229 2180 2262
2178 2269
2176 605 603
3093
3153
265
3183
2289
2288
2287
2222
2230 2158
2159 2263
2177 2268
2162 602
3094
374
1
264
274 269
3182
3142
3143
3146
376
373
370
263 275
3190
2206
2219
2220
2221
2231 2160
2161 2264
2163 2267
2165 601 600
3097
3149 377
208
372
3165
262
247
2207
2218
2185
2184
2181 9
10 2265
2164 2266
606 599
3098
3100
3148
207 369 229
3167
250
3191
2208
2217
2183
2157 11
2149 2148
607 12
609
598 597 3101
3104
2186
2182
2154 2104
2105 2147
608 13
610
596
3109
375
251
261
3192
367 368
3166
289
255
230
290
252
256
2209
2216
2156
2106
574 14
611
3102
2187
2155
2153 2107
2146 612 15
575
3077
3144
3145 381
379 209
240
291 241
307
287 286
285
246
2210
2215
2188 2152
2150 2108
2109 2145 613 16
615
3082
5
384
382
380
3134
3137
3138
3141 383
361
362
363
2211
2214
2189 2151
2103 2110
2111 2144 614 17
616
595 594
3087
3140
210
222 364 347
239
292
234
227 245
221 332
348
365
2212
2213
2190 2102
2070
2112
2113 2143 573 18
617
3106
2205
2192
2191 2069
2071
2114
2115 2142 618 19
593
3091 3136
3139 205
385
346 366
293
232
233
238
294
231
305
345
237
235
226
310
344
236
295 317
316
314
342
343
330 303
315
242
243
3130
3133
539
537 356 386
2204
2193
2101 2068
2072
2116
2117 2141 619
576
591
3099
3135 538
387
2203
2100 2067
2073
2118
2119 2140
20
590
3095
536 218
357
3129
2194
2099 2066
2074
2120
620
621
589
3107
3103
540
217 355
220
2121
2139
21 622
588
3110
3125
3126
2202
2195
2098 2065
2075
2122
2123
572 22
587
592
3131 542 3132
388
353
219 359
3124
2196
2097 2076
2064
2124
2138
624
623
586
3111 3113
3105 3128
543
541
2125
2137
585
3114 3116
3108
3127 545
531
186
358
351
360
349
550
529
530 544
510
509
2201
2197
2096 2077
2063
2126
23
584 3115
3117
3112
546 511
533
534
535
352
336
350
340
341
331 298
338
339
418
299
302
512
532
507
508
214
2127
625
577
583
3118 3119
3123
2095 2062
2136
582
581 481
3120
3122
551
2198 2078
2061
2060
2128
24 627 578
580
482
553
2200 2094
2079 2058
2059 2129
626
2135
2134 628
579
483
555
3121 552
549
548
2199 2093
2080 2057
2056 2130
571
629 2133
565
564
554
524
525
187 547
25 2092
2081 2054
2055
2131
630
484
556
522 517 516 523
514
513
26 2091
2082 2053
2052
185
570 2132
567
566
562
563
520 501
515
527
528
398
389
216
354
334
335
416
421 419
324
323 322
321
333
215
213
211
414 192 415
390
212
413
424
423
392
391
412
425
434
435
436
394
393
411
426
439
438
437
326
27 2090
2083 2050
569
568
498 502 526 503
504
505
506
557
521 518
500
497 496
403 404
402 401
400 399
397 396
395
519
499
495
405
406
407
408
409
190
188
494 204
202
201
200
199
410
184
631
561
189
493 203
452
451
450
449
198
183
632
182 486 487
480
558
453
458
459
485 488 560
559
472
457 464
460
448
197
2051
490
491
492
454
463
461
447
428
427
181
489 478
473
462
446
429
430
431
633
634
476
475
471 474
445
444
443
442
2088
649 2084
479 643 470
455
441
2049
477
469 468
456
28
2085
2089
648
2048
636
639 637
467
30 29
650 2047
635 638
640 641
466
2087
709
654 707
647 646 645 644
2414
2422
2410
2409
2387
2386
2421
2440
2416
2424
2417
2423 2420 2419
2418
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
MOR for multi-terminal RC: “how to” Divide 1. graph-based network partitioning (strongly connected components, bi-connected components etc.) Conquer Per sub-network: 2. partition nodes & matrices selected nodes: xS ∈ Rp+m (terminals + some internal) remaining nodes: xR ∈ Rn−m internal nodes
3. reduce xR , keep xS Advanced option: fill-in minimizing node reorderings → improved sparsity Based on method for R-networks [Rommes, Schilders: IEEE TCAD ’10] /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
RC netlist reduction: the maths [I., Rommes: COMSON ’09]
On full circuit or per sub-network: Original circuit model:
G, C ∈ Rn×n , B ∈ Rn×p
From MNA circuit equations: (G + sC)x = Bu, partition x in: GR GK CR CK xR 0 +s = u. xS BS CTK CS GTK GS Moment revealing transformation [Kerns, Yang: IEEE TCAD ’97]: X = 0
0
G = XT GX, C = XT CX,
0
I 0
−G−1 R GK I
0
B = XT B, x = XT x /centre for analysis, scientific computing and applications
RC netlist reduction: the maths ... 0
0
G =
GR 0 0
GS 0
CS
0 0 GS
0
0
G , C ∈ Rn×n , B ∈ Rn×p
Transformed model:
0
, C =
CR 0 CK
0
CK 0 CS
GS − GTK G−1 R GK ,
=
CS + VT CR V + VT CK + CTK V,
0 BS
0
, x =
xR 0 xS
V = −G−1 R GK
0T
0
, B =
=
0
0
0
CK = CK + CR V 0
0
[(GS + sCS ) − s2 C K (GR + sCR )−1 C K ]x S = BS u {z } | {z } | Y0 R (s)
Y0 S (s)
0
Multiport admittance: Y S (s) already captures first 2 moments! Reduced circuit model:
b C b ∈ Rk×k G,
0 b = G0 , C b = C0 , B b = BS , x b = xS G S S
p≤k n b + sC) b x b b = Bu (G
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
RC netlist reduction: strategy & maths Benefits moment preservation → necessary for accuracy guaranteed stabillity and passivity → necessary in simulation Limitations 4 1. recall transform involving V = G−1 R GK → costly for O(10 ) nodes and beyond
2. fill-in in GS 0 and CS 0 → too many circuit elements! Solution structured reduction (i.e. per sub-networks) to resolve [1.] fill-in minimizing node reorderings to resolve [2.] /centre for analysis, scientific computing and applications
Industrial testcases Parasitic extraction RC models of a Low Noise Amplifier circuit (C45 technology)1 Inventory #ext. nodes #int. nodes #resistors #capacitors Analysis AC Noise SP PSS Speed up 1
TL 1 Original Red
TL 2 Original Red
79 29806 70338 12038
TL 3 Original Red
75 0 117 1047
65.83 s 0.12 s 59.68 s 0.13 s 82.81 s 0.21 s 793.13 s 2.93 s > 270x
33818 81843 12145
79 0 99 920
CPU time 60.11 0.17 s 54.82 0.18 s 110.7 0.19 s 424.28 3.33 s > 127x
27962 66068 9786
0 117 1032
NA2 NA NA NA
0.12 s 0.14 s 0.22 s 3.05 s ∞
Thanks: Gerben de Jong, Dennis Jeurissen (SDR Project, NXP Semiconductors) 2 No DC solution found (no convergence)
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
TL1 original matrices: n = 29885, p = 79 G
C
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
TL1 reduced matrices: k = 79, p = 79 Maximum reduction ↔ eliminate all nodes except terminals b b G C
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
TL1 reduced matrices: k = 2867, p = 79 Partial reduction ↔ preserve some internal nodes b b G C
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
TL1: Reduction Summary 2 reduced netlists: maximum reduced and partially reduced TL 1 Inventory
Original
#ext. nodes #int. nodes #resistors #capacitors
29806 70338 12038
Analysis AC Noise SP PSS
Reduced (max.) 79 0 117 1047
CPU time 65.83 s 0.12 s 59.68 s 0.13 s 82.81 s 0.21 s 793.13 s 2.93 s
Perc.
Reduced (partial)
100% 99.8% 91.3%
2903 4248 4382
90.2% 93.9% 63.6%
Speed Up ∼548x ∼450x ∼394x ∼270x
CPU time 1.46 s 1.71 s 2.74 s 17.46 s
Speed Up ∼45x ∼35x ∼30x ∼45x
Perc. 79
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
AC analysis - node “in”: magnitude Comparison: original, maximum and partially reduced
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
PSS analysis, time domain - node “out” Comparison: original, maximum and partially reduced
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
PSS analysis, freq. domain - node “out” Comparison: original, maximum and partially reduced [matched fundamental harmonics at 2 GHz]
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Noise analysis - NF Comparison: original, maximum and partially reduced → perfect match for range of interest [up to 10GHz]
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Outline 1
Introduction Motivation from electronics industry
2
MOR for multi-terminal circuits Circuit modelling General MOR framework Multi-terminal MOR
3
Results Netlist TL1 Circuit simulations
Netlist TL3 4
Summary
5
Bibliography /centre for analysis, scientific computing and applications
Reduction Summary: TL3 Immediate benefit from reduction! Original simulation not possible → error “NO DC solution found, no convergence” Simulation possible after reduction [quality “predicted” from experiments with TL1 and TL2] Inventory
Original
#ext. nodes #int. nodes #resistors #capacitors
27962 66068 9786
Analysis AC Noise SP PSS 3
TL 3 Reduced
Percentage
79 0 117 1032
CPU time NA3 0.12 s NA 0.14 s NA 0.22 s NA 3.05 s
No DC solution found (no convergence)
100% 99.8% 89.5% Speed Up ∞ ∞ ∞ ∞
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
TL3 reduced circuit simulations PSS: node “Vse2lse 0:2”
Noise: NF
/centre for analysis, scientific computing and applications
Introduction
MOR for multi-terminal circuits
Results
Summary
Bibliography
Summary MOR for multi-terminal circuits electronics industry needs new MOR methods → structure identification and sparsity preservation divide and conquer strategy → efficiency from reducing smaller subnetworks individually reduced netlists for multi-terminal RC networks (up to 100 terminals) significant reduction in #nodes, #Rs, #Cs → much faster simulations partial reduction possible → better approximation
netlists beyond 100 terminals → in progress
/centre for analysis, scientific computing and applications
For further reading Rommes, J., Schilders,W. H. A.: “Efficient methods for large resistor networks” IEEE Trans. CAD Circ. Syst. 29(1), 28–39 (2010) Rommes, J:, “Eigenvalue problems and model order reduction in the electronics industry,” COMSON Autumn School on MOR, presented in Terschelling, the Netherlands, Sept. 21-25, 2009. Kerns, K.J., Yang, A.T.: Stable and efficient reduction of large, multiport networks by pole analysis via congruence transformations. IEEE Trans. on Comp.-Aided Design of Integrated Circuits and Systems 16(7), 734–744 (1997) Ionutiu, R., Rommes, J.: “On synthesis of reduced order models” To appear in Lecture Notes in Electrical Engineering, Springer. Proceedings of Workshop on Model Order Reduction for Circuit Simulation, Hamburg, Germany, Oct. 30-31, 2008. Ionutiu, R., Rommes, J.: “Model order reduction for multi-terminal circuits,” COMSON Handbook (2009).