Dynamic IR analysis » History » Version 1
jun chen, 03/29/2025 09:44 PM
1 | 1 | jun chen | # Dynamic IR analysis |
---|---|---|---|
2 | |||
3 | ## Road map (From Megy Wang): |
||
4 | 470 |
||
5 | |||
6 | ## Correlation plan |
||
7 | |||
8 | ### We need a VCD generator (toggle comb. cells at given times, disable VCD propagation in RK) |
||
9 | |||
10 | a. what if no toggle at all? |
||
11 | b. what if only inverter toggles? |
||
12 | c. what if several inverter toggle? |
||
13 | d. what if FF toggle? (with/without Q toggle?) |
||
14 | |||
15 | ### We need to verify Resistance (enhance analyze_pg_resistance, report_pg_resistance. similar results from RK) |
||
16 | |||
17 | a. min-path compare |
||
18 | b. min-path length compare |
||
19 | c. effective path compare |
||
20 | d. resistance compare with pad (case006) |
||
21 | e. resistance compare with psw (case002) |
||
22 | f. resistance correlation with long via (case00l, 005) |
||
23 | g. resistance check BFFORF IR ANA;rYSIS |
||
24 | |||
25 | ### We need to correlate current, which means: |
||
26 | |||
27 | a. rise/fall charge |
||
28 | b. rise/fall width |
||
29 | c. rise/fall peak |
||
30 | d. FF cells with above |
||
31 | e. rise/fall slope |
||
32 | f. detailed shape polynomial |
||
33 | |||
34 | ### We need to verify capacitance (need investigation) |
||
35 | |||
36 | ### Flow level verify (combined PG algorithm) |
||
37 | |||
38 | a. SPICF level verify |
||
39 | b. node reduction impact |
||
40 | c. by-pass impact |
||
41 | |||
42 | ### VCD correlation method (From Daisy Li) |
||
43 | |||
44 | DVP binary folder□面叩这叩目叩D etc/sh/grep_vcd.sh |
||
45 | 用田D |
||
46 | grep_vcd.sh $vcd_file $nets_file Nets file formatD |
||
47 | instA/instB/instC/pinA instA/instB/instD/netl |
||
48 | |||
49 | ### Cap check |
||
50 | |||
51 | Intrinsic cap can be annotate by below methodology: |
||
52 | |||
53 | set—emir—model [get—cells -hierar -filter "is—memory==true"] -intrinsic—cap l20 |
||
54 | analy江e—power—network -dynamic -vector—file xxx.vcd -start—time xx -end—time xx report—inst—dynamic—irdrop—info > cap—anno.log |
||
55 | show—emir—correlation—result \ |
||
56 | -test—file cap—anno.log \ |
||
57 | -test—index—name inst—name \ |
||
58 | -test—value—name max—ir—drop—among—cycles |
||
59 | -ref—file non—anno.log \ |
||
60 | -ref—index—name inst—name \ |
||
61 | -ref—value—name max—ir—drop—among—cycles |
||
62 | |||
63 | 髻 plot procedure is painful, skip for now |
||
64 | |||
65 | Example max_ir_drop_among_cycles with/without intrinsic cap annotation |
||
66 | Left: ir distribution between annotated IR and un-annotated IR |
||
67 | Right: y axis is annotated IR, x axis is un-annotated IR |
||
68 | |||
69 | ### Current check |
||
70 | In vector based IR analysis, and ref-tool dynamic IR analysis, switching current is generated according toD |
||
71 | |||
72 | switching window (from twf) |
||
73 | load cap (from spef) |
||
74 | clock period (from twf) |
||
75 | slew (from twf) |
||
76 | |||
77 | With above information, tool can generate switching current. |
||
78 | An example for dynamic current replay: |
||
79 | 270 |
||
80 | |||
81 | ## Speed up Dynamic IR Analysis: (From Bob, PiJun, YingShi discussion) |
||
82 | 1.Solver level by-pass (RHS norm check, e.g.) |
||
83 | 2.Step I, V by-pass (Realized by Dr. PiJun Liu, DanPing, 4X speed up) |
||
84 | |||
85 | ``` |
||
86 | 髻 dyn-curr by-pass threshold, default 0.0l mA |
||
87 | set—app—var ir dynamic—sim—abstol 0.0l |
||
88 | 髻 dyn-volt by-pass threshold, default 0.000l V (0.l mV) |
||
89 | set—app—var ir dynamic—sim—vnstol 0.000l |
||
90 | ``` |
||
91 | |||
92 | # The optimal combination of abstol and vnstol (for DTH case) |
||
93 | case003 0-95.45n abstol=1e-4 &vnstol=4e-5 (min_pg_simDmin_pg_tw) results as below: |
||
94 | VP bypassD叩果叩证DDOD度叩叩D且叩影叩DtargetDcorrelation |
||
95 | |||
96 |  |
||
97 | |||
98 | DDDVP bypassD叩能叩D叩升叩DDDDD相叩caseD叩步叩次叩真DtFEDD化叩DDbypass 组叩包叩1e-4/1e-5 D 1-4/4e-5;SC_** 标叩D叩应DSlove Counts□ |
||
99 | |||
100 |  |
||
101 | |||
102 | 3. By-area adjust tolerance |
||
103 | 4. relative and absolute tolerance |
||
104 | 5. Linear fitting between actual solving and by-pass |