ECO and incremental requalification

15.1 Stage purpose

An ECO changes a design that already has its placement, clock tree and routing, and keeps as much of that work as it can. The reason is cost. A late change that restarts placement and routing discards every result that Chapters 1 to 14 qualified, while a small, controlled change keeps most of them. The price is that you must show which results are still true.

The reference describes three ways of making the change. ecoAddRepeater, ecoChangeCell and ecoDeleteRepeater edit one object at a time in the live database. ecoDesign takes a new netlist, restores the saved design and applies the netlist changes. loadECO applies a list of ECO directives to the design that is already in memory, so the baseline must be restored first. ecoPlace and ecoRoute then place the new cells and route the new and modified nets. A separate post-mask flow, used only after masks exist, changes metal and via layers and reuses spare cells. This chapter marks that flow as optional.

A small change touches more than the object you name. The figure below shows one buffer inserted on a routed net. The buffer needs row space, so two filler cells leave. The net is cut, so wires are deleted and new wires and vias are added, and every one of them must be checked again. The amber box is the changed area. The physical difference report, later in the chapter, gives you that box.

The vocabulary of this chapter is short. The baseline is the saved design the ECO starts from. The manifest is your record of what the ECO was meant to do. The difference report is the tool's record of what changed. Incremental requalification means rerunning only the checks that the change can have invalidated, and keeping the evidence of the others.

15.2 Entry prerequisites

15.3 Relevant files and analysis context

An ECO session handles more files than a normal stage, and the order in which they are made matters. The baseline files must exist before the first change, because they cannot be recreated afterwards.

The context that shapes the result is setEcoMode. The reference states that it controls certain behaviours of ecoAddRepeater, ecoChangeCell and ecoDeleteRepeater: whether dont_touch, dont_use and fixed objects are honoured, whether the cell swap must keep the logic equal, whether placement is legalised after each change and whether timing is recomputed. It does not control ecoPlace or ecoRoute. The setting must therefore be read back with getEcoMode and stored with the manifest, because two runs of the same command under different settings can edit different objects.

Timing follows the same rule. The interactive commands recompute timing unless setEcoMode -updateTiming is false, and ecoRoute is timing-driven and SI-driven unless you set otherwise. If -updateTiming is false the commands do not recompute timing, and the reference says that report_timing can recompute it later. Card E-08 starts from extraction, so the post-ECO numbers do not depend on that setting.

15.4 Checks and command cards

The cards follow the order of work. One card comes before the change, three apply it, and four check the result. The apply cards are written in the same form as the checks, because an ECO command that is used without a question and a stop condition tends to leave no evidence.

15.4.1 Pre-stage checks

saveDesign <baseline>.enc
defOut -routing <before>.def
saveNetlist <before>.v
summaryReport -noHtml -outfile <f>
restoreDesign <baseline>.enc.dat <top>

15.4.2 Post-stage checks

The first three cards apply the change. Use the interactive route (E-02) for a small change to named objects, and the netlist route (E-03) when the change is described by a new netlist or an ECO file. E-04 completes either one.

getEcoMode
setEcoMode -honorDontTouch true -honorDontUse true -honorFixedStatus true -LEQCheck true -refinePlace true -updateTiming true
initECO <eco_log_file>
ecoAddRepeater -net <net> -cell <buffer_cell> -loc {<x> <y>}
ecoChangeCell -inst <inst> -cell <new_cell>
ecoDeleteRepeater -inst <inst>
endECO
writeEco -format innovus -output <file>

loadECO -verbose <eco_file>
ecoDesign -noEcoPlace -noEcoRoute -reportFile <f> <session>.enc.dat <top> <new.v>
ecoCompareNetlist -def <ref.def> -outfile <f>

deleteFiller -area {<llx> <lly> <urx> <ury>} -prefix <prefix>
ecoPlace
ecoRoute

checkPlace <report_file>
refinePlace -inst {<inst_list>}
checkFiller -file <f>
addFiller -cell {<filler_cells>} -prefix <prefix> -ecoMode true

verifyConnectivity -type regular -report <f>
verify_drc -limit 0 -report <f>
verify_antenna -report <f>
verify_drc -area {<llx> <lly> <urx> <ury>} -report <f>

defComp <before>.def -defAsGolden -reportFile <f>
summaryReport -noHtml -outfile <f>

timeDesign -postRoute -outDir <dir> -prefix <prefix>
report_annotated_parasitics -list_not_annotated
timeDesign -postRoute -hold -outDir <dir> -prefix <prefix>
check_timing -verbose

15.5 Required reports, artefacts and how to read them

The difference report is the most useful report of the chapter and the least familiar, so here is a short SYNTHETIC example. Its format follows the reference. The names and numbers are invented, and the numbers are database units of an invented DEF, not microns.

VERSION 1                                              ## illustrative excerpt
UNITS 2000
ADDINST u_core/eco_buf_1 BUFX4 120480 100800 N
MOVEINST u_core/i_412 COORD 118080 100800 116880 100800
MOVEINST u_core/i_413 COORD 130080 100800 131280 100800
CHANGEROUTE n1
    DELWIRE METAL2 180 112080 98400 128160 98760
    DELWIRE METAL3 180 128160 98400 128520 104400
    ADDWIRE METAL3 180 112080 98400 112440 101000
ENDCHANGEROUTE
CHANGEROUTE n_ctl_47
    DELWIRE METAL2 180 131040 104400 133200 104760
    ADDWIRE METAL2 180 131040 104400 134400 104760
ENDCHANGEROUTE
CHANGEAREA 111900 98100 134400 104760

Let us read it against the manifest, which lists one buffer on net n1. The ADDINST line is the buffer, and the first CHANGEROUTE block is the cut net, so both are expected. The two MOVEINST lines are not in the manifest. Legalisation moved two neighbours by a small amount to make room. The second CHANGEROUTE block is the real finding: net n_ctl_47 is a different net, and ecoRoute lengthened one of its wires. The excerpt leaves out the ADDNET, ADDPIN and DELPIN lines that cutting n1 produces. The moved cells are a warning until they are listed. Net n_ctl_47 is a hard stop until it is added to the manifest with a reason, a timing look and a DRC look. The CHANGEAREA box now reaches past the buffer, so the area run of verify_drc in card E-06 should use it, extended as card E-07 says.

Keep one short manifest per ECO.

The before and after table is the last piece. Its numbers are SYNTHETIC. Take both columns from reports run with the same views, options and output naming, or the comparison means nothing.

15.6 Healthy, suspicious and hard-stop examples

15.7 Debugging, corrective action and reruns

15.7.1 What to rerun after this change

A change invalidates some of the earlier evidence and leaves the rest alone. The matrix below says which. Rows are the kind of change. Columns are the chapters of this book, numbered 1 to 14: 1 session, 2 import, 3 SDC and MMMC, 4 floorplan, 5 power plan, 6 pre-placement, 7 placement, 8 CTS readiness, 9 CTS, 10 global route, 11 detailed route, 12 post-route optimisation, 13 finishing and 14 extraction.

R means rerun the checks of that stage on the whole design. In the pre-route columns 4 and 6 to 10 it means: confirm on the routed database that the stage's static conditions still hold, such as counts, legality and clock specification. Do not rerun commands that rebuild estimates on a routed design, such as earlyGlobalRoute, optDesign -preCTS or ccopt_design. In column 12, rerun the timing checks, and optDesign -postRoute only to repair what they show. S means spot check: run the stage's checks on the changed objects or the changed area only, for example verify_drc -area, checkFiller -area or verify_antenna -nets, or confirm that the existing report still describes the design. N means not needed: the earlier evidence stays valid, provided the difference report shows that the change did not touch what that stage reads.

The matrix is a recommendation, not tool behaviour, for a routed, flat, single-supply block. A project may tighten it, and should loosen it only with a reason in the manifest. In row E, column 9 is S if no clock pin or clock sink moved. In column 1, the spot check records the release and the setEcoMode settings. Six rules govern its use.

• Choose the row by what the change did. Not by the command used. An ecoRoute that rerouted a neighbouring net is also row G.
• Take the union of rows. When a change fits several rows, each stage takes the strongest letter in its column: R above S above N.
• Send a cell or a wire on a clock net to row H. A swap, a buffer or a route change on the clock network is a clock change, not a data change.
• Restart at Chapter 2 for a library or LEF change. ecoDesign can load replacement LEF files, and a changed abstract can alter every later stage.
• Read column 12 as the gate. Timing is rerun for every row except PG and filler or decap changes. There the spot check covers every view, but only the nets that ecoRoute touched, as defComp lists them. If none, record N.
• Rewrite every deliverable. A change makes the files written before it stale. Write them again through Chapter 16.

15.7.2 Common symptoms

15.7.3 Worked example: a hold fix that looked clean

Suppose one buffer on net n1 fixes a hold violation, as in the figure of Section 15.1. ecoAddRepeater returns the new names, the area run of verify_drc reports nothing, and the hold slack in the motivating view is positive. By the narrow reading, the work is done.

The difference report of card E-07, shown in Section 15.5, says otherwise. Two neighbours moved, and net n_ctl_47 was rerouted. Both belong in the manifest, and they change the reruns. The change is row B, plus row G for the second net. Column 11 is R, so the full-design DRC and connectivity runs are required, not only the area run. Column 12 is R, so setup and hold are rerun in every view, because n_ctl_47 may have a new setup slack. Column 14 is R, so extraction is repeated. Column 13 is S, so filler and antenna are checked in the changed area. The stop decision comes from the union of what changed (rule 2), and the difference report tells you the union.