Design-for-ECO — spare-cell / spare-gate / spare-pad readiness
A taped-out chip will, sooner or later, need a fix: a missed timing path, a
logic bug found in bring-up, a metal-mask trim. The cheapest possible fix is a
metal-only ECO — reusing already-placed transistors and only re-spinning a
few metal/via masks. The most expensive is a base-layer respin — new
diffusion/poly masks, a full mask set, weeks of fab time. Design-for-ECO is the
discipline of pre-investing a small amount of silicon area, BEFORE tape-out,
so that future fixes stay in the cheap (metal-only) regime.
This skill is chip-AGNOSTIC: it describes the methodology and the hard
preservation rule; the deterministic insertion + verification is owned by the
phase3 stage3 Design-for-ECO step and its two checkers (see Flow integration).
Why pre-place spare cells/gates + reserve spare pads
| Fix vehicle |
Masks touched |
Relative cost / turn time |
Needs spares? |
| Metal-only ECO |
a few metal + via |
low — days, partial mask set |
yes — must reuse existing transistors |
| Base-layer ECO |
poly/diffusion + all above |
high — weeks, full mask set |
no (but expensive) |
| Full respin |
entire mask set + re-PnR |
highest |
no |
A metal-only ECO can only add/replace logic if the transistors it needs are
already on the die (placed, powered, tied off) and only need re-wiring.
If no spare cells exist near the bug, even a one-gate fix forces a base-layer
respin. The same logic applies at the chip boundary: a missing test/trim/config
signal after tape-out needs a reserved spare pad (and its ESD + bond
option) already present — you cannot add a pad in metal only.
So Design-for-ECO trades ~1-5% area + a handful of pads now for the option to
do almost any late fix as a metal-only spin later. On most projects that option
is exercised at least once and pays for itself many times over.
What to insert
Spare standard cells / gates (the ECO pool)
Insert a mix of cell types so an arbitrary future fix can be assembled from
nearby spares without re-placing:
- Inverter (
inv) — buffering, polarity, hold padding.
- NAND2 / NOR2 — universal 2-input logic; any combinational function is
reachable by composing these.
- AOI / OAI (and-or-invert / or-and-invert) — compact compound gates for
denser logic fixes.
- MUX2 — select/bypass/observe insertion.
- DFF — sequential fixes (re-time, add a pipeline stage, sample a signal).
Guidance:
- Density target ~1-5% of placed std-cell area (start ~2-3%; raise to 5% for
risky/first-silicon designs, lower toward 1% for mature, low-risk blocks).
- Type ratio roughly: many inverters + NAND2/NOR2 (the workhorses), a useful
count of MUX2 and DFF, a smaller set of AOI/OAI. The pool must be able to
build both combinational and sequential fixes.
- Tie-offs (mandatory): every spare input is tied to a defined level
(VSS via tie-low / VDD via tie-high, or a tie cell), every spare output is
left floating-safe (unconnected or tied per PDK rule). An untied spare
input floats → leakage, latch-up risk, and an LVS/ERC failure. A spare that
is not properly tied off is not a usable spare.
- dont_touch / keep: every spare cell/gate is marked
dont_touch (a.k.a.
keep) so synthesis/PnR optimization will not delete, resize, absorb, or
re-purpose it. This attribute is the contract that the rest of the flow obeys
(see The HARD preservation rule).
Reserved spare pads
Reserve a small number of spare I/O pads (with ESD cells, optional bond
finger, and a tie-off / weak pull) at the chip boundary, marked
dont_touch/keep. These enable post-tape-out access to a new test/trim/
config signal as a metal-only + bond-option change rather than a respin.
Distribution (not a corner dump)
Spares must be spatially distributed across the placeable area — a roughly
uniform grid / per-region quota — so that every region of the die has spares
within ECO routing reach. A bug in a far corner cannot be fixed metal-only by a
spare that sits 500µm away. The coverage checker verifies per-region density
and distribution, not just a global count.
Where in the flow
placement (legalized) ─▶ ★ Step 18: Design-for-ECO insertion ★ ─▶ CTS ─▶ route ─▶ fill ─▶ signoff
(distributed spare pool +
tie-offs + reserved pads,
all dont_touch/keep)
In the canonical 56-step flow this is Step 18 (between placement, Step 17, and
CTS, Step 19); all later numeric steps shift +1 relative to the old 55-step flow.
- After placement, before CTS. Spares need legal placement sites and must be
in the database before CTS/route so the clock tree and routing account for
them. Inserting earlier risks them being optimized away by placement opt;
inserting after route defeats the "already placed/powered" purpose.
- ECO-aware metal fill. When metal fill is added at signoff, it must be
ECO-aware: do not lock fill over the metal tracks above spare cells and
reserved pads, or use slottable/removable fill there, so a future metal-only
ECO can route to the spares without ripping up locked fill. Treat the spare
pool's routing channels as reserved.
The HARD preservation rule
Compare the native netlist or implementation database before and after optimization. Preserve every unused ECO spare, its keep attributes and the reserved pads; an unintentionally removed spare must be restored.
Every optimization skill (synth-doctor, rtl-repair, hold-fix, eco-plan,
ppa-predict, drc-fix) carries a tailored copy of this block.
Flow integration
The deterministic mechanics are owned by the phase3 stage3 Design-for-ECO
step (in the flow runner), not by this skill:
- The stage3 step inserts the distributed, tied-off spare std-cell pool
(inverter / nand2 / nor2 / dff / mux2 / aoi / oai) + reserved ECO pads, all
marked
dont_touch/keep, and emits:
phase3/stage3/pnr/spare_cells.json — the inserted spare inventory, and
the ONLY artefact the stage3 step emits for the readiness question. It
carries the measurements (count, placed_cells_est, target_density,
actual_density, tied_off, tie_off, per-instance coordinates); it
carries no verdict, because the step that inserts the spares does not
grade its own insertion.
spare_cell_coverage_check.py (readiness) verifies the pool meets the
density / distribution / tie-off targets → coverage PASS, and is the
DECLARING PRODUCER of reports/spare_cell_coverage.json: it is the only
writer of that path, and it does not read it back.
- Compare the native netlist or implementation database before and after optimization. Preserve every unused ECO spare, its keep attributes and the reserved pads; an unintentionally removed spare must be restored.
Acceptance (Design-for-ECO readiness gate)
A digital place-and-route IC is Design-for-ECO-ready iff:
reports/spare_cell_coverage.json → readiness PASS (density in ~1-5%
band, distributed, all spares tied off), and
reports/spare_preservation.json → all_keep_attr_intact == true and
removed == 0.
benchmark-verify reports this as its sixth pillar ("Design-for-ECO readiness").
It is N/A only for an IC that genuinely has no place-and-route (analog-only,
no DEF/GDS). A missing spare report is PENDING, never a silent PASS.
Do not
- Do not dump all spares in one corner — distribution is the point.
- Do not leave spare inputs floating — untied spares fail ERC/LVS and are unusable.
- Do not let metal fill lock the tracks over spares/pads — keep ECO routing reachable.
- Do not allow any downstream optimization to remove a keep-marked spare.
1---2name: design-for-eco3description: "Design-for-ECO methodology — pre-place a distributed pool of tied-off spare standard cells/gates and reserve spare ECO pads so that a late-stage bug can be fixed with a cheap metal-only ECO instead of a full base-layer respin. Covers WHY (metal-only vs base-layer respin cost), WHAT to insert (inverter/nand2/nor2/dff/mux2/aoi/oai mix, ~1-5% density, tie-offs, even spatial distribution, reserved ECO pads), WHERE in the flow (after placement, before CTS), ECO-aware metal fill, and the HARD preservation rule that every downstream optimization must obey. Use when: 'design for eco', 'spare cells', 'ECO prep', 'metal-only ECO readiness', 'reserve spare gates', 'spare pad planning', or before CTS on any digital place-and-route flow."4metadata:5 source: https://skillmd.com/skills/vibeic/design-for-eco6---78# Design-for-ECO — spare-cell / spare-gate / spare-pad readiness910A taped-out chip will, sooner or later, need a fix: a missed timing path, a11logic bug found in bring-up, a metal-mask trim. The cheapest possible fix is a12**metal-only ECO** — reusing already-placed transistors and only re-spinning a13few metal/via masks. The most expensive is a **base-layer respin** — new14diffusion/poly masks, a full mask set, weeks of fab time. Design-for-ECO is the15discipline of *pre-investing* a small amount of silicon area, BEFORE tape-out,16so that future fixes stay in the cheap (metal-only) regime.1718This skill is **chip-AGNOSTIC**: it describes the methodology and the hard19preservation rule; the deterministic insertion + verification is owned by the20phase3 stage3 Design-for-ECO step and its two checkers (see *Flow integration*).2122## Why pre-place spare cells/gates + reserve spare pads2324| Fix vehicle | Masks touched | Relative cost / turn time | Needs spares? |25|---|---|---|---|26| Metal-only ECO | a few metal + via | low — days, partial mask set | **yes** — must reuse existing transistors |27| Base-layer ECO | poly/diffusion + all above | high — weeks, full mask set | no (but expensive) |28| Full respin | entire mask set + re-PnR | highest | no |2930A metal-only ECO can only add/replace logic if the transistors it needs are31**already on the die** (placed, powered, tied off) and only need re-wiring.32If no spare cells exist near the bug, even a one-gate fix forces a base-layer33respin. The same logic applies at the chip boundary: a missing test/trim/config34signal after tape-out needs a **reserved spare pad** (and its ESD + bond35option) already present — you cannot add a pad in metal only.3637So Design-for-ECO trades ~1-5% area + a handful of pads now for the option to38do almost any late fix as a metal-only spin later. On most projects that option39is exercised at least once and pays for itself many times over.4041## What to insert4243### Spare standard cells / gates (the ECO pool)44Insert a *mix* of cell types so an arbitrary future fix can be assembled from45nearby spares without re-placing:4647- **Inverter** (`inv`) — buffering, polarity, hold padding.48- **NAND2 / NOR2** — universal 2-input logic; any combinational function is49 reachable by composing these.50- **AOI / OAI** (and-or-invert / or-and-invert) — compact compound gates for51 denser logic fixes.52- **MUX2** — select/bypass/observe insertion.53- **DFF** — sequential fixes (re-time, add a pipeline stage, sample a signal).5455Guidance:56- **Density target ~1-5%** of placed std-cell area (start ~2-3%; raise to 5% for57 risky/first-silicon designs, lower toward 1% for mature, low-risk blocks).58- **Type ratio** roughly: many inverters + NAND2/NOR2 (the workhorses), a useful59 count of MUX2 and DFF, a smaller set of AOI/OAI. The pool must be able to60 build both combinational and sequential fixes.61- **Tie-offs (mandatory):** every spare input is tied to a defined level62 (VSS via tie-low / VDD via tie-high, or a tie cell), every spare output is63 left floating-safe (unconnected or tied per PDK rule). An *untied* spare64 input floats → leakage, latch-up risk, and an LVS/ERC failure. A spare that65 is not properly tied off is **not** a usable spare.66- **dont_touch / keep:** every spare cell/gate is marked `dont_touch` (a.k.a.67 `keep`) so synthesis/PnR optimization will not delete, resize, absorb, or68 re-purpose it. This attribute is the contract that the rest of the flow obeys69 (see *The HARD preservation rule*).7071### Reserved spare pads72Reserve a small number of **spare I/O pads** (with ESD cells, optional bond73finger, and a tie-off / weak pull) at the chip boundary, marked74`dont_touch`/`keep`. These enable post-tape-out access to a new test/trim/75config signal as a metal-only + bond-option change rather than a respin.7677### Distribution (not a corner dump)78Spares must be **spatially distributed** across the placeable area — a roughly79uniform grid / per-region quota — so that every region of the die has spares80within ECO routing reach. A bug in a far corner cannot be fixed metal-only by a81spare that sits 500µm away. The coverage checker verifies per-region density82and distribution, not just a global count.8384## Where in the flow8586```87placement (legalized) ─▶ ★ Step 18: Design-for-ECO insertion ★ ─▶ CTS ─▶ route ─▶ fill ─▶ signoff88 (distributed spare pool +89 tie-offs + reserved pads,90 all dont_touch/keep)91```9293In the canonical 56-step flow this is **Step 18** (between placement, Step 17, and94CTS, Step 19); all later numeric steps shift +1 relative to the old 55-step flow.9596- **After placement, before CTS.** Spares need legal placement sites and must be97 in the database before CTS/route so the clock tree and routing account for98 them. Inserting earlier risks them being optimized away by placement opt;99 inserting after route defeats the "already placed/powered" purpose.100- **ECO-aware metal fill.** When metal fill is added at signoff, it must be101 *ECO-aware*: do not lock fill over the metal tracks above spare cells and102 reserved pads, or use slottable/removable fill there, so a future metal-only103 ECO can route to the spares without ripping up locked fill. Treat the spare104 pool's routing channels as reserved.105106## The HARD preservation rule107108> Compare the native netlist or implementation database before and after optimization. Preserve every unused ECO spare, its keep attributes and the reserved pads; an unintentionally removed spare must be restored.109110Every optimization skill (`synth-doctor`, `rtl-repair`, `hold-fix`, `eco-plan`,111`ppa-predict`, `drc-fix`) carries a tailored copy of this block.112113## Flow integration114115The deterministic mechanics are owned by the phase3 **stage3 Design-for-ECO**116step (in the flow runner), not by this skill:117118- The stage3 step **inserts** the distributed, tied-off spare std-cell pool119 (inverter / nand2 / nor2 / dff / mux2 / aoi / oai) + reserved ECO pads, all120 marked `dont_touch`/`keep`, and emits:121 - `phase3/stage3/pnr/spare_cells.json` — the inserted spare inventory, and122 the ONLY artefact the stage3 step emits for the readiness question. It123 carries the measurements (`count`, `placed_cells_est`, `target_density`,124 `actual_density`, `tied_off`, `tie_off`, per-instance coordinates); it125 carries no verdict, because the step that inserts the spares does not126 grade its own insertion.127- `spare_cell_coverage_check.py` (**readiness**) verifies the pool meets the128 density / distribution / tie-off targets → coverage PASS, and is the129 DECLARING PRODUCER of `reports/spare_cell_coverage.json`: it is the only130 writer of that path, and it does not read it back.131- Compare the native netlist or implementation database before and after optimization. Preserve every unused ECO spare, its keep attributes and the reserved pads; an unintentionally removed spare must be restored.132133## Acceptance (Design-for-ECO readiness gate)134135A digital place-and-route IC is Design-for-ECO-ready iff:136- `reports/spare_cell_coverage.json` → readiness **PASS** (density in ~1-5%137 band, distributed, all spares tied off), **and**138- `reports/spare_preservation.json` → `all_keep_attr_intact == true` **and**139 `removed == 0`.140141`benchmark-verify` reports this as its sixth pillar ("Design-for-ECO readiness").142It is **N/A** only for an IC that genuinely has no place-and-route (analog-only,143no DEF/GDS). A missing spare report is **PENDING**, never a silent PASS.144145## Do not146- Do not dump all spares in one corner — distribution is the point.147- Do not leave spare inputs floating — untied spares fail ERC/LVS and are unusable.148- Do not let metal fill lock the tracks over spares/pads — keep ECO routing reachable.149- Do not allow any downstream optimization to remove a keep-marked spare.