BeginnerQuestion 184 of 187Source: Synopsys ICC2 Implementation User Guide X-2025.06-SP2: Chip Finishing and Design for Manufacturing; Synopsys IC Validator User Guide S-2021.06-SP2: Output Files

What goes into a tapeout handoff?

From PDVerse PnR Interview Handbook · pdVerse Mentor Guide

By Tabish Iqbal · 9+ Years VLSI Physical Design Experience

Short Answer

A tapeout handoff is the final layout plus the evidence that it is clean: a merged GDSII or OASIS stream with every cell, macro and fill shape, the final DRC, LVS and ERC results with an approved waiver list, IR and EM signoff reports, and whatever constraints or documents the foundry or chip owner requires. The stream must be the exact data that was verified, so it is written once from the final block and every signoff run points at that file. Any edit afterwards means rerunning the checks.

A flow from the final ICC2 block to a merged GDSII or OASIS stream, then standalone DRC, LVS and ERC runs on that stream, IR and EM reports and a waiver list, all collected into one handoff package with version stamps. — Technical Reference DiagramWhat goes into a tapeout handoff?

Technical Explanation

The stream has to be complete. ICC2 often works with frame or abstract views of cells and macros, so their full layouts must be merged in: write_gds -merge_files (ICC2) does it on output, and signoff runs inside ICC2 take the same data through signoff.physical.merge_stream_files (ICC2).
Layer mapping must match the foundry's layer and datatype numbers, so the layer map file is part of the handoff. Fill is controlled at stream-out with -fill include, exclude or fill_only, and for cut metal, write_gds -connect_below_cut_metal (ICC2) generates the metal below the cut shapes; it is off by default.
DRC and LVS evidence comes from the standalone icv (ICV) runs on that stream. cell.RESULTS (ICV) opens with 'RESULTS: CLEAN' for DRC, or 'LVS Compare Results: PASS' followed by 'DRC and Extraction Results: CLEAN' for LVS. 'RESULTS: RUN ABORTED' is not a result.
Remaining violations need a waiver list the foundry has accepted, with the rule, location and reason for each. Unexplained violations are not waivers.
Electrical evidence: IR drop and EM reports against the agreed budgets, antenna and density results, and ERC, ESD and latch-up results where the flow owns them.
Version stamps tie it together: runset revision, tool versions, top cell name and the stream file's date or checksum, so anyone can show that the reports belong to the file that was sent.
Masks for advanced nodes are very expensive and a respin costs months, so the package is reviewed as carefully as the design itself.

What To Check

Stream top cell name and that no cell is left as an empty reference.
DRC and LVS headers in cell.RESULTS (ICV) from runs on the exact file being handed off.
Waiver list complete and approved.
IR, EM, antenna and density reports newer than the last layout edit.
Runset and tool versions recorded in the package.

Command Checks & Actions

ICC2 (icc2_shell)write_gds -layer_map layer.map -merge_files {std.gds macro.gds} -fill include top.gds

Writes the final merged stream with foundry layer numbers and fill included.

ICC2 (icc2_shell)write_oasis -layer_map layer.map -merge_files {std.gds macro.gds} top.oas

Writes the same data as OASIS when the foundry prefers the smaller format.

ICC2 (icc2_shell)signoff_check_drc

Last in-design DRC on the final block before stream-out.

IC Validator (icv)icv -i top.oas -f OASIS -c top drc_runset.rs

Standalone signoff DRC on top.oas, the exact stream in the package.

IC Validator (icv)cell.RESULTS

First file to read: the header must say RESULTS: CLEAN, or LVS Compare Results: PASS for LVS.

Healthy, Suspicious & Hard-stop Results

Healthy (illustrative): DRC 'RESULTS: CLEAN' apart from 4 approved waivers, LVS PASS, IR worst 36 mV against 40 mV, EM worst 88%, every report newer than the stream.
Suspicious (illustrative): DRC run on a stream written two days before the last ECO, or waivers without a named approver.
Hard stop: LVS FAIL, DRC 'NOT CLEAN' with unwaived violations, a run that aborted, or macros missing from the stream.

Common Mistake

The Trap: Writing the final GDS without merging the macro layouts. The stream holds only references to the SRAM cells, the in-design checks passed because ICC2 used its own views, and the foundry's intake reports empty cells. The package is rejected, the stream is rewritten and every standalone run repeated, costing days in the tapeout week.

What The Interviewer Is Testing

Knowing that the verified file and the shipped file must be the same file.
Reading cell.RESULTS (ICV) headers correctly.
Treating waivers as documented, approved items, not silent exceptions.

Follow-up Question & Model Response

"Why run final DRC and LVS on the streamed file instead of trusting the ICC2 database?"

Candidate Model Response: The stream is what the foundry builds from, and it goes through steps the database does not: layer mapping, merging of cell and macro layouts, and fill handling. A wrong layer map or a missing merge file breaks the stream while the database stays clean. Running icv (ICV) on the stream proves the file itself is correct. It also gives reports that point at the same file the foundry receives.

Practical Example

Tapeout Scenario: (illustrative) The package holds top.oas, the layer map, cell.RESULTS (ICV) for DRC with 'RESULTS: CLEAN' and 4 approved waivers, cell.RESULTS (ICV) for LVS with 'LVS Compare Results: PASS', IR reports showing 36 mV worst against a 40 mV budget, EM at 88% of the worst limit, antenna and density clean, and a manifest with the runset revision and file checksum. A final ECO for one hold fix forces a new stream, and every DRC and LVS run is repeated on it before the manifest is updated.

PnR Flow Mentor Guide

Master the Physical Design Implementation Flow

Read the complete 8-chapter PnR Flow Mentor Guide free on the web — library setup through placement, clock tree synthesis, routing, chip finishing, hierarchical implementation, and ECO, all the way to stream-out.

Explore PnR Flow Mentor Guide →