7_kakol
TRANSCRIPT
-
7/25/2019 7_kakol
1/16
iDFM Flow: An ECO Implementation of
Metal, Via Filling
Giriraj Kakol, Dibyendu Goswami, Rajesh Karturi, Suryanarayana
Prekke, Intel Corporation, Bangalore, India
-
7/25/2019 7_kakol
2/16
2
Motivation
As the complexity of designs and process design rules are growing
with every generation, getting a correct by construction DFM fill isbecoming increasingly difficult
PDEs (Physical Design Engineers) end up spending huge
manual effort in cleaning-up DRC and density violations
All the manual effort is lost when there is a design ECO that makes
previous fill polygons obsolete as it could potentially create DRCs,
shorts on the ECO-implemented database.
Running DFM flows from scratch is not viable as this would
require spending the same amount of manual effort as before
to clean up the DRCs
Need automation to improve on ECO Layout productivity
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
3/16
3
Prior Art
Manual Effort of cleaning the DRCs, Shorts
Spends nearly two days on a typical ECO per partition
Tedious and Error Prone
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
4/16
4
iDFM Filling
Incremental Fill Flow relies on re-using the old fill data as much as
possible
No re-filling
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
5/16
5
Fill
Legalization Detection and removal of shorts/DRCs
DRCs are detected through foundry runset
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
6/16
6
Cleanup of Shorts Short is detected if a metX fill polygon overlaps or touches a metX
design (route) polygon The portion of the fill polygon that overlaps the design polygon is
removed
In case of metX fill touching the route polygon, it is shrunk along
the preferred routing direction by 1 manufacturing grid so that
touch can be avoided.
Any additional DRCs created at this stage are handled in later
stages of DRC correction
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
7/167
iDFM Flow Steps
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
Identify density failing windows (Metals/Vias)
on trimmed database
incrementally re-fill without creating
additional shorts.
Fill flow doesnt produce clean fill db as it
cant comprehend all complicated DRCs at
UDSM process nodes.
Fill trimming cleans shorts, DRCs but refilling
introduces more DRCs which are addressed
incrementally
Certain cases need addition of small stubs to
fill/cell data depending on the design rule.
TAT (Turn Around time) of iDFM is 6X faster
than rerunning fill from scratch
Trim Done
Density
Error?
Incremental
Filling
DRC analysis
and Fixing
Clean Fill DB
-
7/25/2019 7_kakol
8/168
iDFM Advantages
Short clean fill: Removes the Original FILL polygons overlapping on
ECO polygons for avoiding shorts.
Shorter runtime: Incrementally FILLs Metals and Vias based on
density and DRC requirements.
DRC clean Filling: Set of algorithms to handle individual DRCs.
Addresses Via density issues: Incrementally generates FILL Vias by
connecting only the floating metal fills; without causing any shorts.
Parallelize filling: Option to independently FILL set of Metal and
Vias.
Region Filling: BBOX option to correct only in the specified window.
Easily portable: Easy portability to next process nodes.iDFM Flow: An ECO Implementation of Metal, Via FillingISPD 2013
-
7/25/2019 7_kakol
9/169
Results
Proved solution for ECO implementation on all in-house projects
99% of the DRCs introduced by the fill are being removed
Shorts clean ECO DB with metal fill
Very few density errors
6X improvement in TAT (Turn Around time) for ECOimplementation.
8 person weeks of manual effort reduction on one of the internal
CPU projects.
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
10/16
10
DRC Results
Addressing DRC requirements through iDFM:
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
0
50
100
150
200
250
300
350
400
450
500
Metal1 Metal2 Metal3 Metal4 Metal5 Metal6 Metal7 Metal8
Pre iDFM DRCs
Post iDFM DRCs
-
7/25/2019 7_kakol
11/16
11
Density Results
Addressing density requirements through iDFM:
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
0
10
20
30
40
50
60
70
80
Metal1 Metal2 Metal3 Metal4 Metal5 Metal6 Metal7 Metal8
Pre iDFM Density failing windows
Post iDFM density failing windows
-
7/25/2019 7_kakol
12/16
12
Conclusion
Automated flow that replaces the tedious and error-prone manual
approach of correcting the DRCs, shorts on ECO Database.
Generic approach : Independent of process node.
Judicious re-use of Foundry runsets (rules deck)
Applied special treatment to the errors that are reported by the
runsets to enable Metal/Via DRC and density errors.
Bounding box and layer specific iDFM to improve throughput
Incremental Refill on density failing windows
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
13/16
13
Back up slides for poster
presentation
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
14/16
14
Original FILL and CELL:
FINAL FILL and CELL after incrementally refilling to correct
DRC/density:
Incremental Refill:
Fill Polygon
Signal
Polygon
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
15/16
15
Addressing complex DRVs Addition of stubs to Signal/FILL nets for cleaning DRCs.
Patterning Signal/FILL nets for clean DRCs.
iDFM Flow: An ECO Implementation of Metal, Via Filling
ISPD 2013
-
7/25/2019 7_kakol
16/16
16iDFM Flow: An ECO Implementation of Metal, Via FillingISPD 2013