ISPD 2015 Blockage-Aware Detailed Routing-Driven Placement Contest
[Team Rank History]
October 26, 2014:
Contest topic announcement: Please read to the
call for participation document for details about the contest.
November 10, 2014:
Sample benchmark suite and documentation released.
December 11, 2014:
Please note that the Benchmark Suite will be released on December 23rd rather than December 15th. We apologize for this delay.
We will be presenting a 1.5-minute movie about each team that finishes the contest. We are requesting that you direct a film documentary about
your team. This is your opportunity to shine and be stars
:); so take advantage of it :).
December 23, 2014:
Please note that the Benchmark Suite will be released on December 30th rather than December 23rd due to an error with our macro port definitions
that impedes routability. We are currently resolving the issues and will update the website accordingly. If you are interested in receiving a sample design of what is to come, you may
email to firstname.lastname@example.org.. Again, we apologize for this delay.
December 30, 2014:
The contest benchmark suite
has been released.
There are ten designs with varying degress of difficulty.
These designs are derived from the 2014 ISPD Placement Contest Benchmark-A Suite.
Most designs contain fixed macro blocks.
The floor-plan for each design is included in the released benchmark as JPG files.
Different from previous contests, this year we are introducing the notion of fence regions. These are
common constructs typically used to model voltage islands. A cell that is assigned to an exlusive regions must be
placed within this region. Cells cannot be placed in an fence region if they are nonmembers. The fence regions are
rectilinear (e.g. see design mgc_des_perf_b) and can be disjoint (e.g. region r0 in design mgc_edit_dist_a).
Further detailed documentation shall be released by January 11th.
For any question, please send email to email@example.com..
December 31, 2014:
Please note we will be off January 1-3 for the Holidays. We will process your registration requests on January 5th.
January 11, 2015:
Please note that the final benchmark suite document will be released tomorrow. In the meantime, we have updated the benchmark suite.
It is very important that you download this final version of the benchmark suite.
We have fixed some errors and added five more designs.
January 12, 2015:
Please download the
The benchmark suite documentation.
Again, make sure to download the updated benchmark suite.
January 26, 2015:
Please download the updated version RippleDP that is compatible with this year's benchmark suite.
Many thanks to William Chow and Professor Evangeline Young for supplying us with this new version.
Feburary 3, 2015:
Please download the updated version of the
benchmark suite as we have fixed some errors.
Feburary 9, 2015:
We have updated the API for the placer arguments to include the placement.constraints file as an input.
Thanks to Chau-Chin (NTU) for finding some errors in some of the listed design utilization numbers in the conference preprint paper.
We will email you a corrected version of the table.
We are extending the contest deadline by one day. The new date to submit the file executable is March 11, 2015.
Feburary 10, 2015:
Preliminary benchmark design scores are now available on
Feburary 11, 2015:
We have posted an updated version of RippleDP graciosuly provided to us by Wing-Kai Chow at the Chinese University of Hong Kong.
Please download this version as it fixes a memory error.
Feburary 16, 2015:
- There is a problem with some servers and a write_image command. We have suppressed the generation of the image maps to allow your runs to complete.
We will restore the image generation once this issue is resolved.
- The final contest deadline has been extended to March 15, 2015
Feburary 20, 2015:
- There have been several questions regarding the specified
maximum area utilization limits in the placement.constraints file.
Given a design with a fence region and fixed macro blocks, please use
the following recipe to compute the utilization limits:
Core area utilization limit = max (core area utilization, specified limit)
Fence region utilization limit = max (fence region area utilization, specified limit)
Where core area = chip area - macro area - fence region area
We are publishing the area utilization profiles for all the design in
the benchmark suite. Please visit benchmark suite profile.
March 1, 2015:
Many thanks to Win-Kai Chow (Chinese University of Hong Kong) for finding errors in the row definitions
of the floorplans for the mgc_superble designs. We have updated the benchmark designs with appropriate
fixes. Please download the updated benchmark design suite from here.
March 5, 2015:
The contest deadline has been extended to 11:59pm (California PT), March 17, 2015.
The score sheets are updated with the normalized scores without runtimes.
March 8, 2015:
Thank you for your patience with the server issues this weekend. The server is stable now. Please keep submitting your placement files.
We have noticed that a couple of the mgc_superblue* DEF files were submitted for the older versions of these designs. Please make sure you are working on the most recent version of the benchmark from here.
Finally, we are extending the contest deadline to March 18th to compensate you for the lost time.
March 11, 2015:
Many Thanks to William Chow (CUHK) for updating RippleDP's API to support consecutive execution and for fixing two bugs.
Please download the newest version of RippleDP from here.
Detailed routing evaluation metric change: We have observed that there is insufficient dispersion in the detailed routing score. This is particularly the case when DRmedian is high relative to the best DR score, which can happen when only one or two teams have submitted a good placement. This issue is illustrated by the red curve for dividing by DRmedian of 2000 in the example provided here.
To better distinguish those placements with few detailed routing violations, we are changing to a log10 scale, and not using DRmedian in the calculation:
Sdr = 12.475 (log10 (DR + 100) - 2)
Thus, zero DR violations will result in Sdr of zero, and 10,000 DR violations will result in Sdr of 25.
Note that we have added +100 inside the logarithm to moderate the range of scores for small numbers of DR violations. If instead we used say +1,
Sdr = 6.25 log10 (DR + 1) to have the same 0 to 25 range in scores, then 0 DR violations gives a score of zero, and 10 DR violations would give Sdr of 6.5 which is an excessive penalty for a minor increase in the number of detailed routing violations.
Anything above 10,000 DR violations is considered unroutable and will receive the maximum score of 50 overall.
Please refer to the example provided here for illustration.
March 14, 2015:
The contest grid server is back online. We had to kill all jobs submitted on 03/14/2016. Please resubmit your placement files.
The contest deadline is extended to Marth 19th.
March 18, 2015:
We are extending the deadline one day to March 20th 11:59pm (PST) so that your currently submitted jobs complete. Please submit your placer's binary as follows:
teamID.rippledp [if you need us to run rippledp right after running your placer binary]
Finally, please make sure your C/C++ version is 4.6.1 and GLIBC 2_12. Any other versions are not supported and will not be evaluated.
To see the libraries on a machine type the following:
To see the libraries used by an executable type the following:
Wed Mar 18 14:05:26 PDT 2015
, Current time: