PLE adjusting in Rtl compiler

Hi,

I am not sure what you are looking for exactly but here are a few things I can think about.

• Adjusting aspect ratio:
  
• set_attribute aspect_ratio /
• or read_def
• Controlling the number of metal layer used
  
• set_attribute number_of_routing_layers

Hi,

In addition to the above attributes, set the scale factors to the same ones used in SOC-Encounter. Two scale factors are used to align SOC-E trial route/default extraction with final route and sign off extraction. These R and C scale factors should be set to the same value in RC.

set_attr scale_of_cap_per_unit_len
set_attr scale_of_res_per_unit_len

Remember that there will be differences between the PLE and FE results. The goal of PLE is to give a better starting point for place and route.

Thanks,
Rich

Thanks for your help. I am going to be using the scaling factors for PLE tuning.

This forum is great!

I did not found any explanation about how RC computes net lenght in PLE mode. Therefore I am not sure if it is better use PLE than  wire load models to drive synthesis, can anyone help me?

Regards,
Cristiano.

Cristiano,
You will not find any explanation of how RC computes the net length as it is a proprietary algorithm. About all I can tell you is that PLE uses a combination of the information in the LEF and capTable along with the structure of the logic to model the nets more accurately than any wireload. You will find that the results after Place and Route will be better using PLE's. It is important to note that the PLE flow is meant to make the post P&R (read REAL) results better. Depending on how conservative your WLM's are, the QoR results post synthesis could be either better OR worse.

Hope this helps.

Regards,

-Jeff-

Hi Cristiano,

I tried to find out myself but, as you may have found out, R&D keeps this recipie pretty tight to their belt - understandably so. The information I was provided is pretty similar to what jflieder mentioned. The is no doubt in my mind that using some of the physical information available early on is the right approach. In addition to jflieder's comment I would like to point out the following:
+ PLEs are dynamic so every time RC picks a different architecture the consequences on net topology are recomputed
This is a key difference with WLM which is static in nature
+ I have seen no noticeable impact on runtime or memory usage
+ PLEs can be more localized since they are not bound by hierarchy and fanout only as I understand this
+ In my experience it is a single iteration approach as opposed to constantly tweaking CWLMs over iterations
+ On the designs I have run, PLEs have produce better or comparable results
One issue I have run into is that it has required a little more dilligence at reviewing all the collateral since now I do not only have to review the .libs but also the physical files. Probably not such a bad thing since we have to review them at some point and better earlier than later.

good luck,
GH-

Hello,

1) How to correlate the timing results between RTL Compiler and pre-layout STA (using ETS or FE-CTE) since it is not possible to reproduce PLE wire prediction ?

2) What is the recommended way to generate SDF file for pre-layout verification (functional simulation) if write_sdf is an unsupported command in RTL_Compiler?

Thanks in advance,
Cristiano.

Hi Cristiano,

how are PLEs treating you these days ? I just downloaded RC71 only to be pleasantly surpsied with fully supported write_sdf. Sounds like you will be as exicted as me :)

GH-

GH, it is really great!

Besides that, below is what Cadence support says about correlating PLE results during STA.

Cristiano.

#####
Cristiano.
RC has the 'write_set_load' command. This command will write out a file
which includes a set_load for every net in the design. The user can then load
the netlist and SDCs into the timing tool, followed by this set_load file.
This *should* enable the timing tool to 'see' the same timing as RC PLE. That
is, it uses the same wire delays as RC in PLE mode. RC will not tell how PLE
does it, but it provides the value that PLE picked.
#####

Hello all,

Unfortunately set_load does not work on nets in CTE (SOCE62).

I found the following help messages in sourceLink but they are conflicting:
http://sourcelink.cadence.com/docs/db/kdb/2007/June/11350058.html
http://sourcelink.cadence.com/docs/db/kdb/2006/June/11249762.html

So, what is the recommended way to correlate PLE results with STA using CTE?

Regards,
Cristiano.