PTPX VS. Spectre : about the power analysis

[cmyeda]

Hello everyone, I'm a new guy on this forum. I am a postgraduate in Fudan University in ShangHai. I have encountered some problem about the power analysis, so I regesiter in this forum and look for some help.

My problem is here:

I designed a 2bit pipelined multiplier which is very small and the synthesis result shows it only has 20 standard cells. (smic13HT library , 1.2v supply voltage). Then I written a testbench where the clk period is 10us. And then I got the "saif" file by VCS simulation. Finally I got the power analysis result by PTPX, which shows the total power is 35nW.

To take the power analysis using spectre, I translated the verilog netlist to schematic using the cadence tool in ADE. However the smic13HT library does not provide the schematic cell view of the standard cells. It only provides the symbol cell view of all cells. So I manually draw the schematic cell views of the cells appeared in the netlist base on the CDL file (The CDL file have the spice netlist of all standard cells). Then I written a verilog testbench which is same with the previous one, and save as functional cell view. Finally, I connected the testbench with the multiplier, and started the mix-signal simulation (1.2v supply voltage). The total power of the multiplier was plotted, and average power is 377nW ! Ten times of the PTPX result !!!

Both simulation time are more than 1000 cycles, and the clock periods are same. It was expected that both results should be approximate. Obviously the experiment result was very disappoint. So what's wrong with this contrast experiment?

I hope I have not spelled wrong. Thank you very much for your help!

 

[rca]

1- does the std cell library contains a power info?
2- in Primetime do you used the spef information to include the parasitic?
3- do you check the waveform seen by primetime?
4- could you separate the leakage to the dynamic power in both runs?

 

[cmyeda]

Hello, rca. Thanks for your reply.

1, Yes , the std cell library does have a power info.
2, No, The tools I used are DC , VCS , and PTPX. DC generates sdf and netlist file. VCS takes the gate-level simulation (sdf is annotated) and generates the vcd or saif file. Finally PTPX calculates the power using netlist file, sdf file and saif (or vcd) file.
3, The waveforms in PTPX and Spectre are showed in the following picture:

Zoom in:

The peak power in PTPX is about 12mW , but in Spectre is about 450uW. While the average power in PTPX is 35nW, but in Spectre is 370nW.
4, In PTPX,
Internal power: 28.6nW
Switching power: 5.8nW
Leakage power: 1.1nW total power is about 35.5nW

In Spectre I don't know how to calculate the dynamic and leakage power separately. The waveform tells that the leakage power may be 1nW - 4nW.
Total average power is about 370nW.

 

[rca]

are you able to add parasitic info to Primetime?

 

[cmyeda]

I simulated the INHD1XHT cell in TT corner, and found that the static power is about 38pW. While in the library file (smic13HT_tt.lib) , the value is about 20pW.

I think the multiplier I built in spectre is different with the one I built by DC tool. But I can not say what the difference is.