From arihant at iith.ac.in Fri Feb 17 11:10:39 2012 From: arihant at iith.ac.in (Arihant Jain) Date: Fri, 17 Feb 2012 11:10:39 -0800 Subject: [Hotspot] Reg:Modelling TSV in Hotspot(Help) Message-ID: Hello, I want to model TSV in Hotspot in order to study the effect TSV on the Thermal Profile of the 3D IC.How can I do this? Regards Arihant Jain Class & Hostel Representative 4th Year B.Tech Student(Honors) Electrical Engineering Department Indian Institute of Technology,Hyderabad Ph.No:+919441919112 Email ID: arihant at iith.ac.in -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120217/6749cb5f/attachment-0001.html From Ivaylo.Avramov at uk.fujitsu.com Wed Feb 22 05:40:56 2012 From: Ivaylo.Avramov at uk.fujitsu.com (Avramov, Ivaylo) Date: Wed, 22 Feb 2012 13:40:56 -0000 Subject: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices In-Reply-To: References: Message-ID: Hi, I am modeling a large device with very high power density in certain regions(200mW dissipated in 200um x 2um), causing concern for hotspot issues and reliability. A HotSpot simulation using a very detailed model (80 blocks in an analogue macro) shows a maximum hotspot gradient of 5.17 degC. Using a very basic pyramid model shows that the gradient should be about 15 degC. Using the same detailed model in Mentor FloTHERM gives a gradient of 17 degC. Can you think of any reasons for such a large discrepancy between the results? The substrate is 0.5mm with many circuit block of roughly 150um x 25um. Does the vertical grid get affected by the definitions for horizontal grid or is it a fixed number of effective layers? It seems like blocks with significantly lower power density have correct temperature gradient and this issue is more evident for blocks with very high power density. Regards, Ivaylo. ----------------------------------------------------------------- IVAYLO AVRAMOV Design Engineer Fujitsu Semiconductor Europe GmbH 3 Concorde Park, Concorde Road SL6 4FJ Maidenhead, Berkshire, United Kingdom Email : ivaylo.avramov at uk.fujitsu.com Tel : +44 (0) 1628 50 4737 Fax : +44 (0) 1628 50 4777 www : http://emea.fujitsu.com/semiconductor ----------------------------------------------------------------- ________________________________________ Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: Offenbach/Main HRB 32725 ________________________________________ This e-mail and any attachment contains information which is private and confidential and is intended for the addressee only. If you are not an addressee, you are not authorised to read, copy or use the e-mail or any attachment. If you have received this e-mail in error, please notify the sender by return e-mail and then delete it. ________________________________________ From wh6p at virginia.edu Wed Feb 22 07:46:43 2012 From: wh6p at virginia.edu (Wei Huang) Date: Wed, 22 Feb 2012 09:46:43 -0600 Subject: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices In-Reply-To: References: Message-ID: Hi Ivaylo, It is very interesting to see how HotSpot applies to large devices! Assuming the device does have a heat spreader and a heat sink, one possible reason is the thermal properties of the thermal paste (ie. thermal interface material), which could affect silicon spatial temperature gradient. If you are using the same materials and dimensions in both HotSpot and FloTherm, then it comes to your reasoning about the mismatch of grid thickness and grid lateral size, which makes complete sense. I would suggest chop the silicon into multiple layers using the multi-layer feature in HotSpot, and see if that improves the results. Of course, this makes HotSpot more similar to a conventional FEM/FDM simulator like FloTherm. Hope this helps. -Wei On Wed, Feb 22, 2012 at 7:40 AM, Avramov, Ivaylo < Ivaylo.Avramov at uk.fujitsu.com> wrote: > Hi, > > > I am modeling a large device with very high power density in certain > regions(200mW dissipated in 200um x 2um), causing concern for hotspot > issues and reliability. A HotSpot simulation using a very detailed model > (80 blocks in an analogue macro) shows a maximum hotspot gradient of > 5.17 degC. Using a very basic pyramid model shows that the gradient > should be about 15 degC. Using the same detailed model in Mentor > FloTHERM gives a gradient of 17 degC. > > Can you think of any reasons for such a large discrepancy between the > results? The substrate is 0.5mm with many circuit block of roughly 150um > x 25um. Does the vertical grid get affected by the definitions for > horizontal grid or is it a fixed number of effective layers? > > It seems like blocks with significantly lower power density have correct > temperature gradient and this issue is more evident for blocks with very > high power density. > > > Regards, > Ivaylo. > > ----------------------------------------------------------------- > IVAYLO AVRAMOV > Design Engineer > Fujitsu Semiconductor Europe GmbH > 3 Concorde Park, Concorde Road > SL6 4FJ Maidenhead, Berkshire, United Kingdom > Email : ivaylo.avramov at uk.fujitsu.com > Tel : +44 (0) 1628 50 4737 > Fax : +44 (0) 1628 50 4777 > > www : http://emea.fujitsu.com/semiconductor > ----------------------------------------------------------------- > > > ________________________________________ > Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc > Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida > Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: > Offenbach/Main HRB 32725 > ________________________________________ > This e-mail and any attachment contains information > which is private and confidential and is intended > for the addressee only. If you are not an addressee, > you are not authorised to read, copy or use the e-mail > or any attachment. If you have received this e-mail > in error, please notify the sender by return e-mail > and then delete it. > ________________________________________ > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120222/e56438c4/attachment.html From Ivaylo.Avramov at uk.fujitsu.com Wed Feb 22 09:57:27 2012 From: Ivaylo.Avramov at uk.fujitsu.com (Avramov, Ivaylo) Date: Wed, 22 Feb 2012 17:57:27 -0000 Subject: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices In-Reply-To: References: Message-ID: Hi Wei, Thank you very much for the advice. The models are identical so I tried your suggestion. Going up from 1 to 6 to 11 to 22 layers increased the accuracy at every step and it's a necessity for devices with high-power-density circuits. Regards, Ivaylo. IVAYLO AVRAMOV Design Engineer Fujitsu Semiconductor Europe GmbH 3 Concorde Park, Concorde Road SL6 4FJ Maidenhead, Berkshire, United Kingdom Phone: +44 (0) 1628 50 4737 Fax: +44 (0) 1628 50 4777 Mailto: ivaylo.avramov at uk.fujitsu.com Internet: http://emea.fujitsu.com/semiconductor From: Wei Huang [mailto:wh6p at virginia.edu] Sent: 22 February 2012 15:47 To: Avramov, Ivaylo Cc: hotspot at mail.cs.virginia.edu Subject: Re: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices Hi Ivaylo, It is very interesting to see how HotSpot applies to large devices! Assuming the device does have a heat spreader and a heat sink, one possible reason is the thermal properties of the thermal paste (ie. thermal interface material), which could affect silicon spatial temperature gradient. If you are using the same materials and dimensions in both HotSpot and FloTherm, then it comes to your reasoning about the mismatch of grid thickness and grid lateral size, which makes complete sense. I would suggest chop the silicon into multiple layers using the multi-layer feature in HotSpot, and see if that improves the results. Of course, this makes HotSpot more similar to a conventional FEM/FDM simulator like FloTherm. Hope this helps. -Wei On Wed, Feb 22, 2012 at 7:40 AM, Avramov, Ivaylo wrote: Hi, I am modeling a large device with very high power density in certain regions(200mW dissipated in 200um x 2um), causing concern for hotspot issues and reliability. A HotSpot simulation using a very detailed model (80 blocks in an analogue macro) shows a maximum hotspot gradient of 5.17 degC. Using a very basic pyramid model shows that the gradient should be about 15 degC. Using the same detailed model in Mentor FloTHERM gives a gradient of 17 degC. Can you think of any reasons for such a large discrepancy between the results? The substrate is 0.5mm with many circuit block of roughly 150um x 25um. Does the vertical grid get affected by the definitions for horizontal grid or is it a fixed number of effective layers? It seems like blocks with significantly lower power density have correct temperature gradient and this issue is more evident for blocks with very high power density. Regards, Ivaylo. ----------------------------------------------------------------- IVAYLO AVRAMOV Design Engineer Fujitsu Semiconductor Europe GmbH 3 Concorde Park, Concorde Road SL6 4FJ Maidenhead, Berkshire, United Kingdom Email : ivaylo.avramov at uk.fujitsu.com Tel : +44 (0) 1628 50 4737 Fax : +44 (0) 1628 50 4777 www : http://emea.fujitsu.com/semiconductor ----------------------------------------------------------------- ________________________________________ Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: Offenbach/Main HRB 32725 ________________________________________ This e-mail and any attachment contains information which is private and confidential and is intended for the addressee only. If you are not an addressee, you are not authorised to read, copy or use the e-mail or any attachment. If you have received this e-mail in error, please notify the sender by return e-mail and then delete it. ________________________________________ _______________________________________________ HotSpot mailing list HotSpot at mail.cs.virginia.edu http://www.cs.virginia.edu/mailman/listinfo/hotspot ________________________________________ Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: Offenbach/Main HRB 32725 ________________________________________ This e-mail and any attachment contains information which is private and confidential and is intended for the addressee only. If you are not an addressee, you are not authorised to read, copy or use the e-mail or any attachment. If you have received this e-mail in error, please notify the sender by return e-mail and then delete it. ________________________________________ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120222/daa57044/attachment-0001.html From ks7h at virginia.edu Wed Feb 22 10:10:12 2012 From: ks7h at virginia.edu (Kevin Skadron) Date: Wed, 22 Feb 2012 13:10:12 -0500 Subject: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices In-Reply-To: References: Message-ID: <4F452F84.9000303@virginia.edu> Thanks for letting us know. We haven't tested such high power densities before, so this is great to know. /K On 2/22/12 12:57 PM, Avramov, Ivaylo wrote: > Hi Wei, > > Thank you very much for the advice. The models are identical so I tried > your suggestion. Going up from 1 to 6 to 11 to 22 layers increased the > accuracy at every step and it?s a necessity for devices with > high-power-density circuits. > > Regards, > > Ivaylo. > > *IVAYLO AVRAMOV* > > Design Engineer > > *Fujitsu Semiconductor Europe GmbH* > > 3 Concorde Park, Concorde Road > > SL6 4FJ Maidenhead, Berkshire, United Kingdom > > Phone: +44 (0) 1628 50 4737 > > Fax: +44 (0) 1628 50 4777 > > Mailto: ivaylo.avramov at uk.fujitsu.com > > Internet: http://emea.fujitsu.com/semiconductor > > *From:*Wei Huang [mailto:wh6p at virginia.edu] > *Sent:* 22 February 2012 15:47 > *To:* Avramov, Ivaylo > *Cc:* hotspot at mail.cs.virginia.edu > *Subject:* Re: [Hotspot] HotSpot accuracy comparison with FloTHERM for > high-power-density devices > > Hi Ivaylo, > > It is very interesting to see how HotSpot applies to large devices! > > Assuming the device does have a heat spreader and a heat sink, one > possible reason is the thermal properties of the thermal paste (ie. > thermal interface material), which could affect silicon spatial > temperature gradient. > > If you are using the same materials and dimensions in both HotSpot and > FloTherm, then it comes to your reasoning about the mismatch of grid > thickness and grid lateral size, which makes complete sense. I would > suggest chop the silicon into multiple layers using the multi-layer > feature in HotSpot, and see if that improves the results. Of course, > this makes HotSpot more similar to a conventional FEM/FDM simulator like > FloTherm. > > Hope this helps. > > -Wei > > On Wed, Feb 22, 2012 at 7:40 AM, Avramov, Ivaylo > > > wrote: > > Hi, > > > I am modeling a large device with very high power density in certain > regions(200mW dissipated in 200um x 2um), causing concern for hotspot > issues and reliability. A HotSpot simulation using a very detailed model > (80 blocks in an analogue macro) shows a maximum hotspot gradient of > 5.17 degC. Using a very basic pyramid model shows that the gradient > should be about 15 degC. Using the same detailed model in Mentor > FloTHERM gives a gradient of 17 degC. > > Can you think of any reasons for such a large discrepancy between the > results? The substrate is 0.5mm with many circuit block of roughly 150um > x 25um. Does the vertical grid get affected by the definitions for > horizontal grid or is it a fixed number of effective layers? > > It seems like blocks with significantly lower power density have correct > temperature gradient and this issue is more evident for blocks with very > high power density. > > > Regards, > Ivaylo. > > ----------------------------------------------------------------- > IVAYLO AVRAMOV > Design Engineer > Fujitsu Semiconductor Europe GmbH > 3 Concorde Park, Concorde Road > SL6 4FJ Maidenhead, Berkshire, United Kingdom > Email : ivaylo.avramov at uk.fujitsu.com > Tel : +44 (0) 1628 50 4737 > Fax : +44 (0) 1628 50 4777 > > www : http://emea.fujitsu.com/semiconductor > ----------------------------------------------------------------- > > > ________________________________________ > Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc > Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida > Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: > Offenbach/Main HRB 32725 > ________________________________________ > This e-mail and any attachment contains information > which is private and confidential and is intended > for the addressee only. If you are not an addressee, > you are not authorised to read, copy or use the e-mail > or any attachment. If you have received this e-mail > in error, please notify the sender by return e-mail > and then delete it. > ________________________________________ > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > ________________________________________ > Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc > Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida > Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: > Offenbach/Main HRB 32725 > ________________________________________ > This e-mail and any attachment contains information > which is private and confidential and is intended > for the addressee only. If you are not an addressee, > you are not authorised to read, copy or use the e-mail > or any attachment. If you have received this e-mail > in error, please notify the sender by return e-mail > and then delete it. > ________________________________________ > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From wh6p at virginia.edu Wed Feb 22 10:26:47 2012 From: wh6p at virginia.edu (Wei Huang) Date: Wed, 22 Feb 2012 12:26:47 -0600 Subject: [Hotspot] HotSpot accuracy comparison with FloTHERM for high-power-density devices In-Reply-To: References: Message-ID: Hi Ivaylo, Great it works with more layers. Thanks for the update. -Wei On Wed, Feb 22, 2012 at 11:57 AM, Avramov, Ivaylo < Ivaylo.Avramov at uk.fujitsu.com> wrote: > Hi Wei, **** > > ** ** > > ** ** > > Thank you very much for the advice. The models are identical so I tried > your suggestion. Going up from 1 to 6 to 11 to 22 layers increased the > accuracy at every step and it?s a necessity for devices with > high-power-density circuits.**** > > ** ** > > ** ** > > Regards,**** > > Ivaylo.**** > > **** > > *IVAYLO AVRAMOV***** > > Design Engineer**** > > *Fujitsu Semiconductor Europe GmbH***** > > 3 Concorde Park, Concorde Road**** > > SL6 4FJ Maidenhead, Berkshire, United Kingdom**** > > Phone: +44 (0) 1628 50 4737**** > > Fax: +44 (0) 1628 50 4777**** > > Mailto: ivaylo.avramov at uk.fujitsu.com**** > > Internet: http://emea.fujitsu.com/semiconductor**** > > *From:* Wei Huang [mailto:wh6p at virginia.edu] > *Sent:* 22 February 2012 15:47 > *To:* Avramov, Ivaylo > *Cc:* hotspot at mail.cs.virginia.edu > *Subject:* Re: [Hotspot] HotSpot accuracy comparison with FloTHERM for > high-power-density devices**** > > ** ** > > Hi Ivaylo,**** > > ** ** > > It is very interesting to see how HotSpot applies to large devices!**** > > ** ** > > Assuming the device does have a heat spreader and a heat sink, one > possible reason is the thermal properties of the thermal paste (ie. thermal > interface material), which could affect silicon spatial temperature > gradient. **** > > ** ** > > If you are using the same materials and dimensions in both HotSpot and > FloTherm, then it comes to your reasoning about the mismatch of grid > thickness and grid lateral size, which makes complete sense. I would > suggest chop the silicon into multiple layers using the multi-layer feature > in HotSpot, and see if that improves the results. Of course, this makes > HotSpot more similar to a conventional FEM/FDM simulator like FloTherm.*** > * > > ** ** > > Hope this helps.**** > > ** ** > > -Wei**** > > ** ** > > On Wed, Feb 22, 2012 at 7:40 AM, Avramov, Ivaylo < > Ivaylo.Avramov at uk.fujitsu.com> wrote:**** > > Hi, > > > I am modeling a large device with very high power density in certain > regions(200mW dissipated in 200um x 2um), causing concern for hotspot > issues and reliability. A HotSpot simulation using a very detailed model > (80 blocks in an analogue macro) shows a maximum hotspot gradient of > 5.17 degC. Using a very basic pyramid model shows that the gradient > should be about 15 degC. Using the same detailed model in Mentor > FloTHERM gives a gradient of 17 degC. > > Can you think of any reasons for such a large discrepancy between the > results? The substrate is 0.5mm with many circuit block of roughly 150um > x 25um. Does the vertical grid get affected by the definitions for > horizontal grid or is it a fixed number of effective layers? > > It seems like blocks with significantly lower power density have correct > temperature gradient and this issue is more evident for blocks with very > high power density. > > > Regards, > Ivaylo. > > ----------------------------------------------------------------- > IVAYLO AVRAMOV > Design Engineer > Fujitsu Semiconductor Europe GmbH > 3 Concorde Park, Concorde Road > SL6 4FJ Maidenhead, Berkshire, United Kingdom > Email : ivaylo.avramov at uk.fujitsu.com > Tel : +44 (0) 1628 50 4737 > Fax : +44 (0) 1628 50 4777 > > www : http://emea.fujitsu.com/semiconductor > ----------------------------------------------------------------- > > > ________________________________________ > Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc > Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida > Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: > Offenbach/Main HRB 32725 > ________________________________________ > This e-mail and any attachment contains information > which is private and confidential and is intended > for the addressee only. If you are not an addressee, > you are not authorised to read, copy or use the e-mail > or any attachment. If you have received this e-mail > in error, please notify the sender by return e-mail > and then delete it. > ________________________________________ > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot**** > > ** ** > ________________________________________ > Geschaeftsfuehrer/Managing Directors: Dr. Joji Murakami, Brendan Mc > Kearney, Satoru Yamaguchi, Kagemasa Magaribuchi, Shinichi Machida > Sitz/Seat: Langen, Hessen; Registergericht/Commercial Register: > Offenbach/Main HRB 32725 > ________________________________________ > This e-mail and any attachment contains information > which is private and confidential and is intended > for the addressee only. If you are not an addressee, > you are not authorised to read, copy or use the e-mail > or any attachment. If you have received this e-mail > in error, please notify the sender by return e-mail > and then delete it. > ________________________________________ > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120222/f9b6a49d/attachment.html From alexli8 at rocketmail.com Wed Feb 22 18:05:20 2012 From: alexli8 at rocketmail.com (Alex Li) Date: Wed, 22 Feb 2012 18:05:20 -0800 (PST) Subject: [Hotspot] power trace file format Message-ID: <1329962720.30836.YahooMailNeo@web121504.mail.ne1.yahoo.com> Hi, ? First of all thanks a lot for providing such a great tool.? ? I have a?question?regarding the values in Power Trace file. What exactly is unit of the value assigned to each block at each time step? On the FAQ page, it says it should be average power?(#events*Pevent)/timestep, but it doesn't make sense since Power is energy divided by time, thus unit of?(#events*Pevent)/timestep would be J/s^2? ? Right now I have the power of each module, in watt. How should I convert this power consumption in watt to the value in power trace? Does time step matters in this conversion? ? Thanks so much for your help Alex -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120222/e7132f48/attachment-0001.html From Runjie at virginia.edu Thu Feb 23 20:25:25 2012 From: Runjie at virginia.edu (Runjie Zhang) Date: Thu, 23 Feb 2012 23:25:25 -0500 Subject: [Hotspot] power trace file format Message-ID: Hi, Alex The value in power trace is in W. The explanation about average power in FAQ is actually a mistake. It should be (#event*Eevent)/timestep. Thanks for pointing that out! We've corrected it. HotSpot has two modes: transient state mode and steady state mode. In steady state mode, you only need to provide one power value for each module and HotSpot will assume power consumption keeps unchanged for infinite amount of time and then compute the temperature for each module after temperature map become steady. For transient state mode, you will need a trace of multiple power consumption values for each module and HotSpot will give you the corresponding temperature trace. The power trace describes the power activity of the chip within a certain amount of time. Time step does matter here because it defines the duration of each power value in your power trace. Hope this helps. Runjie -------- Original Message -------- Subject: [Hotspot] power trace file format Date: Wed, 22 Feb 2012 18:05:20 -0800 (PST) From: Alex Li Reply-To: Alex Li To: hotspot at mail.cs.virginia.edu Hi, First of all thanks a lot for providing such a great tool. I have a questionregarding the values in Power Trace file. What exactly is unit of the value assigned to each block at each time step? On the FAQ page, it says it should be average power (#events*Pevent)/timestep, but it doesn't make sense since Power is energy divided by time, thus unit of (#events*Pevent)/timestep would be J/s^2? Right now I have the power of each module, in watt. How should I convert this power consumption in watt to the value in power trace? Does time step matters in this conversion? Thanks so much for your help Alex -- Runjie Zhang Computer Engineering University of Virginia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120223/00c831e4/attachment.html From alexli8 at rocketmail.com Thu Feb 23 22:11:46 2012 From: alexli8 at rocketmail.com (Alex Li) Date: Thu, 23 Feb 2012 22:11:46 -0800 (PST) Subject: [Hotspot] power trace file format In-Reply-To: References: Message-ID: <1330063906.48520.YahooMailNeo@web121501.mail.ne1.yahoo.com> Thanks so much Runjie. One follow up question, if I only provide one power value, will Hotspot automatically do steady state simulation instead of transient analysis? Do I have to specify steady state or transient analysis? Thanks again. Alex ________________________________ From: Runjie Zhang To: Alex Li Cc: hotspot Sent: Thursday, February 23, 2012 10:25 PM Subject: power trace file format Hi, Alex ? The value in power trace is in W. The?explanation about average power in FAQ is actually a mistake. It should be (#event*Eevent)/timestep. Thanks for pointing that out! We've corrected it. ? HotSpot has two modes: transient state mode and steady state mode. In steady state mode, you only need to provide one power value for each module and HotSpot will assume power consumption keeps unchanged for?infinite amount of time and then?compute the temperature for each module after temperature map become steady. For transient state mode, you will need a trace of multiple power consumption values for each module and HotSpot will give you the corresponding temperature trace. The power trace describes the power activity of the chip within a certain amount of time. Time step does matter here because it defines the duration of each power value in your power trace. Hope this helps. Runjie? -------- Original Message -------- Subject: ? ? ? ?[Hotspot] power trace file format Date: ? Wed, 22 Feb 2012 18:05:20 -0800 (PST) From: ? Alex Li Reply-To: ? ? ? Alex Li To: ? ??hotspot at mail.cs.virginia.edu? Hi, First of all thanks a lot for providing such a great tool. I have a questionregarding the values in Power Trace file. What exactly is unit of the value assigned to each block at each time step? On the FAQ page, it says it should be average power (#events*Pevent)/timestep, but it doesn't make sense since Power is energy divided by time, thus unit of (#events*Pevent)/timestep would be J/s^2? Right now I have the power of each module, in watt. How should I convert this power consumption in watt to the value in power trace? Does time step matters in this conversion? Thanks so much for your help Alex -- Runjie Zhang Computer Engineering University of Virginia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120223/e3a65d7f/attachment.html From pheonixashes11 at gmail.com Fri Feb 24 01:52:04 2012 From: pheonixashes11 at gmail.com (Bijju Kranthi Veduruparthi) Date: Fri, 24 Feb 2012 04:52:04 -0500 Subject: [Hotspot] Calculating temperature of each block for different initial temperatures Message-ID: Hello Team, Using HotSpot 5.0.1, for a given power profile, how do we obtain the final temperatures of each block given a initial temperature supplied with the -init_file option. I followed the HotSpot/HowTo supplying the following command: ./hotspot -c hotspot.config -f ev6.flp -p gcc.ptrace -init_file gcc.steady -model_type grid -grid_steady_file gcc.grid.steady -o gcc.ttrace I expect to obtain the final temperatures in the gcc.ttrace file? Is my assumption correct? For different init_file 's I am getting the same values in gcc.ttrace. Why is it so? -- -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120224/8d257d6f/attachment.html From Runjie at virginia.edu Tue Feb 28 07:32:50 2012 From: Runjie at virginia.edu (Runjie Zhang) Date: Tue, 28 Feb 2012 10:32:50 -0500 Subject: [Hotspot] Reg:Modelling TSV in Hotspot(Help) In-Reply-To: References: Message-ID: Hi, Arihant Currently HotSpot does not support enough features to model TSV in 3D IC. However, you can check with Prof. Coskun's 3D extension, which can be found here: http://lava.cs.virginia.edu/HotSpot/links.htm . We will integrate this in to HotSpot in the future. Runjie On Fri, Feb 17, 2012 at 2:10 PM, Arihant Jain wrote: > Hello, > > > I want to model TSV in Hotspot in order to study the effect TSV on the > Thermal Profile of the 3D IC.How can I do this? > > Regards > Arihant Jain > Class & Hostel Representative > 4th Year B.Tech Student(Honors) > Electrical Engineering Department > Indian Institute of Technology,Hyderabad > Ph.No:+919441919112 > Email ID: arihant at iith.ac.in > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > -- Runjie Zhang Computer Engineering University of Virginia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120228/797ba2a9/attachment.html From skadron at cs.virginia.edu Tue Feb 28 18:53:04 2012 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Tue, 28 Feb 2012 21:53:04 -0500 Subject: [Hotspot] Calculating temperature of each block for different initial temperatures In-Reply-To: References: Message-ID: <4F4D9310.4040505@cs.virginia.edu> The temperatures will converge to the steady state. The steady state is a function of the power and the configuration and is independent of the initial conditions. /K On 2/24/2012 4:52 AM, Bijju Kranthi Veduruparthi wrote: > Hello Team, > > Using HotSpot 5.0.1, for a given power profile, how do we obtain the > final temperatures of each block given a initial temperature supplied > with the -init_file option. > > I followed the HotSpot/HowTo supplying the following command: > > ./hotspot -c hotspot.config -f ev6.flp -p gcc.ptrace -init_file > gcc.steady -model_type grid -grid_steady_file gcc.grid.steady -o gcc.ttrace > > I expect to obtain the final temperatures in the gcc.ttrace file? Is my > assumption correct? > For different init_file 's I am getting the same values in gcc.ttrace. > Why is it so? > > -- > > > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From vivek2983 at gmail.com Tue Feb 28 19:22:25 2012 From: vivek2983 at gmail.com (vivek chaturvedi) Date: Tue, 28 Feb 2012 22:22:25 -0500 Subject: [Hotspot] technology node in hotpsot? Message-ID: Hello, I am using HotSpot 5.0.1 to perform simulations. I would appreciate if you could please tell me the technology node (nm) used in ev6.flp and default parameters. Also please suggest, if we want to scale it to 45nm or lower, in that case what all parameters should be changed. Thanks & regards Vivek -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120228/8576f744/attachment.html From skadron at cs.virginia.edu Tue Feb 28 19:46:22 2012 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Tue, 28 Feb 2012 22:46:22 -0500 Subject: [Hotspot] technology node in hotpsot? In-Reply-To: References: Message-ID: <4F4D9F8E.3090407@cs.virginia.edu> 130 nm If you wanted to scale the sample materials to another node, you would need to generate a new floorplan and/or change the die size, possibly scale the hotspot configuration (die thickness, etc.), and scale the power trace, which was obtained from a version of Wattch calibrated as described in the ISCA 2003 paper. /K On 2/28/2012 10:22 PM, vivek chaturvedi wrote: > Hello, > I am using HotSpot 5.0.1 to perform simulations. I would appreciate if > you could please tell me the technology node (nm) used in ev6.flp and > default parameters. > Also please suggest, if we want to scale it to 45nm or lower, in that > case what all parameters should be changed. > Thanks & regards > Vivek > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From nedahpg at gmail.com Wed Feb 29 12:36:46 2012 From: nedahpg at gmail.com (Neda Hassanpour) Date: Thu, 1 Mar 2012 00:06:46 +0330 Subject: [Hotspot] question about HotSpot 3D stacking feature Message-ID: Hi, I am new in working with HotSpot thermal model. I want to use HotSpot in 3D NoC architecture as a sensor. The chip I?m working with has 4 layers and there are 16 (4*4) processing cores in each layer. In fact I have a 3D mesh topology (4*4*4) in the chip and I want to calculate each core?s temperature using HotSpot . I have downloaded HotSpot 5.2 and read the How To page. I have also checked the files and some parts of the software code but still have some questions about how to use 3D stacking capability of HotSpot which I would be really thankful if anyone helps me. 1. There are some example files like ev6.flp or gcc.ptrace in the folder of software. Are these files referring to a single processor system? As I mentioned before, I have 64 core processors in a chip and I want to calculate their temperature. Other parts like different levels of cache are not related to my work. How should I prepare suitable inputs (for ptrace and flp file) in this situation? I did not find any place to give HotSpot the number of cores in each layer. 2. Although in the examples in the How To page has been referred to gcc.ptrace.orig and ev6.flp.orig files, I can?t find them in the software folder. My other question is about gcc.grid.steady file. According to the manual, this file?s content relates to steady state temperature in grid model but file includes nearly 4000 lines. What does each line of this file represent? 3. If all the chip layers were the same (for example all have 2D mesh topology and 16 cores), can use the same flp file for all of them? 4. In a flp file, using for each layer, should I consider each processing core as a unit? (Like what has been done in example.flp ) 5. Is it necessary to run any other commands before running the ones below, to use 3D stacking feature of HotSot? *hotspot -c hotspot.config -f -p example.ptrace \* * -o example.ttrace -model_type grid \* * -grid_layer_file example.lcf* 6. How should I generate the initial values in the example.ttrace file, for the first run of above commands ? Can I fill the file with ambient temperature for each element at the beginning? 7. In lcf file, should I consider the heat sink as a layer too? For instance in my case study with 4 layers of process elements, should I define 5 layers in lcf file? Thanks in advance Neda -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120301/fa1045e5/attachment.html