Minutes from Sep 2011 ibis-atm meetings are attached. Mike
IBIS Macromodel Task Group Meeting date: 20 Sep 2011 Members (asterisk for those attending): Agilent: Fangyi Rao * Radek Biernacki Altera: * David Banas Ansys: Samuel Mertens * Dan Dvorscak * Curtis Clark Arrow Electronics: Ian Dodd Cadence Design Systems: Terry Jernberg * Ambrish Varma Celsionix: Kellee Crisafulli Cisco Systems: Ashwin Vasudevan Syed Huq Ericsson: Anders Ekholm IBM: * Greg Edlund Intel: Michael Mirmak LSI Logic: Wenyi Jin Mentor Graphics: * John Angulo Zhen Mu * Arpad Muranyi Vladimir Dmitriev-Zdorov Micron Technology: Randy Wolff NetLogic Microsystems: Ryan Couts Nokia-Siemens Networks: * Eckhard Lenski QLogic Corp. * James Zhou Sigrity: Brad Brim Kumar Keshavan Ken Willis SiSoft: * Walter Katz * Todd Westerhoff Doug Burns Snowbush IP: Marcus Van Ierssel ST Micro: Syed Sadeghi Teraspeed Consulting Group: Scott McMorrow * Bob Ross TI: Casey Morrison Alfred Chong Vitesse Semiconductor: Eric Sweetman Xilinx: Mustansir Fanaswalla unaffiliated: * Mike LaBonte The meeting was lead by Todd Westerhoff ------------------------------------------------------------------------ Opens: - None -------------------------- Call for patent disclosure: - None ------------- Review of ARs: - Walter update BIRD 123.2.4 and send to Mike for posting - In progress ------------- New Discussion: Todd showed a presentation from Greg: - Slide 2: - Greg: Should we keep the terminator block on slide 2? - Walter: We should not - David: Does that mean not series pin in a TX model? - Walter: Sometimes people add a termination to the buffer - But it's really part of the buffer model - David: It would be good to have terminator options instead of several buffers - Bob: One way to have terminations is through sub-models - Todd: We usually need diff termination - IBIS [Model] declarations are single ended - Slide 3: - Greg: The simulated PLL and CDR will model RX jitter - Walter: This is accurate for stat - For time domain CDR is in the ALG box - Getwave usually does not have CDR in it - Don't know if this is output or input to EQ2 - Kumar and Ambrish needed the files, no web client running - Walter: We have a RX model of type Terminator - Radek: That should be allowed - Todd: If clock_ticks are not provided by the model, RX Clock_Recovery params are used to provide a clock budget - The main question is what point in the circuit jitter is injected - Discussion of which parameters go where - David: Why no TX voltage noise parameters? - What protection would there be against disclosing jitter specifics? - Is TX Rj from EDA tool to model or vice-versa? - Is it from the user or from the model? - Walter: It is from the model - Kumar: RX Sj is modeled at the CDR - The EDA tool should not add Sj - David: How are refclk imperfections modeled? - Todd: TX Rj is an Info param - In the current BIRD the model is not expected to do anything with RX Sj - It never sees the param - BIRD 123 is about the EDA tool modulating clock_ticks - Walter: When CDR does not include RX Sj DCD on refclk CDR ??? - If RX Sj is modeled in the algorithmic model's CDR it should not be an info param - Todd: New params can be created to tell the model how much Sj to generate - Walter: It would have a new name, Model_Specific - Todd: Does the simulator use only Info parameters? - What if the DLL accepts an Input Rx_Sj parameter? - Walter: It would be confusing if we overloaded that - Arpad: The spec doesn't say that can't be done - Mike: Slides 2 & 3 should have the jitter params annotated at the right places Walter showed an email to explain Sj in BIRD 123.2.5: - David: We need to specify this at multiple frequencies - Walter: Fangyi brought that up - A new BIRD would be needed for that - Kumar: Would multiple instances of Sj do it? - Walter: We would need something like a separate file or PDF - Dual-Dirac has just 2 points - In statistical mode a clock PDF might be adjusted - David: Is it guaranteed simulators will observe clock_ticks? - Walter: Yes - Todd: Is there a difference between the simulator and the DLL modulating Sj? - Kumar: The simulator will be more pessimistic - The DLL will remove it - Walter: At high freq the model might ask the EDA tool to do it - Todd: Injecting at the latch and refclk are different - Kumar: If the tool does the jittering it can't be tracked - Walter: BUJ is jitter than the DLL can't model - But it needs to be incorporated into the final result - David: Why defend the request to the EDA tool to add jitter? - Kumar: Often the CDR tracks jitter and takes it out - Todd: It helps to know the physical basis of these jitters - Power supply noise will not track out - Walter: Anything that can't be tracked should be disclosed - Todd: Is frequency required if RX Sj is specified? - David: How is a simple magnitude of Sj helpful? - That depends on how much goes into the CDR - Walter: This is spread across all frequencies - David; No CDR has a white noise jitter function - Walter: It's a budget, which distribution should it be? - We had discussed whether these should be Info or Out - Amplitude noise changes with time where AGC is used Meeting ended. ------------- Next meeting: 27 Sep 2011 12:00pm PT Next agenda: 1) Task list item discussions ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives
IBIS Macromodel Task Group Meeting date: 06 Sep 2011 Members (asterisk for those attending): Agilent: Fangyi Rao * Radek Biernacki Altera: * David Banas Ansys: Samuel Mertens * Dan Dvorscak * Curtis Clark Arrow Electronics: Ian Dodd Cadence Design Systems: Terry Jernberg * Ambrish Varma Celsionix: Kellee Crisafulli Cisco Systems: Ashwin Vasudevan Syed Huq Ericsson: Anders Ekholm IBM: * Greg Edlund Intel: Michael Mirmak LSI Logic: Wenyi Jin Mentor Graphics: * John Angulo Zhen Mu * Arpad Muranyi Vladimir Dmitriev-Zdorov Micron Technology: Randy Wolff NetLogic Microsystems: Ryan Couts Nokia-Siemens Networks: * Eckhard Lenski QLogic Corp. * James Zhou Sigrity: Brad Brim * Kumar Keshavan * Ken Willis SiSoft: * Walter Katz * Todd Westerhoff Doug Burns Snowbush IP: Marcus Van Ierssel ST Micro: Syed Sadeghi Teraspeed Consulting Group: Scott McMorrow * Bob Ross TI: Casey Morrison Alfred Chong Vitesse Semiconductor: Eric Sweetman Xilinx: Mustansir Fanaswalla unaffiliated: * Mike LaBonte The meeting was lead by Arpad Muranyi ------------------------------------------------------------------------ Opens: - Arpad: Last week we discussed having an IBIS section on version differences - Walter: IBIS 5.1 should only discuss AMI 5.1 - We should take a resolution to the Open Forum - Arpad: We should have this on the agenda - Greg Edlund wanted to show some slides on jitter -------------------------- Call for patent disclosure: - None ------------- Review of ARs: - Arpad: Inform Open Forum that BIRDs 137.1 and 143 will not change - In progress - Fangyi: to send suggestions on Jitter BIRD to Walter - Done - Walter: Send updated BIRDs 123.2.4 to Mike for posting - Was sent to list AR: Mike post latest BIRD 123.2.3 to ATM web ------------- New Discussion: Arpad showed BIRD 137.2: - One minor typographical fix - Arpad described the changes - Bob: This is ready to go - There should be discussion on how we deal with version changes - Walter motioned to submit the BIRD to Open Forum - Curtis seconded AR: Arpad submit BIRD 137.2 to Open Forum Arpad showed BIRD 143.1: - This has the same changes as BIRD 137.2 - Arpad described the changes - Walter motioned to submit the BIRD to Open Forum - Curtis seconded AR: Arpad submit BIRD 143.1 to Open Forum Arpad showed BIRD 140.1: - Arpad: This has new wording that Walter and I agreed to - [External Model] and [External Circuit] have "corner_name" - We only need to reference these names (Typ, Min, Max) - This could have another paragraph about dependency tables - We are not yet ready for that in this BIRD - AMI have have to map to those names to 0, 1, 2 - Walter: EDA tools all have an intrinsic corner variable - In one place Corner is a Format - In another it's a parameter - Bob: For Type String Corner would be quoted - Walter described a Corner parameter - Arpad: The question is if we want to untable this BIRD - We can discuss next meeting - Walter motioned to discuss BIRD 140.1 next meeting - Bob and David seconded - Ambrish asked if we should wait for dependency tables first Arpad showed a slide presentation from Greg: - Greg: Have been discussing AMY issues in Quality task group meetings - Have been doing jab work to verify models - Studying where jitter comes from - Users will see a list of parameters but will not know know what they are controlling - BIRD 123 are jitter terms external to the DLL - Walter; Rx_Noise is an exception - Todd: That is still external to the DLL - Slide 1: - Greg explain the diagram - Walter: The analog model is dashed - The PKG is in there but it's usually in the IBIS file - EQ may be between Pre and Out - Greg: A 1 tap would have 2 parallel output stages - Walter: That depends on how the TX is implemented - Todd: EQ should be before Pre - Walter: Replace Pre with EQ - Todd: - Arpad: What is the purpose of the Quality discussion? - Greg: What does the DLL correspond to? - Walter: The Pre - Arpad changed Pre to EQ - Todd: Jitter is only what comes out of the EQ box - Ambrish: So no jitter can be in the model? - Walter: It can't be in Init - If it is Getwave the EDA tool must not add more jitter - The refclk path is controlled by AVDD and VDD - David: Can we tell the DLL what noise to assume? - Walter: We have params to tell the EDA tool what the DLL models - Todd: The block marked AMI should be Algorithmic - Radek: The pkg is not part of the model? - Walter: It is in IBIS - Arpad: Those are often not good enough - Ambrish: Will PKG affect IR? - Todd: yes - David: For FPGAs the PKG has to be kept separable - Radek: The PKG is to be used if it is in the IBIS file - Arpad made further block diagram changes The call dropped saying that the chairperson disconnected. We resumed: - Todd: When we say "model" it means to the edge of the die -The analog channel includes all interconnect - Walter: It also includes Term - James: Where is the analog data saved? - Walter: We know where the PKG and ALG models are - The buffer analog in 5.0 is either IV/VT or [External Model] - BIRD 122 proposes new AMI parameters to describe intrinsic models - It will be something like a Thevenin or Touchstone model - It might be ISS at some point - The jitter could be internal or external to the AMI file - James: Model makers have to provide TX DCD, etc.? - Kumar: They only have to give corner values - Walter: Model makers will know what jitter params are and are not included - The ones not included must be parameters - Radek: Users can not modify that - Todd: If jitter is a range it implies the user can pick a number - James: So users can change model jitter parameters? - Walter: Users can select jitter if the model allows it - Todd: Should users change AMI files? - Kumar: Users do not have unlimited freedom to change AMI files? - Walter: If IP is placed in different ICs the jitter changes - James: The model maker can't know what the power supply is like - Ambrish: Some jitter comes from the system - The EDA tool has to add that - Can the user add more DCD? - Walter: Yes - We should remove Out as an option for jitter params - Kumar: If the AMI gives a range how will the user know he can add more jitter? - Walter: The manual should give enough info - Todd: The proposal is for the EDA tool to modulate stimulus - The DLL can modify internally - It can have an AMI that sets jitter to zero - Users can edit that file - Ambrish: What does the user do the add more system jitter - Todd: If users do not edit AMI files the value can be a corner or range - The EDA tool can still prompt for additional jitter - James: The spec and EDA tool can do many things - It's hard to know what it is actually doing though - Todd: They may have trade secrets - James: I need to know what is in and not in the model - Todd: Anything handled by the DLL should be zero in the AMI file - Walter: There will be a PS component that will be low frequency - The EDA tool has to account for that - James: Can we tell the user which parameters not to change? - Walter: They should not change AMI files - Ambrish: The EDA tool could handle it internally - Todd: We have a model standard for ICs - We can't tell and EDA tool what it can't do - Greg: The DLL would have to model refclk jitter - Walter: The DLL does not handle that - It gets transition times - Todd: It is modifying the output - James: Walter was saying the opposite - Walter: Transition times will have any jitter added by the tool - Todd: Nothing on the table now talks about that refclk input - Greg: This was an issue in the early days of PCIe - Walter: They gave budgets for jitter at the output - There was nothing at the inputs - Now they have to describe jitter propagation - It can't be a compliance test any more - James: It is still not clear what the DLL is not doing - Walter: It would be appropriate for the DLL to say what it's input should include - Ambrish: These models are supposed to be black box - Todd: We can standardize how jitter is injected - Models can be made with parameters that together model all jitter - James: It should be clear that these jitter params do not mean the model handles it - Even for Rj - Walter: There is a debate about Rj - Applying it in time domain up front will not be accurate enough to get the tails - Post-processing may be better - Walter read from BIRD 123 to clarify this Slide 2 was not discussed Meeting ended. ------------- Next meeting: 6 Sep 2011 12:00pm PT Next agenda: 1) Task list item discussions ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives
IBIS Macromodel Task Group Meeting date: 13 Sep 2011 Members (asterisk for those attending): Agilent: * Fangyi Rao * Radek Biernacki Altera: * David Banas Ansys: Samuel Mertens * Dan Dvorscak * Curtis Clark Arrow Electronics: Ian Dodd Cadence Design Systems: Terry Jernberg * Ambrish Varma Celsionix: Kellee Crisafulli Cisco Systems: Ashwin Vasudevan Syed Huq Ericsson: Anders Ekholm IBM: * Greg Edlund Intel: Michael Mirmak LSI Logic: Wenyi Jin Mentor Graphics: * John Angulo Zhen Mu * Arpad Muranyi Vladimir Dmitriev-Zdorov Micron Technology: Randy Wolff NetLogic Microsystems: Ryan Couts Nokia-Siemens Networks: * Eckhard Lenski QLogic Corp. * James Zhou Sigrity: Brad Brim * Kumar Keshavan Ken Willis SiSoft: Walter Katz * Todd Westerhoff Doug Burns Snowbush IP: Marcus Van Ierssel ST Micro: Syed Sadeghi Teraspeed Consulting Group: Scott McMorrow * Bob Ross TI: Casey Morrison Alfred Chong Vitesse Semiconductor: Eric Sweetman Xilinx: Mustansir Fanaswalla unaffiliated: * Mike LaBonte The meeting was lead by Arpad Muranyi ------------------------------------------------------------------------ Opens: - Arpad: May have to leave soon, will leave meeting line open - Greg will discuss jitter parameters -------------------------- Call for patent disclosure: - None ------------- Review of ARs: - Arpad submit BIRD 137.2 to Open Forum - Done - Arpad submit BIRD 143.1 to Open Forum - Done - Mike post latest BIRD 123.2.3 to ATM web - Done just now 123.2.5 - Arpad: Inform Open Forum that BIRDs 137.1 and 143 will not change - Done, will be discussed Friday ------------- New Discussion: Arpad showed BIRD 140.1: - Arpad: Changes are near the bottom with ** - A_to_D and D_to_A have another column - Has Format Corner specs - Examples added - New analysis added - Radek: The official names of the corners are typ, min and max? - Arpad: We might have to add mappings - Radek: Min and max are OK, but we would want to avoid using slow and fast - Arpad: We are really saying non-verbally what the mapping is - Todd: In IBIS min means two things: - For most params slow/weak - For C_comp just the min value - Tools like SPICE have more then 3 combinations - I prefer slow/typ/fast - Arpad: We can't change typ/min/max as the selector names - Todd: It would be better to say min always means slow/weak except for C_comp - Arpad: This BIRD is not written to solve that problem - Ambrish: That one line should say typ/slow/fast - Radek: If the model maker wants to use min and max values they will not use Corner - Arpad: In the first section we don't explain what Corner does - Todd: I drop my objection - Arpad is saying slow and fast are coming out of nowhere - IBIS says what min and max mean - Arpad just wants to extend that top AMI - We will go beyond 3 corners - The IBIS analog model can't go beyond 3 corners - Arpad: Nothing keeps us from adding more columns - Radek: Tying them together may be wrong for the future - Ambrish: We would only have to add one line - James: We should be able to do that - The real issue is what min/max means - Ambrish: We only have to point to section 9 paragraph 2 - James: It should be clear what the model maker did to create each column - Arpad: Is this BIRD acceptable as is? - We will continue this topic next week Todd showed Greg's jitter parameter document: - Greg: These are all external to the DLL? - Todd: Agree - Kumar: Disagree - Assuming this is about BIRD 123 - Todd: This is about 5.0 - Kumar: RX Sj has to be handled by the CDR - Also RX_Receiver_Sensitivity - Todd: What does the simulator do if it's in the AMI - Kumar: It depends on the model - David: Does external mean the DLL gives the EDA tool options? - Or does the tool tell the model how much it should be? - It would not make sense for the tool to make up values - Arpad: Only the model maker knows - The tool should not have a say - Todd showed the data format table - Range allows any value within constraints - David: It comes from the model maker to say what the silicon does - Todd: Yes - External means the DLL knows nothing about it - The simulator is expected to do something - Vih and Vil are handled that way - David: "The model knows nothing about it" is confusing - It implies the model maker did not provide it - Kumar: These are often like eye mask parameters - Todd: The RX is guiding the post-processing of data - But the simulator handles it - Arpad: Usually the outcome is an eye - How does RX_Receiver_Sensitivity fit in? - David: It gives what you need to define eye width - Arpad: Is this a poor man's way to define eye mask? - Usually Vinh and Vinl are used - Todd: Anything between RX_RS would treated as impinging on an eye mask - Fangyi: RX sens is used to calculate BER - That is not device behavior - The model should represent only device behavior - The EDA tool has to use this - Todd: The RX models delivers clock_times and waveform at the sample latch - RX_RS tells us how to read the value there - No part of IBIS specifies an eye mask - Fangyi: The eye is not sufficient - You have to know the model - Todd: Yes several things are need to create an eye - Todd slightly modified the first slide Meeting ended. ------------- Next meeting: 20 Sep 2011 12:00pm PT Next agenda: 1) Task list item discussions ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives