IBIS Macromodel Task Group Meeting date: 22 April 2014 Members (asterisk for those attending): Agilent: Fangyi Rao * Radek Biernacki Altera: * David Banas ANSYS: * Dan Dvorscak * Curtis Clark Cadence Design Systems: * Ambrish Varma Brad Brim * Kumar Keshavan Ken Willis Scott Huss Ericsson: Anders Ekholm Intel: * Michael Mirmak LSI Amaresh Malipatil Dai Xingdong Maxim Integrated Products: Hassan Rafat Mentor Graphics: John Angulo * Arpad Muranyi Andrey Matvienko Vladimir Dmitriev-Zdorov Micron Technology: * Randy Wolff Justin Butterfield QLogic Corp. James Zhou Andy Joy SiSoft: * Walter Katz * Todd Westerhoff * Mike LaBonte Teraspeed Consulting Group: Scott McMorrow * Bob Ross The meeting was led by Arpad Muranyi ------------------------------------------------------------------------ Opens: - Walter asked which of two web meeting services to use for ibis-interconn meetings. - Michael M. said the Intel meeting service would be used - -------------------------- Call for patent disclosure: - None ------------- Review of ARs: - None ------------- New Discussion: DLL dependency checking: - Arpad: Should we require models to be checked for code library dependencies? - David: There is a utility with a dumpbin command for that. - Arpad: It should be built into our own checking. - Walter: Maybe one of the test suites could be used. - Ambrish: The DLL is too much of a black box. - Todd: The spec says there should be no dependencies. - We should at least make sure that part is clear. Backchannel: - Arpad asked that we focus on the issue at hand to avoid going in circles. - Walter showed two example AMI files. - Walter: The BCI file requirement poses maintenance problems. - A very general TX AMI file would eliminate the need for that. - One shown here is generic, the other for PCIe3. - There are three co-optimization modes. - Manual: The TX does what it's told. - Auto: The TX can modify based on impulse response inputs. - Co-optimize: RX inputs and impulse response are used. - TX taps indexes are usually given as integers from 0 to N. - Coefficients are float numbers. - There would be reserved parameters to say what the names of key Model_Specific parameters are. - Also the ranges would be given, even the target parameters have these. - Michael M.: Coefficients do not assume summation to 1.0? - Walter: Yes they must sum up to 1.0. - Michael M.: Shouldn't the index ranges map up to 1.0? - Walter: Tap -1 shown here has 32 index settings. - The register may have a different set, not discussed here. - The 32 values may not map the whole range,and maybe not linearly. - Converting from coefficient to index can be very complicated. - This is why Tap_Conversion is proposed. - Rarely are both indexes and coefficients used together. - Michael M.: You may have those reversed. - Walter: The RX sends out tap coefficients. - Michael M.: One of the 4 phases of EQ is stating max index values each side supports. - Walter: We need to look at that. - To exchange tap coefficients we only need to know the parameter to use. - RX Init can effectively only return tap coefficients. - Walter showed an 802.3KR example. - The TX has co-optimize mode which means it accepts tap increments. - The increment parameter is a table showing the range for each tap. - The RX can only increment and decrement. - Michael indicates it can also set index values. - This should work for the foreseeable future, at least four years. - We need to know if the RX Init can co-optimize the TX. - Likewise for RX GetWave. - Another parameter tells whether the impulse response includes TX EQ. - David: Is this a new information flow? - Walter: It is the output of TX Init with no EQ. - This is an important detail. - Arpad: Is this related to Use_Init_Output? - Walter: There is no connection. - The RX defines training requirements, similar to BIRD 147. - BIRD 147 adds the Bit type, we should not add new types. - This proposal uses reserved String parameter names. - Tap_Increment is declared by the RX too. - This has all of the BIRD 147 functionality, eliminating the BCI file. - Ambrish: How you the RX and TX support multiple protocols? - Walter: This will not support that. - Mike: BIRD 147 has Backchannel_Protocol pointing to a single BCI file. - Ambrish: There can be multiple List values. - Walter: If we have a BCI file it should belong to the RX. - Ambrish: The TX and RX must point to the same BCI file to insure they support the same protocol. - Walter: Is there any way other than increment to send information to a TX? - Ambrish: The BCI allows that to be defined, we don't know what will be done in the future. - Walter: Do we know of any new requirements in development? - Ambrish: The BCI contains all parameters for the TX and RX for whatever protocol is used. - Walter: This requires us to standardize the BCI files. - Ambrish: A subcommittee could meet every 6 months to set that. - Walter: This eliminates the need for the subcommittee. - Ambrish: Your proposal will require more changes as new protocols arise. - Walter: I am on several standards committees, none of them need new parameters. - Ambrish: We don't want to make changes soon after a standard comes out. - Walter: No Reserved_Parameter has been suggested since the 5.1 changes. - Some people have been co-optimizing for years. - We need to decide which method is easier for IC and EDA tool vendors. - Bob: A key item would be the need for a new file checker. - Both proposal involved about 8 new Reserved_Parameters. - We would need TapIndex as a new type. - Kumar: The EDA tool should not have to translate what the RX sends to the TX. - Also indexes are very architecture specific. - This has a lot of new parameters. - BCI has the minimum set of parameters for communication. - Standards committees have no hard rules that coefficients will always work the same way. - Ambrish: And the BCI file makes it easier for models to support multiple protocols. - Walter: Co-optimization can be done in different places. - Each has different requirements. - Ambrish: It is not that complicated, Init works one way and GetWave works another way. - Kumar: It is important to support the Init flow with coefficients. - Arpad: These should be posted online. AR: Walter send backchannel example files, Mike post these. - Michael M.: We need to define how presets fit in. - Walter: We might need a table for that. - Kumar: That is a good one for the BCI file, it is protocol-specific. - We should not need a Reserved_Parameter for that. - Walter: If it's in the BCI file there is a reserved parameter. - Ambrish: I would like to present next week. - Michael M.: We need to work out more details on PCIe too. ------------- Next meeting: 29 April 2014 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives