******************************************************************************* ******************************************************************************* Buffer Issue Resolution Document (BIRD) BIRD ID#: 11.2 ISSUE TITLE: Improving common error detection in IBIS_CHK program. REQUESTOR: Kellee Crisafulli, HyperLynx Inc. DATE SUBMITTED: 03-28-94 DATE REVISED: 04-21-94 DATE ACCEPTED BY IBIS OPEN FORUM: April 29, 1994 ******************************************************************************* ******************************************************************************* STATEMENT OF THE ISSUE: Several common problems with IBIS models are not detected with the present version of IBIS_CHK. Two main problems include: 1) Incorrect 'I'(current) signs in the V/I tables. 2) Pullup and POWER_clamp V/I tables are not VCC relative. This BIRD is directed at problem 1 only. A 2nd separate BIRD will be generated to address problem 2. ******************************************************************************* STATEMENT OF THE RESOLVED SPECIFICATIONS: The following changes apply to version 1.1 and forward versions of the IBIS_CHK program and including testing of some parameters for BIRD 7.2 for ECL model types *************************************************************************** Change 1- Add verbage on current direction. *************************************************************************** Keywords: [Pullup], [Pulldown], [GND_clamp], [POWER_clamp] Required: Yes, if they exist in the device Description: The data points under these keywords define the V/I curves of the pulldown and pullup structures of an output buffer and the V/I curves of the clamping diodes connected to the GND and the POWER pins, respectively. Currents are considered positive when their direction is into the component. *************************************************************************** Change 2- Add detection to IBIS_CHK program for V/I table 'I' sign errors. *************************************************************************** For each of the following V/I tables: Pullup, Pulldown, POWER_clamp, GND_clamp 1) Find the minimum and maximum voltage points (Vmin, Vmax) in the table. 2) IF:The current in the TYPICAL column corresponding to Vmax is less than the current in the TYPICAL column corresponding to Vmin than the table is assumed to have decreasing current. ELSE IF:The current in the TYPICAL column corresponding to Vmax is greater than the current in the TYPICAL column corresponding to Vmin than the table is assumed to have increasing current. ELSE: The table is assumed to have equal current." Note: This works for all cases of discontinuities unless the magnitude of discontinuity is such that this model is in all probability competely unrealistic. Examples: *** example with non-monotonic data at the end point V: I: 0.00 0.0 4.90 50.0ma 4.91 49.9ma 4.93 56.7ma 5.00 3.0ma -> V/I table has increasing current (3.0 > 0) Vmax = 5.0, I =3.0mA Vmin = 0.0, I =0.0 *** example with negative to positve voltages with negative first V: I: -5.00 -0.1ma 0.00 0.0 5.00 100.0ma -> V/I table has increasing current (100 > -0.1) Vmax = 5.0, I=100mA Vmin = -5.0, I=-0.1mA *** example with table data entered postive voltages first V: I: 5.00 10.1ma 0.00 0.0 -5.00 -10.1ma -> V/I table has increasing current (10.1 > -10.1) Vmax = 5.0, I=10.1mA Vmin = -5.0, I=-10.1mA *** example with only two entrys V: I: 0.00 0.0 -5.00 10.1ma -> V/I table has decreasing current (0 < 10.1) Vmax = 0.0, I=0 Vmin = -5.0, I=10.1mA *** ECL example [Pullup] Voltage I(typ) I(min) I(max) 0.0 0 0 0 0.7 -0.2m -0.2m -0.2m 0.73 -0.4m -0.4m -0.4m 0.75 -0.8m -0.8m -0.8m 0.76 -1.2m -1.2m -1.2m 0.77 -1.6m -1.6m -1.6m 0.8 -4.4m -4.4m -4.4m 0.82 -7.6m -7.6m -7.6m 0.85 -14.2m -14.2m -14.2m 0.9 -30.0m -30.0m -30.0m 1.0 -58.0m -50.0m -68.0m -> V/I table has decreasing current ( -58 < 0) Vmax = 1.0, Ityp=-58mA Vmin = 0, Ityp=0 [Pulldown] Voltage I(typ) I(min) I(max) 0.0 0 0 0 1.6 -0.2m -0.2m -0.2m 1.62 -0.4m -0.4m -0.4m 1.64 -0.6m -0.6m -0.6m 1.65 -0.8m -0.8m -0.8m 1.66 -1.2m -1.2m -1.2m 1.67 -1.6m -1.6m -1.6m 1.68 -2.4m -2.4m -2.4m 1.69 -3.2m -3.2m -3.2m 1.70 -4.4m -4.4m -4.4m 1.72 -7.4m -7.4m -7.4m 1.75 -14.2m -14.2m -14.2m 1.8 -30.5m -30.5m -30.5m 1.9 -65.0m -60.0m -75.0m -> V/I table has decreasing current ( -65 < 0) Vmax = 1.9, Ityp=-65mA Vmin = 0.0, Ityp= 0 *** An abreviated INTEL model for a CMOS output |**************************************************************************** [Pulldown] | Voltage I(typ) I(min) I(max) -5.00V -38.70mA -29.47mA -51.22mA -1.00V -24.88mA -19.18mA -32.90mA -0.50V -14.35mA -11.06mA -19.05mA 0.00V -11.84pA -554.66pA -11.03pA 100.00mV 3.20mA 2.47mA 4.27mA 200.00mV 6.24mA 4.80mA 8.30mA 4.90V 38.68mA 29.45mA 51.18mA 5.00V 38.70mA 29.47mA 51.22mA 10.00V 39.96mA 30.37mA 53.06mA -> V/I table increasing [GND_clamp] | Voltage I(typ) I(min) I(max) -5.00V -680.00mA NA NA -1.10V -75.50mA NA NA -600.00mV -950.00uA NA NA -500.00mV -78.00uA NA NA -200.00mV 0.00pA NA NA -100.00mV 0.00pA NA NA 0.00V 0.00pA NA NA 5.00V 0.00pA NA NA -> V/I table increasing (0 > -680) [Pullup] | Voltage I(typ) I(min) I(max) -5.00V 38.14mA 27.33mA 54.76mA -4.50V 37.49mA 26.87mA 53.79mA -1.00V 17.13mA 12.81mA 23.55mA -0.50V 9.26mA 6.96mA 12.66mA 0.00V 13.57pA 613.51pA 11.04pA 100.00mV -1.96mA -1.48mA -2.67mA 200.00mV -3.87mA -2.92mA -5.27mA 500.00mV -9.26mA -6.96mA -12.66mA 1.80V -26.79mA -19.79mA -37.25mA 1.90V -27.74mA -20.46mA -38.64mA 4.60V -37.62mA -26.97mA -54.00mA 4.70V -37.76mA -27.06mA -54.20mA 5.00V -38.14mA -27.33mA -54.76mA 10.00V -44.52mA -33.72mA -61.15mA -> V/I table decreasing [POWER_clamp] | Voltage I(typ) I(min) I(max) -5.00V 1.05A NA NA -1.10V 79.00mA NA NA -1.00V 54.00mA NA NA -900.00mV 29.00mA NA NA -800.00mV 10.40mA NA NA -200.00mV 0.00uA NA NA -100.00mV 0.00uA NA NA 0.00V 0.00pA NA NA -> V/I table decreasing 3) IF the model is any of the following types:(Input_ECL, Output_ECL, I/O_ECL) { Verify that: - Pullup V/I table has equal or decreasing current - POWER_clamp V/I table has equal or decreasing current - Pulldown V/I table has equal or decreasing current - GND_clamp V/I table has equal or increasing current } ELSE { Verify that: - Pullup V/I table has equal or decreasing current - POWER_clamp V/I table has equal or decreasing current - Pulldown V/I table has equal or increasing current - GND_clamp V/I table has equal or increasing current } Note: This specifically allows constant current generators and 0 current tables. 0 current tables may be used to indicate table is unused. 4) If any table verification fails report the following error message: 'Error found in xxx V/I table at line number nnn!'. Where xxx is one of the following Pullup, Pulldown, POWER_clamp, GND_clamp. Where nnn is the line number. *************************************************************************** Change 3- Add a header comment statement at the TOP of the IBIS_CHK program to insure that new changes to the IBIS_CHK program donot break tests that worked in old MAJOR versions. This approach makes the program larger however it insures the parser always works the same on older versions of IBIS. This apporach uses more memory, but has the reward of low maintaining costs. The IBIS_CHK program is very small and would not be effected by this until many revisions have occured. *************************************************************************** NOTICE TO ANY PERSON MODIFING THIS PROGRAM! ------------------------------------------- This program SHALL NOT BE MODIFIED unless there is an associated IBIS BIRD. Said BIRD shall be agreed upon by IBIS committee vote. Only the currently elected IBIS_CHK 'czar and programmer' is allowed to modify the source code. The present CZAR is Jon Powell (April 1994). The IBIS committee may also hire programmers from time to time to make major changes to the source code. Note: Source licensees are free to modify their own copies of this source code in any way they choose. Source licensees shall not redistribute the source code modified or otherwise. Source licensing is available from the IBIS open forum. The IBIS open forum is non-profit. The code for each MAJOR version of the IBIS_CHK program SHALL NOT BE MODIFIED when adding code for the next version of the IBIS specification. Instead completely new code for all functions and features shall be created. This may require duplication of numerous functions. Each function shall be preceded by VXX_ where XX is the MAJOR version of the IBIS specification which is being parsed and tested. A MAJOR version would for example be 1.x going to 2.x. A MINOR version would for example be 1.1 to 1.2. Functions using the above syntax would look as follows: V01_GetValue MINOR revisions DO NOT required new code. Startup code shall be provided at the top of the program which reads the version number from the IBIS file and runs the portion of the program corresponding to that MAJOR version. Code which is used only by the program startup function is not duplicated. ******************************************************************************* ANALYSIS PATH/DATA THAT LED TO SPECIFICATION: Change 1) Suggested from inputs of several people in Email and at the at previous IBIS meetings. Change 2) Defective IBIS models are slipping through the IBIS_CHK program. A method for determining sign errors in V/I table was determined based on inputs from several people including: Kellee Crisafulli, HyperLynx Jon Powell, Quad Design Arpad, Intel Bob Ward, TI Bob Ross, Interconnectix Maah Sango, Contec I reviewed all the comments that have been submitted and I believe this method will work with all conditions mentioned so far. Change 3) I am proposing a method of insuring that future IBIS_CHK modifications donot effect the parsing of older IBIS files. This only applies to MAJOR revisions in the specification, like 2.0 comming up. This would not apply to this update, it would however be added to the program header information now and would apply to all future MAJOR updates to the IBIS specification. ******************************************************************************* ANY OTHER BACKGROUND INFORMATION: Comments about EMAIL leading up to this BIRD -------------------------------------------- From: Jon Powel, qdt Cases that will not work for proposed algorithm V: I: 0.0 0.0 3.5 50.0 3.51 49.9 3.52 49.8 3.53 49.7 5.0 100.0 any OC any OD If current is not negative at GND for the pullup then it cannot pullup? (except for ECL OC etc?). Action: problems fixed in BIRD 11.0 ****************** From: Arpad, Intel What happens if there are equal number of increases and decreases in a curve? Most of the curves I am generating lately do that and I do not think I am doing it wrong. Action: problems fixed in BIRD 11.0 ******************** Bob Ward, TI I think the proposed method of current sign determination might run into trouble in at least two cases. One is when there are exactly the same number of rising and falling segments. The other is the same problem one runs into testing for monotonicity of the independent variable. That problem is that very small oscillations occur on the "flat" part of a numerically generated curve, not so much because they are real, but because of the nature of small numbers, finite precision, and floating point round off. Action: Problems fixed in BIRD 11.0 ****************************** Bob Ross, Interconnectix, Inc. If this proposal is adopted, it would apply to just IBIS Version1.1. The polarity rules do not comply with proposed IBIS extensions BIRDS 3 and 4 for ECL type devices. For ECL type devices, the polarities of both the Pullup and Pulldown tables will go in the same (decreasing) direction because BOTH tables are tabulated referenced to Vcc using Vtable = Vcc - Vout. I am sure this will become another area of confusion, justifying a test. Action: Problem fixed BIRD 11.0 ************************** Maah Sango, Contec Microelectronics USA Inc 2. I agree with Bob Ross that the proposed scheme for enforcing data integrity will not work for ECL, nor for a few other device types. (See item 3 of the proposal). Either the "pullup" or the "pulldown" data for the ECL and other device types will violate these requirements. Action: Problem fixed Bird 11.0 3. The proposed scheme, item 3, will not always work even for CMOS unless we use only the magnitudes of the currents. Existing data for ageing (old) devices is sometimes presented with positive currents and sometimes with negative currents for CMOS pullup devices. Action: Problem fixed Bird 11.0 4. I am not sure that checking only the two end points will always guarantee the conclusions we are assuming here. Current(I) data in between these two points may increase or decrease and still be perfectly valid, particularly if we think of device types other than CMOS. Action: Numerous examples cited, none found that present a problem ******************************* Arpad, Intel If we JUST want a polarity checker for the MOST COMMON error, I have another most common error to check for: VCC-relative! That needed more explaining than the polarity of the current in my experience, and I have seen more people being confused about that. Should that also be checked for then? Action: Another Bird needs to be generated separate from Bird 11 ******************************************************************************* Changes from the April-1-94 IBIS meeting BIRD 11, IBIS_CHK Changes Kellee Crisafulli proposes three changes in this BIRD, each related to the sign of VI table data, which has proven to be a common pitfall among new modelers. First, the BIRD proposes adding a comment to the IBIS specification discussing the sign of VI table data; second, add checking to ibis_chk for the correct sign, using the algorithm tuned through reflector feedback; and third, add a note to the ibis_chk source saying a BIRD is required to change the source. Proposed changes 1 and 3 appear OK to group, with the additional comment that source licensees are free to modify their own source, but not to post the result. Only the ibis_chk czar (Jon Powell) can post modified versions of ibis_chk, and that requires a ratified BIRD to initiate. AR:Kellee, change the proposed ibis_chk notice to indicate that it is OK to change your own source code, but not the global one without a supporting BIRD. Action: Done Regarding change 2, there is controversy about the sign of the ECL example. Kellee pulled the example out of the IBIS V1.1 spec, which was a CMOS example, and indicated that if this was an ECL example then there would be a sign error because for ECL pulldown, VI is Vcc-V per BIRD 4. Kellee feels that right thing to do is to change the sign and use ECL type data and present a correct ECL table, but doesn't have ECL data. Kumar has ECL data which he offered to supply for inclusion in the BIRD. AR:Kumar, post ECL model to reflector, Kellee, use that data in the BIRD. Action: I have incorporated the model. AR Kellee, Two people noted that the directions of the current tests for the ECL were wrong. Kellee thought they were correct during the phone conversation. Action: I was wrong, and have changed them to correct the problem. ******************************************************************************* Changes April 15 meeting and inputs from Bob Ross, Interconnectix Inc. Will tables composed of only 0 mA entries or of only constant current entries be flagged as an error in Version 2.0? They are acceptable in Version 1.1, and could be useful to get around some Model_type omissions and do some (beyond IBIS) constant current source modeling or biasing of tables modeling. Input at the meeting 4-15-94 indicated that 0mA and constant current should be allowed. Change 2, section 4) states: If any table moves in the wrong direction report the following error message: 'Error found in xxx V/I table at line number nnn!'. Where xxx is one of the following: Pullup, Pulldown, POWER_clamp, GND_clamp. Where nnn is the line number. Note: It is acceptable to stop the parser after the first line found with this error." I could technically be very happy with the wording as is, and yet we may have entirely different expectations. To me "If any table MOVES in the wrong direction ..." means an error is NOT reported if there is NO table movement - e.g., NO error is reported if the tests reveal EQUAL currents. I believe you really intended "If any table verification fails ...". I would prefer that the IBIS_CHK parser not stop, but continue to check the file for all errors (including other occurances of this error). So I would delete the "Note" entirely. AR Kellee, Modify the test to allow constant current and 0mA as valid. Action: I have made changes in the V/I table tests to allow 0mA and constant current. AR Kellee, Clarify wording and remove requirment to stop IBIS_CHK on first error. Action: Modified wording to say "If any table verification fails..." Deleted the Note requiring the parser to stop after the first error line. *************************************************************************************