IEEE P1149.4 Mixed-Signal Test Bus Working Group Meeting Minutes for October 20, 1996 Generously sponsored by Panasonic Semiconductor --------------------------------------------------------------------------- Access the Minutes of the May, 1996 Working Group Meeting. --------------------------------------------------------------------------- Meeting Agenda Time Topic Responsibility 8:00am Continental Breakfast 8:00am Arrival and Introductions Adam Cron 8:15am Approval of May, 1996 Minutes Adam Cron 8:30am Number of Bits in Register Lee Whetsel 9:15am BSDL Proposal Ken Parker 10:00am Break 10:30am New Test Chip Design Effort Akira Matsuzawa Katsuhiro Hirayama 11:30am IC Test Progress Keith Lofstrom 12:00pm Lunch - Thank you Panasonic 1:00pm Draft Review Brian Wilkins 3:30pm Break 4:00pm Software for Test Chip C.J. Clark 4:15pm Officer Affirmation Vote Mani Soma 4:30pm Chapter 8.4: Switches, Vref, and ESD Steve Sunter 5:00pm Ballot Status Adam Cron 5:15pm Next Meeting at ITC Adam Cron 5:29pm Miscellaneous Adam Cron 5:30pm Adjourn Adam Cron --------------------------------------------------------------------------- Working Group Statistics Working Group Members 36 Total Subscribers 341 Total Subscribers on "esd" reflector 253 Countries Participating 32 Companies/Universities Participating ~214 Funds Available $876.79 --------------------------------------------------------------------------- Meeting Attendees Name Company Sponsor John Andrews National Semiconductor Terry Borroz Teradyne, Inc. John Braden Stratus CJ Clark Intellitech Bill Coombe Medtronic Adam Cron Motorola Dan Dandapani University of Colorado Craig Danes Guidant/CPI Ted Eaton Intellitech John Ford Silicon Systems Ren Franse Panasonic Semiconductor Yasuo Furukawa Advantest Allen Heiden Motorola Katsuhiro Hirayama Panasonic Semiconductor Terry Junge Seagate International Jake Karrfalt Alternative System Concepts Adam Ley Texas Instruments Keith Lofstrom KLIC Akira Matsuzawa Matsushita Colin Maunder BT Laboratories John McDermid Hewlett-Packard Math Muris Philips Naveena Nagi LogicVision Elbert Nhan Johns Hopkins University Kozo Nuriya Matsushita Adam Osseiran Ecole d'Ingenieurs de Geneve Ken Parker Hewlett-Packard Adam Sheppard ASSET InterTech Mani Soma University of Washington Steve Sunter LogicVision Tony Suto GenRad Jon Turino Integrated Measurement Systems Lee Whetsel Texas Instruments Harry Whittemore nCHIP Brian Wilkins Southampton University Tom Williams IBM Sending Regrets Name Company Steve Dollens International Microelectronic Products Michel Parot Thomson-CSF --------------------------------------------------------------------------- Arrival and Introductions Adam Cron and the WG thanked Panasonic for sponsoring today's meeting (approximately $52 per person for food and refreshments). Adam announced there would be a P1149.4 technology demonstration for interested people on Tuesday, 10/22/96, in the same room (Idaho) from 3 to 6 p.m. Introductions. Review meeting agenda. Keith Lofstrom was set to present two papers on Tuesday afternoon and Wednesday morning. Ballot: Invitation was not closed yet, and therefore more can still be added to the ballot list. Adam reiterated the voting policy: only WG members may vote. Approval of October, 1995 Minutes Adam received a motion to approve the 05/96 minutes. Seconded. Unaminous approval. Number of Bits in Register Before turning the stage over to Lee Whetsel, Adam said he would like to have a consensus in the WG on the number of bits in the Standard. The motivation for this proposal is to standardize the definition of a test cell in both 1149.1 and P1149.4. The analog test cell is similar to the digital test cell in 1149.1. The task is to come up with a definition without significantly changing the analog boundary module (ABM) with its digital functions already well defined. Viewgraph VG1: Example P1149.4 Test Architecture. Steve Sunter said there is no switch between the output of 3SOA and the output pad, looking at the internal structure of the analog test cell (ATC) on the viewgraph. Lee said P1149.4 interconnect testing includes the capability of testing wires, inductors and resistors, and checking capacitors for shorts. The P1149.4 interconnect test also covers high and low termination drive tests. The following viewgraphs show the various proposed analog test cells for 1149.1. Viewgraph VG2: 1149.1 2-State Analog Test Cell. This 1149.1 analog test cell will not support the full P1149.4 metrology since no tri-state is available, but it is compliant with all other 1149.1 instructions. Viewgraph VG3: 1149.1 3-State Analog Test Cell. This modified 1149.1 analog test cell supports full P1149.4 metrology. With the addition of a single-input analog output test cell (AOTC). Tri-state for 3SOA is enabled through the single-input AOTC. Viewgraph VG4: 1149.1 Bidirectional Analog Test Cell. This ATC is similar to the 1149.1 bi-directional digital test cells. Capture Select (CS) is used during Sample/Preload. This ATC is compliant with all 1149.1 instructions. Viewgraph VG5: 1149.1 Input Analog Test Cell. This viewgraph shows a proposed 1149.1 input ATC that is similar to the existing 1149.1 Capture-Only Digital Test Cell. Viewgraph VG6: 1149.1 Buffered Analog Test Cell for Signals Without Functional Output Amplifiers. Test output buffer (TOB) provides strong test drive and reduced test loading. Ken Parker asked how strong the buffer has to be. The identity of the pin might be different in the normal mode and the test mode, and P1149.4 assumes that. Lee would like to see it shown as an example in the Standard. Keith joined the discussion, saying we are arguing over something that is imaginary. Treat it as a black box (we should not care whether the implementation is a transmission gate or a buffer). Keith pointed out an amplifier/buffer has stability problems if it is not carefully used. Buffers are neither required nor precluded in the Standard. John McDermid asked if there is a V- for return. Lee said yes. He added that this is not for measurement, but for setting up test. John was concerned about the buffer having offset and stability problems. Sunter said we can only specify what can be measured off-chip, not what is inside. We are dealing with implementation details here! Viewgraph VG7: Use of Functional Output Amplifier Solves Low/High Drive limitations. This slide shows 1149.1 analog interconnect test examples. The output amplifier may be employed to handle 50-Ohm termination resistor and 470-Ohm pull-up resistor loadings. Again, whether a transmission gate or a buffer should be used is an implementation issue. The following viewgraphs deal with the various methods to upgrade 1149.1 chips and test cells to P1149.4. Viewgraph VG8: Upgrading a Chip from 1149.1 to P1149.4. To transform a chip from 1149.1 to P1149.4, all one has to do is add an ABD, AT1 and AT2 pins (to make a 6-pin TAP), analog measurement test cells (AMTCs), and internal AB1 and AB2 bussing. Viewgraph VG9: Upgrading 1149.1 2-State Analog Test Cells to P1149.4. To add tri-state feature, replace the amplifier with a 3-state amplifier, add an analog measurement test cell and a high-Z control. Ken asked what the difference is between a 4-state cell proposal back in 1993 and this one. He would like to see separation of analog and digital test cells. A 3-cell structure is probably more difficult to implement than one with 4 cells. Ken said the elegance of separation is that it appeals to 1149.1 WG, and is easier to convince those folks. Adam asked what then if we have a stability problem with a 3SOA? Lee said this is a design/engineering issue. Lee wanted to keep the cell as simple as possible and assume good engineering practices. Viewgraph VG10: Upgrading 1149.1 3-State Analog Test Cells to P1149.4. One only has to add an AMTC in this case. Viewgraph VG11: Upgrading 1149.1 Bi-Direct Analog Test Cells to P1149.4. This scenario only requires the addition of an AMTC. Viewgraph VG12: Upgrade 1149.1 Buffered Analog Test Cells to P1149.4. Simply add an AMTC. Ken commented that there is a design constraint on the buffer. The following viewgraphs present upgraded 1149.1 measurement examples: Viewgraph VG13: Once upgraded from 1149.1 to P1149.4, when measuring a resistor, disable 3SOA using an AOTC control; output V- from AITC's 3SOA to sink Idc; force a known Idc through the unknown R to V-; then close the AB2 switch of the output to measure V2 and close the AB2 switch of the input to measure V1. John said some devices don't come with ground. Terry Junge said a power pin can be used instead of a ground pin. If test cells can truly be shared between 1149.1 and P1149.4, things will be simplified significantly. Steve Sunter asked if there is anything in P1149.4 that would preclude the implementation of this scheme. Lee said he wasn't certain. Steve said as the draft stands, nothing prevents us from using this scheme. Adam Cron wants the number of bits mandated so that Brian Wilkins can write it into the draft. Brian asked if 4 bits are now mandated. Viewgraph VG14: Required Step to Standardizing Analog Test Cells Between 1149.1 and P1149.4. A P1149.4 ATC must be partitioned into 2 cells (1149.1/P1149.4 ATC and P1149.4 AMTC). A 1149.1/P1149.4 ATC satisfies 1149.1's input, output and Input/output interconnect test requirement. A 1149.1/4 ATC in combination with P1149.4's AMTC support full P1149.4 measurements. Viewgraph VG15: Obstacle Confronting Development of Standardized Analog Test Cells between 1149.1 and P1149.4. To partition 1149.1 and P1149.4, it is necessary to eliminate decode and add another cell. Adam said he wants a motion on the number of bits. Ken suggested to wait until end of the day to vote on it so that other parties may present their materials and viewpoints that may have a bearing on the outcome of the vote. Steve asked what the advantage of this scheme as a user (chip designer) is. Lee said 3-bit will not allow partitioning but 4-bit will. One advantage is to get 1149.1 and P1149.4 on the same page. It would simplify matters considerably. Lee reemphasized that standardizing is the prime objective of his proposal. BSDL Proposal BSDL was rewritten in 1993 (with lots of pain and grief). If standardization is dealt with and handled properly, it would make life infinitely easier for software. The issue here is more on what it takes to support the Standard (to create a viable standard), and not so much on the standpoint of saving silicon and so forth. In a standard such as P1149.4, software is equally important as hardware. Viewgraph VG16: BSDL Changes to Support P1149.4. In the current draft, there is presently nothing on 3-bit vs. 4-bit. From the software standpoint, the number of bits is important to know and should be mandated. If software development costs can be amortized over a large number of people, that would be ideal. People rely on the standard. The software needs to know what to do for a standard. It needs to know how to get data from measurements and process it. We must provide the software designer with rock-solid definitions and language to create useful and meaningful test software. BSDL is a subset of VHDL. BSDL, until a few years ago, completely ignored analog pins. It treated analog pins, no-connects, power, and ground pins as "linkage" pins (essentially "garbage" pins that were ignored). One approach is to include analog pins and then figure out how to handle them later. The P1149.4 Standard has new terms that require new definitions. There is probably even a need to add new functions; anything but optioning the new features in P1149.4. The reason is that the optional features will grow and eventually become excessive as will the software. If an ABM is an amorphous object in terms of not having the number of bits specified, then there will be problems. It doesn't matter if it is 3- or 4-bit, we need to nail it down. Between Keith and Panasonic, we will hear about an experience concerning real silicon test chips. A question was raised about whether BSDL was written only because the number of bits was mandated in 1149.1. The answer is yes. Whenever you add an option, the complexity of the software correspondingly increases. If there was no limit to the number of bits, then a viable and practical standard cannot be written. In Chapter 10 of 1149.1, it looks complicated for software but is actually implementable. Ken proposed that we identify each flip-flop with a purpose and a name. If pins cannot be identified, then it is impossible to produce software. Ken wanted to make a short list that will solve all the important problems. There is also global options vs. per-pin. A logical question is: if VHDL has a way to handle this, should we use it for BSDL? Ken said he had bad experiences with it. A question was raised about the 1076 and whether we need to take a look at it. Adam asked if we should go to ballot without BSDL. Ken said this would be better answered later. Viewgraph VG17: For P1149.4, there will be new function definitions. A way to identify new function control. Ken said if we followed the elegant way to extend 1149.1 into P1149.4 , then the software should follow very quickly. Viewgraph VG18: How to describe pin-to-pin impedances. In the Standard, guidelines will be provided, but at what point should we say the user is allowed to do this but not that. As an example, when measuring resistors from pin to pin, what if a FET is in between? What about global options? Are they all-encompassing? Are there any exceptions? It is evident that this is a complex issue. With that, the floor is open for discussion. Software has to be written in a very concise way and the standard has to reflect that. Reviewing the draft is important for people such as Panasonic and Keith. A lot of people out there will have only a draft to look at. Therefore, it has to be unambiguous. Keith pointed out that we can't restrict too much on the mission circuits. There are people out there that still leave pins out of 1149.1 chips and we have to tolerate it. The same thing is predicted to happen in P1149.4. The Standard says to do it a certain way, but the market determines the winners and losers. A motion was on the table: We will go to ballot prior to defining BSDL for P1149.4 Standard - Yes or No? Seconded. Discussion followed. It took 3 years to write BSDL and no one now cares anymore. Modified motion by adding the word "formally" before "defining ....." Brian said the real reason for the delay of the draft is nobody devotes full time to it. MOTION: We will go to ballot prior to formally defining BSDL for P1149.4 Standard. Seconded. Yea: 16. No: 0. Abstained: 0. Unaminous approval. MOTION: We will define a BSDL for P1149.4 Standard. Seconded. Yea: 17. No: 0. Unaminous approval. Before going to ballot, we need to at least clearly define the number of bits. We can write BSDL today, if we wanted to, that will do interconnect tests. Balloting will sound an alarm to the industry that a standard is being established and will probably lead to more critical reading of the draft that will result in feedback and data for the WG. Adam asked if we can write BSDL for draft 10 as it stands today. The answer is no. New Test Chip Design Effort Viewgraph VG19: The complete specifications for the MEI test chip (MNABST-1) were distributed in handouts. The objective is to evaluate the analog test bus structure with various switches. This chip is fully compatible with 1149.1. Only basic functions (Bypass, Sample/Preload, EXTEST) are implemented. Viewgraph VG20: Shown is the ETAP controller architecture for the test chip MNABST-1. Viewgraph VG21: The analog boundary module is shown. The diodes at the pin and core are ESD protection diodes. Viewgraph VG22: A photograph of the evaluation board. Viewgraph VG23: Matsushita tested the chip using BSDL and test patterns (which were distributed on the reflector by Ken) provided by HP on an HP3070 tester. The test patterns were successfully used to test the chip. Viewgraph VG24: Why was MNABST-1 implemented using 0.35 micron process? Because it is a conventional process and Matsushita expected it to be the most widely used process. In addition, 0.35 micron CMOS serves as the base for future deep quarter-micron process and therefore there are merits to experiment with the 0.35 micron. Experiences gained from working with the 0.35-micron can be applied to the quarter-micron process. Viewgraph VG25: Shown is the complete chip layout with the various metal layers. A question was raised about the number of metal layers implemented. The answer is 3 for now, 5 in the future. Viewgraph VG26: Size estimation. By optimizing the chip layout for the differential comparator (VG25) resulted in a space saving of 8% using .35-micron CMOS process for one ABM. When optimizing, 100-Ohm resistance value was selected out of 4 possible values. There are layout restrictions for an ABM. In a deep submicron process, the metal conductor resistance increases because of the decrease in width. This can possibly affect measurement results. The total path resistance is 43 Ohms. In short, the guideline for layout must be observed, i.e., the pad-ABM interconnection has to be laid out as short as possible. Steve said there is no need for these guidelines if using 2 wires. For P1149.4, the cost of wiring will be higher than for 1149.1. Viewgraph VG27: Distribution of evaluation board, BSDL, and specifications. The test chips will be available only to P1149.4 WG members free of charge. Interested individuals should fill out a registration sheet provided by MEI. Forty sets are available at present. Test chip recipients are requested to report any results to P1149.4 WG. For requests from within the U.S., contact Hewlett Packard. Test chip requests will be filled for requestors outside the U.S. by MEI. One condition attached to the receipt of these test chips is that they are solely for P1149.4 evaluation purposes. Viewgraph VG28: The on-resistance of a 120 Ohm analog switch is 70 Ohm. For 2 KOhm resistors, the on-resistance is 1.6 KOhm. For N- and P-channel switches, the on-resistance increases as a function of the gate voltage. The resistances increase to 120 Ohms and 11 KOhms, respectively, at a voltage of 2 V. The gate on-conductance is maximum at 0 V gate voltage and is minimum at about 2 V. Fluctuation of switch impedances at low voltage levels is an important issue. A thorough understanding of the behavior of the on-resistance at low voltages is required to avoid obtaining erroneous or extraneous measurement results. Viewgraph VG29: Distortion measurement of the core-disconnect switch. The distortion is approximately 1% for 8-bit accuracy and 0.01% for 16-bit accuracy. The point here is to select switches with appropriate distortion levels. The designer should choose several types of switches. The offchip load impedance is 1MOhms. Viewgraph VG30: Cost of Boundary Scan Test. This is a table comparing the cost factors between 1149.1 and P1149.4. For 1149.1, the cost is minimal and the TAP controller and DBMs can be made small in submicron technology. However, P1149.4 requires large wiring area which can be reduced using multi-metal layers. In short, P1149.4 will require more silicon and may cost more than 1149.1. Ken said that the chips have been available for 2 months now. There was a demonstration of the MEI chip on HP tester at the HP booth at the ITC. The demo performed indicated that 1149.1 interconnect tests work with 1149.1 BSDL. The HP P1149.4 exhibit at the show included demonstration of measurement techniques for a 1.5-mH inductor (to an error of less than 1%) and a 10% capacitor. Also, measurement of a 3-resistor delta was performed. Parasitic impedances caused the measurements to be a little high. There were also AC measurements. The demo used 1.6-KOhm switches. With this arrangement, very low resistance values can be obtained. The objective here, again, is to test out the P1149.4 metrology. Ken invited everyone to stop by the HP booth. More experiments were to follow. Adam thanked MEI/Panasonic for the test work. IC Test Progress Keith will have presentations this week. Keith used 1.2-micron process (but have 1.5-micron cells). The ETAP is "crummy". The emphasis was on time and not so much on creating a perfect chip. Drifts during test: The conventional wisdom that one can obtain more accuracy in test if averaging is used is not necessarily true. On a blank viewgraph (VG31), Keith illustrated the variation of switch resistance with time. There are stored charges in silicon that move around which can affect switch resistances. Averaging can actually make it worse since there is more time for resistance to drift. Averaging can mask random noises but cannot shield drifting resistances. Thermal effects along with mobile charges and other factors are responsible for resistance drift in MOS. Keith reported a 0.3% measurement error with not-so-great equipment. One can run production types of tests with PCMCIA and a laptop, which is great for field engineers. Keith mentioned he came across P1451.2 standard which basically is a smart transducer interface for sensors & actuators. Stan Woods of HP is the contact if more information is desired. Steve Dollens' demo board hasn't been scanned through yet. The software for interfacing with the board is available on the SPAsystem. More polished software may be purchased from Intellitech. If anyone is interested in experimenting with Keith's software, then get his. Draft Review Draft 10 had been e-mailed to the reflector. We should base the draft on the most current revision of 1149.1 and thus we should remove the "P1149.4 must follow 1149.1" phrases. A better way to say it is "we are consistent with 1149.1". Lots of changes have been proposed to 1149.1 but they have not been written in yet. Whenever 1149.1 is updated, P1149.4 will also be. Ken mentioned he sent out an e-mail last week with philosophical statements about P1149.4 and 1149.1 compatibility (i.e., if protocols are compatible, then the same algorithms may be adopted). Section 1.2 addresses that. But Brian wanted to address the issue that if a chip is 1149.1 compatible, does it mean all rules of 1149.1 apply to it? The bottom line is: Brian will synchronize the P1149.4 draft with the latest version of 1149.1 instead of "known changes" that WILL be made to 1149.1. No motion on this discussion is necessary. A motion was initiated by Lee Whetsel on having an ABM consisting of 2 distinct cells -- digital and analog. Ken said 1149.1 has a number of ways to deal w/ bidirectionality. Brian said we are removing requirement to use 3 bits (cells) with the option of having an unlimited number of bits. The next motion will focus on defining the function of each bit. Colin said 1149.1 only specifies functions, not implementation details. Debates on whether it is important to define bit functions ensued. Ken said 3 bits vs. 4 bits means measurable overhead and likes the separability of 1149.1 & P1149.4 cells. More debates followed. Steve said AMODE is an option whether we have 3 bits or 4 bits. ken said 3 bits, yes, AMODE is required because 3 bits plus AMODE will be necessary for all 11 states. Some spare states are available for production IC parameter testing. Control cells cannot be shared in P1149.4. Instead, P1149.4 needs dedicated controls. Brian urged to close this discussion so that the motion can be voted on. MOTION: The P1149.4 analog boundary module will comprise 2 separate cells: 1. A 2-bit 1149.1 and BSDL compliant analog or digital test cell capable of controlling a pad wire (during EXTEST) to Vh, Vl, and Z, and observing the pad wire. 2. A 2-bit analog measurement cell capable of connecting a pad wire to AB1, AB2, or AB1 and AB2. Seconded. Yea: 15. Nay: 0. Abstained: 0. Unaminous approval. Terminology: Ken said in P1149.4, we call a group of cells analog boundary module. Colin said basically we are using 1149.1 DBM to describe something not existent in 1149.1. The word "cell" raises a lot of confusion. An ABM has both a digital and an analog section. Adam asked if we can delete Section 8.2 in the draft. This is an editorial issue. Adam suggested to adhere with the term "DBM" and proceed onward. Steve disagreed saying that it would create a "mess" if we equated a DBM to a digital boundary cell. More on terminology: Core circuit vs. system circuit. The only concern is that the circuit is not just digital but analog also. Colin would like to see consistent terminology in both 1149.1 and P1149.4 eventually. Thus, use "core circuit", "core disconnect" as in 1149.1. Another issue: Differential signals (John Andrews). These signals exist in analog, digital and mixed-signal circuits. Brian drew on a viewgraph (VG32) a diagram of 2 chips interconnected with ATC's. Scenario: If we have to test the recommended ABM's (denoted by "R"), should we use the PROBE instruction. Colin suggested 2 types of EXTEST instructions, one for digital and the other for analog. Adam asked if anyone wanted to make a suggestion on the optionality of the recommended register and instruction. In a sense, it could be treated as an EXTEST. In 1149.1, each digital differential pin is a digital pin. However, in P1149.4, the mandated cells take precedence over the recommended cells. The fact remains that we need to consider and take into account testing in noisy environments. Steve said we have resolved this issues before. The Ms on the viewgraph represent ABMs. It was suggested to form a subgroup that focuses on the issue of whether to have DBCs for the Rs in the diagram. Adam recommended that Steve Sunter and Brian Wilkins get together and decide what they should be (ABM or ABC, etc.). The bottom line is to standardize the terminology. Instruction issue: Lots of disagreements here. What do we mean by INTEST? Do we need to have it? Lee said yes. 1149.1 has INTEST. Is INTEST worth defining? Lee said yes. If so, what does it mean then in a P1149.4 chip? Ken asked if the INTEST instruction is loaded, should core disconnect be open or closed? When open, the ABMs on the pad side are disconnected from the mission circuitry. Maybe we need to caution that INTEST is really defined only in 1149.1 and one should be careful every time INTEST is loaded to make sure the chip does not get damaged. INTEST has to be defined unambiguously because of the digital and analog interface. Adam asked if anyone wanted to work on the INTEST issue. Subsequently, a subgroup consisting of Lee Whetsel, Keith Lofstrom, Steve Sunter will hash out the topic of INTEST. Core disconnect (CD) on input pin: Do we need to mandate CD on input pins. Steve said we have not mandated a physical switch there but do require anything that happens outside chip will not affect chip. From the outside world, one cannot tell if there is a switch in the chip. Keith suggested to always show an explicit switch to avoid confusion. Lee said an output pad does require a conceptual switch but an input pad does not. Lee is not comfortable with Figure 15 in the draft and will work with Brian on it. Steve does not have any new presentation on current/voltage buffers. If there is a current on AT1, a voltage on AT2, how do we handle that? This issue was deferred to e-mail discussion. Steve said it is more efficient to debate over the e-mail and then vote on proposals in WG meetings. Brian said it would help if some other people would also contribute. CJ preferred to discuss the proposals complete with pictures and diagrams in WG meetings. Adam said he can put pictures on the web if necessary. Ken referred to Figure 10 on pages 9 to 11 in the MEI test chip specifications document. He suggested to consider this concept of an AT port. Here is a physical implementation that works and should be a valuable addition to the Standard. There are currently no measurement instructions. EXTEST can do the basic tests, and if complex testing is required, then we add options (e.g., for bipolar). The consequence is that different flavors of EXTEST (e.g., measure) can be invented for various purposes. Ken started the discussion of an AT port having a boundary scan cell. Steve said at one point we had that but it somehow dropped out. EXTEST connects TDI and TDO but also AT1. Pages 9-11 of the MEI specifications document show one implementation that potentially has some good capabilities. Action Item: Ken will help Brian make the 4-cell scheme more consistent. Anyone interested in reading PostScript format should visit this site: http://www.cs.wisc.edu/~ghost/. This web site contains downloadable software that interprets PostScript, GhostScript, GSView, etc. Software for Test Chip CJ thanked Lee for pushing for a 4-bit ABM. A 4-bit ABM would really help today. He has been working on demo software along with Ted. CJ had problems with tables 2 and 3; working with 1149.1 ASSET Software. In this situation, one cannot just take a regular 1149.1 tool and do whatever desired. The software needs to be modified. The bottom line is that there was a major change in today's meeting -- 4-bit ABM. Lee solved the problems. CJ was to demo the software at the show. CJ wirewraps his own board for use in testing out his software. Regarding the demo board issue: It was interpreted in a different way. It should be noted that CJ's demo software will not work with Steve Dollens' board. CJ said the boards should just be given to WG members. His software will work with 3-bit or 4-bit cells. Officer Affirmation Vote It is ballot time. Nominations for Chair, Vice Chair, Editor, and Secretary are the incumbents. A motion was initiated to approve the elections for 1997: * Adam Cron, Chair * Steve Sunter, Vice Chair * Brian Wilkins, Editor * Elbert Nhan, Secretary MOTION: The above officers for P1149.4 Working Group for 1997. Yea: 13. Nay: 0. Abstained: 2. Motion carried. Chapter 8.4: Switches, Vref, and ESD Steve had sent Brian's request for ESD requirements over the e-mail. Viewgraph VG33: Electro-static discharge protection. In the diagram on VG33, Rcom cannot be cancelled out or calibrated for measurement. Rcom represents the 43-Ohm path resistance on the MEI/Panasonic test chip. When forcing a current through AT1 and AB1 and monitoring the voltage at AB2, assuming a small current through resistors, then accurate measurements can be obtained. We should make sure Rcom is less than 1 Ohm for 1% accuracy. If this structure was connected to another similar structure, then would the accuracy of the reading be 2%? The answer is yes. Rcom depends on the output driver impedance. This rule is very design dependent. MEI thinks this scheme is reasonable. This Rcom is the "undocumented" resistance. If any more impedance is added, we would need to document it (this was agreed upon and was one of the results from the switch subgroup meeting). Draft D10 does not have any implementation details. However, some examples will be included. Mani asked continuous technology improvement will demand the requirement to be less than 100 uA. The total resistance from AT1 to the pin should be limited to 10 KOhms. Rload should be than 1 KOhms. These are just wire resistances, not physical resistors. Lee pointed out this is midband model. If a resistor is in series, that has to be documented. If the path resistance met specifications, it does not have to be documented. Otherwise, it would have to be. If Rcom can somehow be bypassed, then no documentation is necessary. Adam Ley summarized the main point of this discussion: The rule is such that if the path resistance does not meet the specifications as given, then it must be documented. One ESD resistor is needed for driver, and another for the entity being driven. Adam Osseiran asked if we are talking about CMOS here. The answer is no. Ken added that this is for the general case for ESD which is independent of technology being employed. Viewgraph VG34: Switch limitations. Debates and discussions resulted in the following modified form of switch rules: Rpin is impedance measured at the func. pin due to pin's function driver or receiver. Rsw includes the TBIC switch and is measured at the func. pin while AT1 is driven to any DC voltage between Vmax and Vmin. Rules: * (a) When Tx gates are used, the max resistance (2 switches in series) between AT1 pin and func. pin (Rload) shall be: Rsw = 50 x Rpin, or 10 KOhms, whichever is less. * (c) Minimum current drive of AB1 and AT1, when current buffers are used: Imax = (Vmax - Vmin) / Rsw * (a) and (c) are complementary. * (b), (d), (e) leakage and offset limits: TBD. * (g) The characteristics and functions of a switch shall not be altered if a function pin is connected via a short circuit to any voltage between Vmax and Vmin for any period of time. This applies to on-chip only, but not on-board. This does not include bondwires (.1 Ohm). We can document our way around the Rcom but still have to adhere to these rules. Ken would prefer not to have too many restrictions placed on the switches that may turn out to be unnecessary or even unrealistic. However, there has to be some kinds of limits. If a 1 Ohm driver is to drive a 1 Ohm load offchip requiring a 0.1 % accuracy, then the switch resistance, as a rule of thumb, has to be 50 times smaller. Lee wanted to simplify this diagram but this represents actual situations (refer to VG33). Action Item: Steve will take this offline with the switch subgroup on the contents of VG34. As for the materials in VG33, Brian will attempt to word it. Terry Junge suggested a guardband by reducing the tolerance to less than 1%. That is, set the overall impedance to be less than 1%. In this way, there does not have to be an automatic built-in 1% error. Several WG members concurred with this point. To view postscript files on screen, visit the WWW URL mentioned earlier. Adam Cron will e-mail to the reflector more information. Brian will send the draft out in postscript and in WordPerfect format. Adam will also accommodate with the request that ASCII files be made available. Ballot Status The current head count is 32. Invitation is still open. Just send e-mail to Adam Cron if interested in being a balloting member. Adam showed a list of the people currently in the balloting group. There were several omissions that will be addressed. Adam also showed a list of non-eligible voters that don't have membership. Ken said we will need at least 100. Other people agreed. Adam said he can send out another invitation to solicit more balloting members. Mani asked when the WG goes to ballot. Adam said it was originally scheduled to be last June. CJ pointed out that 1394 had just gotten approved after 10 years of standard development. Adam said an EISA Draft Standard he was just handed was dated 1990. Dan Dandapani suggested sending invitation to subscribers. Next Meeting February time frame was suggested. Possible sites and dates were proposed. Bill Coombe will check if Phoenix is possible in February. He will also investigate the possibility of a Motorola meeting site (Al Heiden). The meeting format was tentatively agreed to be 2 half-days. Miscellaneous Ken Parker commended Brian on his work. The Editor position is labor-intensive. The draft needs to be clear and unambiguous. Ken thanked Brian for his diligence. CJ missed the BSDL discussion because of being late to the meeting this morning. He has already started doing some BSDL and has software for a 3-bit structure. Ken, saying it is an important issue, will talk with his management at HP and ask that he be able to devote more time to BSDL. Lee asked how the test chip people feel about the progress? Ken said for the Panasonic chips, large impedances were employed to drive highs and lows. There is probably a need for more delay in the capture time. The pin capacitances were not a major problem. Adam asked if anyone will do attempt the measurement of the parameters suggested by Steve Sunter in the meeting. So far, there are no volunteers. Regarding the next meeting dates. There was a general consensus on the meeting date not to fall on Valentine's day or Presidents' Day, or any other holiday, for that matter. The actual meeting dates and site will be determined over e-mail. Lee suggested we need to target system engineers, not just designers since system designers do influence chip designers. Adjourn Seconded. Unanimous approval. Meeting officially adjourned. --------------------------------------------------------------------------- Access the Minutes of the March, 1997 Working Group Meeting. --------------------------------------------------------------------------- To reach the Chair of the IEEE P1149.4 Working Group: Adam Cron Motorola 50 East Commerce Drive, Suite M5 Schaumburg, Illinois 60173 USA Phone: (847)576-3092 Fax: (847)538-4801 E-mail: fac601@email.mot.com