Minutes of SPI-2 LVD SCSI Working Group Meeting -- 1/29-30/96

Lohmeyer, John JLOHMEYE at cosmpdaero.ftcollinsco.ncr.com
Mon Feb 19 11:15:00 PST 1996


* From the SCSI Reflector, posted by:
* "Lohmeyer, John" <JLOHMEYE at cosmpdaero.ftcollinsco.ncr.com>
*

Minutes of SPI-2 LVDS Working Group Meeting                   X3T10/96-122 
r0

Accredited Standards Committee*
X3, Information Technology
                                                   Doc. No.: X3T10/96-122r0
                                                       Date: February 4, 
1996
                                                    Project: 1142-D
                                                  Ref. Doc.:
                                                   Reply to: John Lohmeyer

To:         Membership of X3T10

From:       John Lohmeyer, Chair X3T10
            Bill Ham, SPI-2 Technical Editor

Subject:    Minutes of SPI-2 LVD SCSI Working Group Meeting
            Denver, CO -- January 29-30, 1996


                                       Agenda

1. Opening Remarks

2. Approval of Agenda

3. Attendance and Membership

4. Releasing Bus from Active Negation [Uber]
   4.1 Digital Filtering of  Negation Release Glitches [Lohmeyer]
   4.2 Symmetric vs. Asymmetric Drivers
   4.3 Timing Budget [Jander]

5. Three-Resistor Terminator Equations [Haynes]
   5.1 Additional Termination Issue (96-125) [Aloisi]

6. Transmission Line Model (96-123 & -124) [Gingerich]

7. Voltage Mode Drivers [Bridgewater]

8. SPI-2 Document Review (X3T10/1142D) [Ham]

9. Test Data (96-127) [Ham]

10. Standing Waves [Bridgewater]

11. Meeting Schedule

12. Adjournment




                              Results of Meeting

1.    Opening Remarks

John Lohmeyer, the X3T10 Chair, called the meeting to order at 9:00 a.m.,
Monday January 29, 1996.  He thanked the host, Symbios Logic, perhaps too
profusely.

As is customary, the people attending introduced themselves and a copy of 
the
attendance list was circulated.

2.    Approval of Agenda


3.    Attendance and Membership

Attendance at working group meetings does not count toward minimum 
attendance
requirements for X3T10 membership.  Working group meetings are open to any
person or organization directly and materially affected by X3T10's scope of
work.  The following people attended the meeting:

         Name          S        Organization         Electronic Mail Address
 ---------------------- -- ------------------------- 
 -------------------------
Mr. Norm Harris        P  Adaptec, Inc.             nharris at eng.adaptec.com
Mr. Wally Bridgewater  V  Adaptec, Inc.             wally at eng.adaptec.com
Mr. Richard Moore      V  Adaptec, Inc. 
            richard_moore at corp.adaptec
                                                    .com
Mr. Dennis R. Haynes   O  Burr-Brown Corp.          haynes_dennis at bbrown.com
Mr. Justin McEldowney  V  Burr-Brown Corp. 
         mceldowney_justin at bbrown.
                                                    com
Mr. Louis Grantham     P  Dallas Semiconductor      grantham at dalsemi.com
Dr. William Ham        A# Digital Equipment Corp.   ham at subsys.enet.dec.com
Mr. Dan Colegrove      A# IBM Corp.                 colegrove at vnet.ibm.com
Mr. Richard Greenberg  V  IBM Corp.                 richg at vnet.ibm.com
Mr. Dean Wallace       P  Linfinity Micro 
          75671.3443 at compuserve.com
Mr. Ting Li Chan       A  QLogic Corp.              t_chan at qlc.com
Mr. Richard Uber       V  Quantum Corp.             duber at tdh.qntm.com
Mr. John F. Fobel      O  Rancho Technology, Inc.   scsi at tstonramp.com
Mr. Gene Milligan      P  Seagate Technology        Gene_Milligan at notes.
                                                    seagate.com
Mr. Dave Guss          A  Silicon Systems, Inc.     dave.guss at tus.ssi1.com
Mr. Vit Novak          A  Sun Microsystems, Inc.    vit.novak at sun.com
Mr. John Lohmeyer      P  Symbios Logic Inc. 
       john.lohmeyer at symbios.com
Mr. Kevin Bruno        V  Symbios Logic Inc.        kevin.bruno at symbios.com
Mr. Kelly Bryant       V  Symbios Logic Inc.        kelly.bryant at symbios.com
Mr. Brian Day          V  Symbios Logic Inc.        brian.day at symbios.com
Mr. Frank Gasparik     V  Symbios Logic Inc. 
       frank.gasparik at symbios.com
Mr. Mark Jander        V  Symbios Logic Inc.        mark.jander at symbios.com
Mr. Richard Mourn      V  Symbios Logic Inc. 
       richard.mourn at symbios.com
Mr. Tracy Spitler      V  Symbios Logic Inc. 
       tracy.spitler at symbios.com
Mr. Kevin Gingerich    V  Texas Instruments         4307725 at mcimail.com
Mr. Paul D. Aloisi     P  Unitrode Integrated       Aloisi at uicc.com
                          Circuits
Mr. Tak Asami          A  Western Digital           asami at dt.wdc.com
                          Corporation

27 People Present

Status Key:  P    -  Principal
             A,A# -  Alternate
             O    -  Observer
             L    -  Liaison
             V    -  Visitor

4.    Releasing Bus from Active Negation [Uber]

Dick Uber presented some simulation data (96-126) regarding the release
glitches (active negation to high impedance).

Dick also included data on glitches that can occur on an assertion to high
impedance transition.  There glitches are different from the "wired-OR"
glitches already documented in SPI because they can occur when only one 
device
is driving the bus.  Bill Ham noted that SPI-2 already requires all 
non-wired-
OR signals to be negated for at least a Bus Settle Delay before being 
released
and therefore the point becomes moot.

Dick also had some data that indicated attenuation puts a limit on how
asymmetrical drivers can be.

He had built a cable attenuation model for the REQ/ACK pulses that run at 40
MHz.  Simple skin effects and relavant harmonics up to 440 MHz were
considered.  With a worst-case cable and symmetric drivers, 32 mA drivers
would be needed.  With worst-case cable and asymmetric drivers, an 8/4 mA
driver would be required.  Dick's `worst-case' cable consisted of 30 Gage
single-strand wire, 110 ohms minimum line impedance, and 12 meters long. 
 The
skin effect is stronger where there is less "skin".  The use of non-stranded
wire provides the minimum surface area for a given copper cross section and
makes stranded more appealing for high frequency uses.  It will be important
not to restrict the lengths of cables arbrtrarily

      4.1   Digital Filtering of  Negation Release Glitches [Lohmeyer]

John Lohmeyer spoke about the message he had sent to the SCSI reflector
identifying where thenegation release glitches would occur.  He said that he
believes the negation release glitches can be filtered digitally should the
group agree on an analog LVD SCSI solution that produces these glitches. 
 SPI-
2 wording will need careful review to minimize the places where negation
release glitches would be permitted and to clearly identify where digital
filtering would be required.

Later in the meeting, it was concluded that negation release glitches may
still occur.  It will be necessary to specify release glitch filtering in 
SPI-
2.  It is also important to consider if the present requirement for actively
negating all assertions for a bus settle delay has any protocol impact.

      4.2   Symmetric vs. Asymmetric Drivers

Earlier in the month, the SPI-2 LDV SCSI working group started investigating
using asymmetric drivers (weaker negation driver than assertion driver). 
 This
was in part because some people believed this would help eliminate the
negation release glitches (false assertions when going from active negation 
to
released).  The SCSI working group had recommended that analog people not go
to asymmetric drivers just to remove these glitches -- they believed that 
the
glitches can be filtered digitally.  Most of the meeting danced around 
whether
there are good analog reasons to switch to asymmetric drivers.  The most
important reason that evolved was that considerable power can be saved in 
the
negation driver.  This also would deliver a more symmetric signal to the
receiver, which would help with setup and hold times.

The eventual solution permitted either symmetric or asymmetric driver
implementations.  However, asymmetric drivers should consume less power.

      4.3   Timing Budget [Jander]

Mark Jander asked that the setup and hold time budgets be re-visited based 
on
the changes to the driver specifications.  Bill Ham said that such 
discussions
really need a setup and hold time measurement figure like the one in Annex B
of Fast-20 in order to be useful.

Bill drew part of a figure that represented the receiver setup and hold 
times.
He started to draw a similar figure for the driver and ran into some
discussion from folks who thought the driver should have different critical
voltage levels from the receiver.  This indicated that some folks had not
carefully considered how the Fast-20 timing diagrams were derived.  In the
Fast-20 case timing allowance was made for signal degradation during its
journey to the receiver.  For example the pulse widths at the driver are
required to be significantly longer than at the receiver.  Actually this
timing difference was first introduced in the SCSI-3 SPI document.

Bill reasserted that this model should be continued for LVD but that the
voltage levels would need to be adjusted for the LVD receiver case.  There 
is
no voltage level at the driver that can be used as a reference other than 
the
receiver switching levels Bill claimed.

In a related discussion, Gene Miligan querried whether there was a 
requirement
that the signal pattern used during timing measurements be regular (i.e.
alternating assertions and negations of equal length).  Bill Ham noted that
there is presently no requirement on the data pattern for SPI or Fast-20 and
that this hole that should be plugged in SPI-2.  Bill suggested a model 
where
a regular pattern is used for the timing measurements for the driver and 
where
the receiver timings would include the cable plant and intersymbol
interference timing budget as well as that needed by the receiver silicon.
(The intersymbol interference arises when non-regular pattern are used.)
There may need to be some requirements on the driver concerning how much
deviation from "regularity" is allowed.

Bill took the action item to proposed a more detailed timing measurement
scheme for LVD for the March meeting.

5.    Three-Resistor Terminator Equations [Haynes]

Dennis Haynes presented a foil describing the derivation for his SCSI
reflector message describing a three-resistor terminator.  The conclusion 
was
that using a single voltage applied directly to one side of a three-resistor
totem pole stack terminator (with the other side connected to local ground)
will not meet all of the LDV SCSI requirements.  Such terminators will need 
to
have two non-ground voltage levels, one applied to each side of the resistor
stack, to meet all of the specification requirements.  This does not imply a
complication for the distribution of TERMPWR for LVD terminators but it does
imply that LVD terminators will need two voltage levels that are derived 
from
the TERMPWR line (or a local source) and that "active" or "regulated" LVD
terminators will generally be needed.

      5.1   Additional Termination Issue (96-125) [Aloisi]

Paul presented colorful slides (96-125) describing a potential problem if a
terminator were designed with a fairly high common-mode resistor of 500 ohms
and if the driver is capable of sourcing 24 mA (which is permitted in the
EIA/TIA standard).   This could result in a 12 v pulse when two drivers are
turned off at the same time, which would likely be harmful to the drivers 
and
receivers.  Paul proposed that the common-mode resistor range be specified 
to
be in the 100 - 150 ohm range instead of today's 100 - 500 ohm range.

Wally Bridgewater was concerned about some of the terminators shown during 
the
discussion.  They appeared to be quite different loads as seen by the 
drivers.

Kevin Gingerich noted that having two drivers on at the same time would have
the effect of adding another driver to his model.  The impedance seen would
depend on the driver type.  The voltage spike problem is worse with voltage-
mode drivers because they have a much lower impedance.  It may be necessary 
to
specify a maximum current requirement to limit this problem.  The other 
factor
that affects this problem is the amount of common mode voltage on the cable.

After considerable more discussion, the specification on the common-mode
resistance was adjusted to be 100 - 300 ohms.

6.    Transmission Line Model (96-123 & -124) [Gingerich]

Kevin Gingerich presented a model for LVD SCSI (see 96-123).  It includes 
both
AC and DC models and assumes a lossless cable.  Kevin used the solver 
feature
in Excel to optimize several parameters.  From the results, Kevin prepared 
96-
124, which proposes how SPI-2 should be modified to use his results in
specifying the driver and receiver characteristics.  This document was
reviewed under agenda item 8.

7.    Voltage Mode Drivers [Bridgewater]

Wally asked that this topic be combined with the Standing Wave topic (see
agenda item 10).

8.    SPI-2 Document Review (X3T10/1142D) [Ham]

Bill Ham lead a discussion of the terminator specifications in Rev 3 of 
SPI-2.
Figures 2 and 3  were revised to use a voltage source instead of a current
source.  V3 was specified at 1.0 volts and V4 was specified at -1.0 volts.
Note that higher voltages may be seen under wired-OR conditions, but these
should not matter as the protocol permits a bus settle delay for the 
wired-OR
signals to settle.

Kevin Gingerich questioned why the common-mode test circuit for terminators 
in
Figure 6 specified 0 to 1 MHz instead of a driver-speed pulse, which would
have much higher frequency components.  The 1MHz number was used from the 
last
meeting as a reasonable guess for the frequency components of externally
generated noise.  No decision was made at this point and the frequency
specification for the terminators was left To Be Defined after we settled on
the driver specifications.  At the end of the meeting we agreed to use the
maximum frequency of 120 Mhz for all the terminator tests in the next draft
document.  This frequency was chosen as the third harmonic of the 
fundamental
for the Fast 40 REQ/ACK signals.  This was viewed as reasonable since the
drivers themselves contribute common mode noise due their imperfect balance.

On Tuesday morning, Bill lead a discussion of the driver and receiver 
sections
of SPI-2.  Figure 15 was revised to use different terminology.  Kevin
Gingerich lead a detailed review of his 96-124 document that contains
suggested changes to the driver and receiver sections based on the data he
developed in 96-123, LVD SCSI Model.  Bill will edit all the figures, 
tables,
and text to conform to the new terminology.

There was considerable debate over whether Kevin's model included enough
margin for attenuation.  Kevin included 70 mV which he claimed was 2 dB. 
 Dick
Uber wanted at least 3 dB and he asserted that this should be measured over
the entire signal swing instead of the 0 to peak range.  After reviewing 
Bill
Ham's lab data (see item 9) and considering that the attenuation does really
apply to the entire wavefront it was agreed to drop the use of db's and to
increase the budget for attenuation to 130 mV for subsequent work.

Kevin proposed a complete set of tests and test limits based on the
assumptions of the last meeting.

A great deal of time was spent reviewing the remainder of Kevin's document.
We eventually agreed to wait on specifying the minimum rise and fall times 
 --
these will be discussed on the SCSI reflector.  The basic issue is whether 
the
test circuits are an adequate model for designing drivers or whether
additional reactive components are needed in the test circuit to better
simulate a real cable.  The use of a real cable was seen as problematic due
the the need to test the extreme ranges of the characterictic impadances and
it is not easy to produce precise enough cables so that different test sites
would get similar results.

The common-mode voltage specifications need to be revised to 0.7 to 1.8 
volts
(1.25 +/- 0.55) for the silicon tests.  Accomplishing this may require
reducing the allowed ground shifIt.  It was agreed that the test circuits 
were
OK but that we needed the new numbers for the test conditions and allowed 
test
result limits.  Kevin agreed to incorporate all these new agreements into a
new version of his model and to send these to Bill for inclusion in the next
draft of SPI-2.

For the first time we have a model that is complete enough to use for making
tradeoffs.

Bill called for an Annex to be created that contains the details of the 
model.
This was generally agreed as useful but it was not clear how we could 
document
the Excel program.  Gene Miligan suggest a "C" code listing as one
possibility.

The specification of LVD SCSI bus loading was reviewed with the result that
the sections in Rev 3 are OK.


9.    Test Data (96-127) [Ham]

This item was covered early on Monday.  Bill Ham presented 96-127, which
contains the results of his testing of the TI LVDS drivers (for the EIA/TIA
standard) with biasing terminators.  Bill's previous data had not used 
biasing
terminators and was therefore very optimistic.  While the drivers do not
exactly match the current SPI-2 LVD SCSI specifications, they are close 
enough
to yield very useful data.  The data show that bus signals resulting from
symmetric drivers used with biasing terminators have little assertion margin
(due to fighting the termination bias) and great (excessive?) negation 
margin.
These drivers were calculated to have approximately 6 mA symmetrical current
for both assertion and negation and therefore are a very close match to the
earlier LVD SCSI driver specification.

With symmeteric drivers Bill's data suggests that 27 meters of heavily 
loaded
bus is possible for Fast 40.  He also showed that Fast 80 could not reach
these lengths unless steps were taken to improve the regularity of the REQ 
and
ACK signals.  The attenuation levels reached were those using multistranded
primary copper conductors from the present SCSI 30 guage P cables.  With 
solid
conductors one would expect to experience higher attenuation.  This whole 
area
needs more work to arrive at the tradeoffs.

A major effect on the bias delivered from the terminators was noted due to
driver loading.  For the drivers used a single driver reduced the bias from
112 mV to 82 mV.  This would clearly be bad news if many drivers were 
present
on the bus at the same time (in a HIZ state).  Kevin called back to his 
office
and found that these drivers actually have nearly 0.5 mA "leakage" caused by
an internal circuit intended to pull the bus back to its original condition
when the driver is not driving.  As these drivers were developed for use 
only
in point to point applications this was not a serious issue.  Actual LVD 
SCSI
drivers would have less than 20 uA leakage and 15 would therefore cause 0.3 
mA
total for all drivers.  This is 60% of that for the single LVDS driver used
for these tests.  One can expect no more than 18 mV reduction in terminator
bias from a maximally leaky LVD system.  The test data exactly matches this
case for a terminator delivering an unloaded bias of 100 mV (the present 
lower
limit).  This data confirms earlier projections that leakage would have some
effect.  Fortunately, this level of leakage is easily accommodated.

This means that LVD SCSI is set up for using the newest silicon technologies
that are expected to have higher input leakage.

Bill's conclusions were that asymmetric drivers would significantly benefit
noise margins without causing any increase in power consumption.  They are
also a key for operating at longer lengths.

Bill passed around a new test board developed at Digital that will be used 
for
the creation of precise LVD SCSI bus loading conditions.  This test board 
will
allow adjustment of the LVD loading over a broad range and will enable the
testing of simultaneous lines.  Bill may have data for the March meeting.

Bill also showed a nearly comprehensive set of data describing what happens
during negation to HIZ and other HIZ related transitions.  This data showed
evidence of false assertions significantly after the HIZ transitions caused 
by
reflections.  The general conclusion was that software based glitch 
filtering
should be required (for a bus settle delay).

For the next meeting Bill expects to have some data with asymmetric drivers.

10.   Standing Waves [Bridgewater]

Wally Bridgewater actually had three items:

1. Can the REQ/ACK Offset be greater than 16?  Yes, the REQ/ACK Offset can 
be
up to 254.

2. Are the single-ended cables different than LVD cables because the
impedances are specified differently?  No, the impedances are measured
differently, so the cables can be the same.

3. He has noticed that single-ended cables exhibit standing waves under
certain condtions.  He plans to do some testing to see if differential 
cables
also exhibit standing waves.  Bill Ham will also review his earlier data 
where
some evidence of anomolous frequency dependence was seen on the single ended
signals but not on the differential signals.  This could be a very important
area of concern if it turns out that the differential signals are affected.

11.   Meeting Schedule

The next meeting of SPI-2 Working Group will be Monday March 11, 1996, in 
San
Diego, CA at the Hyatt Islandia (619) 224-1234, hosted by QLogic Corp.
Another SPI-2 Working Group meeting was tentatively scheduled for April 
22-23,
1996 in the San Jose area.

12.   Adjournment

The meeting was adjourned at 4:25 p.m. on Tuesday January 30, 1996.

 --
John Lohmeyer             E-Mail:  john.lohmeyer at symbios.com
Symbios Logic Inc.         Voice:  719-573-3362
1635 Aeroplaza Dr.           Fax:  719-573-3037
Colo Spgs, CO 80916     SCSI BBS:  719-574-0424 300--14400 baud







More information about the T10 mailing list