All,
I think we can make the following conclusions from the Figure 16
presentation and Bob's presentation this week in IBIS-ATM and various
discussions:
1. IBIS is a specification of a data format that contains the
following behavior of I/O buffer models:
a. IV and VT curves, and measurement/threshold values are of a
Device Under Test (DUT)
b. The conditions of a DUT are:
i.
Rail voltages are kept constant relative to a test fixture reference
"Node"
ii.
These Rail voltages are specified by the IBIS Parameters
1. [Voltage Range]
2. [Pullup Reference]
3. [Power Clamp Reference]
4. [Pulldown Reference]
5. [GND Clamp Reference]
iii. IV
and VT tables are measures at the I/O buffer, and exclude all package and
other parasitics that may be required to perform the test.
iv. Since
the rail voltages are kept constant, C_comp compensation may assume that
C_comp is connected to the test fixture ground, or any other rail
terminal.
2. What happens if a device is operating at conditions other than
the DUT conditions?
a. Case one is the rail voltages are not the same as the DUT rail
voltages, but are kept constant
i.
Since the rail voltages are kept constant, simulation results will not
change if C_comp is connect to any of the local rail voltages or any
simulator DC reference node (include Node GND or Node 0). Similarly,
package models may reference any simulator DC node.
ii.
Simulators need to apply the voltage measure between the I/O pad and the
actual rail voltage when using the IV curves
b. Case two is the rail voltages are not the same as the DUT rail
voltages, and in fact are changing in time
i.
This is due to the rail voltages being modified in time by the simulator,
and can be done in multiple ways including but not limited to
1. Full power aware simulations with package and on die power
distribution and time varying loading due to composite current from this
or other I/O buffers in the simulation
2. Voltage variations in time injected by the simulator using I/O
buffers that do not have composite current data.
ii. If
there is a C_comp (not split C_comp), errors will be introduced no matter
what a simulator chooses to connect C_comp to. Options are:
1. A simulator DC reference Node (e.g. Node 0)
2. A rail voltage that is most likely GND (e.g. GND_Clamp_Ref)
3. Split between the various connected rails
4. General agreement that 1. Is incorrect, and that EDA tool could
do 2.b.ii.2 or 2.b.ii.3 or give user choice between 2.b.ii.2 or
2.b.ii.3)
iii. It
is logical that package models should reference the "GND" signal name that
is the reference "GND" for the buffers the package model is connected to.
iv. The
model maker has the ability to split C_comp so that it is clear what the
EDA tool does for C_comp/
v. The
model maker has the ability to choose the correct reference signal_names
for his package model interconnect.
vi. If the
model maker does not split C_comp or reference package models "correctly"
then the EDA tool has two choices:
1. Assume a "Ground Referenced System"
2. Make best guess for the "GND" reference signal_names for C_comp
and package models
3. Have user select signal_names for C_comp and package models
First, can we agree to the above statements?
If we can agree, then can we say that everything in IBIS is talking about
DUT, and GND because simple, and then add a separate, new section,
discussing the section 2 above (Device in Action).
Walter
Walter Katz
<mailto:wkatz@xxxxxxxxxx> wkatz@xxxxxxxxxx
Phone 303.449-2308
Mobile 303.335-6156
Attachment:
Figure_16.pdf
Description: Adobe PDF document
