IBIS: Issues with SSO Noise
IBIS Issues with SSO Noise
A powerful logic driver, when it switches from one state to another, draws a huge surge of current through its power pins. Some of this current is due to overlap in the ON conditions of the two transistors in a totem-pole driver circuit. This overlap current flows from V CC straight through to GND.
In addition to the overlap current, any current flowing to or from the load must also traverse the power pins (see Figure 13.8). When pulling low, current surges from the load, through the I/O pin, through the low-side FET, and from there flushes out the ground pin of the package back to the ground plane on the pcb and back to the load. When sourcing logic high, current surges from the power rail, through the power pin of the package, through the high-side FET, and from there through the I/O pin to the load. The point is, whatever current traverses an I/O pin must also somehow pass through the power or ground pins of the package.
Figure 13.8. Current surging through the I/O pin of the device must also traverse the power and ground connections, creating noise on the internal power and ground rails as it passes .
As current surges through the finite inductance of the chip's power and ground connections (the lead inductance of the package), it perturbs the voltage between the true ground plane on the pcb and the chip's internal ground substrate (or between the true power plane on the pcb and the chip's internal power net). These voltages are called simultaneous switching output noise (SSO noise) [101] , [102] . On the ground rail inside the chip, the magnitude of the SSO noise equals
Equation 13.2
|
where |
v SSO is the noise voltage on the ground substrate inside the chip package, |
|
i GROUND is the current flowing through the ground pins of the chip package, and |
|
|
L GROUND is the effective inductance of the array of ground pins. |
Equation [13.2] indicates that an SSO simulator needs to know (a) the effective cumulative inductance of all the ground pins in the package, and (b) the rate of change of current flowing through the ground pins. So far, so good. Here's the problem:
Item (a) can be measured in a variety of ways, but often the measured results are not available. Without this crucial piece of information, you can't make ground-bounce calculations. In this author's opinion, information about ground-pin inductance should be provided as a matter of routine by all package vendors . What I specifically want is a complete matrix of pin inductances and mutual pin inductances for the chip package. From this information, anyone can compute the effective inductance of any combination of ground pins (or power pins). There are optional provisions made in IBIS for incorporating such packaging information, and I wish everyone would provide the data .
Item (b) may be approximated if one knows the core currents required by the chip and also the sum of all I/O currents being drawn low by the output drivers. The conventional approach to computing this quantity is to first neglect the core currents on the assumption that the core currents are small compared to the massive I/O currents surging through the power and ground pins (a reasonable assumption for any but the largest VLSI circuits). Then use a ringing simulator to calculate the output current waveform for each output. From the output current waveforms, the peak di / dt due to each individual output may be estimated. The worst-case sum of all peak di / dt values may then be used for item (b), provided that the output currents sum linearly. The problem is, they don't. The drivers interact.
The driver interaction is caused by the shared power rails, which are bussed among I/O cells . When one output switches, it reduces the on-chip supply voltage available to the other drivers. When a driver is faced with a reduced power voltage, it switches more slowly than one with a full power voltage available. Therefore, when the second driver switches, the net maximum di / dt from the pair is less than twice the di / dt from a single driver. In other words, the assumption of independence used by IBIS simulators overestimates ground bounce, sometimes dramatically.
You can get a feeling for the sensitivity of a driver to ground bounce by simply operating the chip at a reduced DC power supply voltage. As you reduce the voltage, the output switching time usually slows down.
The IBIS committee has investigated the possibility of improving its specification file to incorporate the driver-to-driver interactions necessary to properly compute SSO noise, but to date has chosen not to do so.
POINT TO REMEMBER
- IBIS simulators don't yet properly compute SSO noise.