Tip:
Highlight text to annotate it
X
Thank you for watching this video clip. It’s entitled Signal Integrity Design Using Fast
Channel Simulation and Statistical Eye Diagrams. My name is Colin Warwick, signal integrity
product manager with Agilent EEsof EDA. The problem I’ll address today is typical of
pre-layout signal integrity design, namely: how do you optimize the many transmitter,
channel, and receiver characteristics and settings to meet your design goal at the lowest
cost?
To evaluate even one set of settings, we need to compare the BER contours down to a low
value – maybe 10 to the -10 or 10 to the -12 – versus the eye mask. And then we need
to repeat that simulation thousands of times. For example, if you have only ten parameters,
each with only two settings, an exhaustive search would take 1,024 simulations. With
a real project with smooth variations of the figure of merit versus optimization parameters
you can avoid exhaustive techniques.
But still, the bottom line is you have to run a lot of simulations to find the optimum
set of settings. Multimillion-bit SPICE- like simulations are out of the question; they
take far too long. Bit-by-bit channel simulation using brute force pulse super position over
the length of the channel memory is better. Maybe you’ll get a million bits per minute.
But statistical channel simulation is much better.
The simulation length need only be as long as the channel memories – the pulse response
length of the victim and crosstalk channels, plus some time to perform statistical calculations
that I’ll go into later on. Here’s a comparison of the techniques I discussed in the previous
slide: SPICE-like simulation, bit-by-bit channel simulation, and statistical channel simulation.
This first row compares the methods. SPICE uses modified nodal analysis of Kirchhoff’s
Current Law for every time step in every bit. It’s quite expensive, computationally.
A bit-by-bit channel simulation runs a short SPICE simulation, just to get the pulse response
of the channels. And after that, it uses bit-by-bit super position of the step responses for each
bit in the bit pattern. In a statistical channel simulation, the statistical calculation is
based on the short SPICE step response, or pulse response. The statistical calculations,
I’ll go into in the next slide.
The applicability of the methods is shown in this next row. SPICE is very good for linear
and non-linear circuits. Bit-by-bit channel simulation can only be used for linear time-invariant
circuits. It can be used for any specific bit pattern. And you can let the adaptive
equalizer taps run during a bit-by-bit channel simulation.
The statistical channel simulation, again, is only applicable to linear time-invariant
channels and systems. The bit pattern is fixed. It actually exhaustively covers all the bit
patterns in the memory of the channels. It’s a general bit pattern. And you have to fix
the equalizer taps. You can’t let them adapt during a statistical channel simulation. This
last row shows the advantage of statistical channel simulation.
What BER floor can you measure in a one-minute simulation? Well, with SPICE, you’d make
it a thousand-bits-per-minute. So, the minimum noise floor you can measure is 10 to the -3
bit-error rate. In a minute’s simulation of bit-by-bit channel simulation at a million-bits-per-minute,
maybe you’ll get 10 to the -6 floor for your BER. But the statistical channel simulation
can give you an arbitrarily low BER floor in about a minute.
To get these low BER contours in about a minute, we’ve added a new capability to ADS 2009
Update 1. It’s a statistical mode for our channel simulator. Unlike competitors’ point
tools, the channel simulator is integrated, in that it acts on regular schematic and layout
components that you’d use with ADS Transient Convolution and ADS Momentum. You can model
lumped and distributed elements, including elements with artwork, like these lookalike
components here.
The source is a special binary source, where you can set the details of the binary pattern,
the line coding, the equalization, the electrical properties, and the jitter. Likewise, the
receiver is specific to signal integrity, and it lets you set the equalization and so
on, including adaptive equalization, like decision feedback equalizers. A new custom
IPRO component works with both the channel simulators shown here and the traditional
SPICE-like Transient Convolution Simulator.
You can drive the simulator manually, or from the Tuning, Optimization, or Batch Mode controllers.
In bit-by-bit mode, Channel Simulator works in two steps. For if you probe your schematic
in the artwork, if there is any, using a single pulse source and the regular transient simulator,
we’re pulling the convolution option, if there are any components defined in the frequency
domain, and appropriate EM simulators, if there are any distributed or artwork or layout
components.
The simulator need only run for a short length of time, equal to the pulse response or, equivalently,
the channel memory. Step two is we use the pulse response as a linear time-invariant
model of the channel, and then use signal processing and superposition techniques to
calculate the output for millions of bits, without having to call the SPICE simulator
again at all.
In statistical mode, again, we have two steps. The first step is the same we saw for bit-by-bit
mode. But the second step is much quicker. There’s no need for brute force super position
of each bit, just some calculation based on, firstly, inter-symbol interference and crosstalk
implied in the through and crosstalk pulse responses. Secondly, the jitter specifications.
Thirdly, the equalizer specifications. And, fourth, the line coding. So that’s the concept.
Now let’s look at a demonstration.