Author: Donald Telian, SI Guys - Guest Blogger
Great question. Now
that I’ve devoted well over 30 years to the field, I’ll venture an answer. My simplistic definition has always been “the
analog side of digital”. I still like
that description, because in practice the binary world of “ones” and “zeroes”
only exists in theory. And, hopefully,
in system memory - assuming the interface’s Signal Integrity (SI) was done
My answer to the question will describe Signal Integrity
from angles not typically articulated. My
statements are best understood against the backdrop of my series Signal
Integrity, In Practice, which provides a detailed description of what
Signal Integrity is. My hope is that the
series equips you to not only understand and succeed performing SI, but also
empowers you to carry the practice into new and uncharted territory. In practice, that is non-optional. As the face of SI is always changing, one
aspect of what SI “is” is participating in defining how it is done. Stay in Signal Integrity for more than a decade
and you will see what I mean. The field
has always provided ample room for contribution, and the larger SI community assumes
collaboration as a given.
In my full-day class
I start with the dictionary’s definition of Signal Integrity: a “signal” is “an electrical impulse
transmitted or received”, and “integrity” is “an unimpaired condition”. Putting it together, the practice of SI
involves getting an electrical impulse to travel from a Tx to an Rx
unimpaired. As that definition is overly
simplistic, we can then spend the rest of the day discussing how to figure out
what impairments are problematic, and how such impairments can be predicted,
understood, and corrected. In many ways,
SI is all about the impairments.
I like SI textbooks, and buy them regularly. Because the world of SI is small most of the
books were written by friends or acquaintances, and I expect to write one
too. But the problem with SI is that it
keeps re-inventing itself, rendering textbooks quickly out-of-date. It’s not that Maxwell’s Equations and most of
SI’s equations change, it’s just that – just like the technological world that
SI attempts to tame – the practice of SI keeps changing. Truth be told, this is what I like about
SI: to survive in the craft you have to
be nimble and creative, as I’ll explain in more detail below.
How has Signal Integrity
As frequencies have increased 4 orders of magnitude
throughout my career, perhaps a better question is how could SI not change? And frequency is not the only thing that has
changed; we’ve also seen orders of magnitude transitions in integration,
materials, and measurement equipment.
These forces have influenced change, and are inter-related. In the same time frame, items that have
remained fairly constant are PCB fabrication dimensions and computer
architecture. As a result, we are still
moving signals between the same things (e.g., CPU, memory, IO) amidst roughly
the same dimensions (few things resist change more than PCB tooling).
The most obvious SI changes are the interface transitions
from common-clock to source-synchronous
links. Said another way, the “clock”
(i.e., the signal that defines when the others are valid) has transitioned from
traveling “to” the data to “next to” the data to “inside” the data. These changes brought our required precision
from nanoseconds to hundreds of picoseconds to tens of picoseconds. As these transitions are adequately described
elsewhere (textbooks that are becoming out of date), I will not belabor this
point here. Thankfully, models, simulators
and measurement equipment have kept pace – always keeping us in the ~3% accuracy
realm. As changed happened, it has been interesting
to watch the interplay between measurement
measurement being more bandwidth-limited than the simulation, while simulation’s
lack of limitation helped foreshadow the future while simultaneously leaning on
measurement for self-correction. Say
what? Great fun.
Figure 1: Signal Integrity, Across the Decades
What Signal Integrity
As SI emerged as a design task in the 1990s it unfortunately
became shrouded in mystery. This was
further perpetuated by the seminal 1993
SI book that subtitled itself “A Handbook of Black Magic”, a myth even the
authors quickly dispel in their preface.
I still wish they hadn’t likened our craft to “black magic”. If there is any mystique associated with SI
it is around how it is done, which
I’ve already established to be dynamic.
Signal Integrity is not black
In reality the principles and physics that govern SI are not
mysterious. The engineer willing to step
beyond the tidy “digital” world quickly discovers the many SI principles and
equations that existed before digital electronics began. The problem is modern electronics engineers
are constrained by either their job description or inclinations, relegating
digital “fuzziness” to the SI engineer.
And we haven’t always given clear and timely answers, which has further
fueled the mysterious side of SI.
Indeed, our favorite answers are not the binary “yes” or “no” but rather
“it depends”. I guess that’s our analog
expression of how we view the digital world.
So what is signal integrity?
Looking at SI at the highest level we see three things: Signal Integrity is essential, a
differentiator, and a practice that requires creativity.
Figure 2: Three Aspects of Signal Integrity
Signal Integrity is
I still remember the day the CEO of Intel passed down
guidance to the divisions to invest in Signal Integrity. In the 1990s the key metric for each new
processor – even from the consumer’s viewpoint – was frequency. As increasing frequency demanded faster edge
rates a new wave of SI problems emerged that had to be contained; SSO,
crosstalk, ringing, non-monotonicity, flight times, and meta-stability were
affecting system performance, schedules, and volume production ramps. SI had become essential.
During this time, it was interesting to watch the responsibility
of SI jump from system to IC companies.
Integration moved an increasing amount of system design inside the ICs,
and IC companies discovered their products could not succeed in the market
unless they credibly resolved SI issues – both inside and outside their
components. And so IC companies began
publishing “design guidelines” and “reference designs” to detail the boundaries
in which reliable signal integrity could be achieved. We began calling this required design space
“the box”. Make the box too small and system
companies could not design in the product.
Make the box too big and SI was lost.
The burden of achieving reliable SI within a realizable design space
became the burden of silicon design, and this is why – even to this day – many of
the best SI engineers work for IC companies.
The requirement to hand-off a reliable design space (i.e.,
“the box”) blurred the practice of SI.
While EDA vendors may assert that every system design must perform SI
simulation before production sign-off, in practice this is not the sense in
which SI is essential. What is essential
is that SI must be resolved at some point in the design flow, and it may be
done by the IC company, the system company, or both. System companies using components from
world-class IC vendors and staying within the box often choose not
to simulate. However, if every
system company stayed within the box there is no differentiation in product
design and the market is flooded with too many instances of the same thing. Find a way to implement different form
factors, higher frequency, or increased functionality and a product stands out
among the others. But these ideas
require robust SI analyses to engineer beyond the box. As such, Signal Integrity is a great way to
differentiate a product.
Signal Integrity is a
While the IC company finds Signal Integrity an essential
part of design, the systems company that invests in Signal Integrity finds it to
be a significant differentiator. Systems
companies use SI to cut costs (pages
23-24), improve performance,
add design and manufacturing margin (pages
20-23), and – most importantly – to expand the design space (“box”) defined
by their IC vendors and develop a better product than their competitors using
the exact same components. I think this
is where SI shines the brightest, and adds the most value.
It’s no secret that “the box” defined by IC companies is overly
constrained and slanted in their favor.
And good for them; they need to ensure the system company with no SI
capability succeeds implementing their product.
And so many design engineers willingly stretch a component’s rules and
guidelines, knowing they are guard-banded and not as authoritative as they may
seem. (When I worked at IC companies, we
often quipped “I used to think datasheets were perfect, and then I wrote
one”.) And yet where is the edge? How do I understand, and even quantify, the
risk I am taking in my journey outside the box?
Better yet, is there a way to re-think the assumptions and engineer
these components into an expanded or entirely new application? For those ready to think creatively, Signal
Integrity provides the answers.
If you step into SI you will consistently be asked to craft
clear answers from insufficient data based on ill-defined measures of
success. Does that require resilience
and creativity? You bet. It’s not that our simulators can’t provide
answers to 14 decimal places; indeed they do, but that is not an indication of
precision. And so the process of
learning to what extent SI’s data can be trusted is a journey unto itself. The models are never
perfect, simulators and measurement tools can easily lead you astray (or,
Mike says, “pick your pocket”), and – even if you learn how to use all that
to derive a valid answer – it is often unclear what to compare it to. That’s a great job description, huh?
I believe my first
career as a musician
and artist helped me here. In music
you learn to create something that makes sense from a palette of notes, sounds,
harmonies and rhythm. You can combine it
all without skill and frustrate those around you, or you can blend them into
something that feels right. And so it is
with SI. Artistic creativity is required
because – even though it is an “engineering discipline” – the practice of
Signal Integrity and its deliverables are not as measurably precise as say a
schematic or a layout. Those tasks have
a well-defined and literal set of inputs and outputs; there’s no fuzziness to
success because everything is either connected or it’s not. While I admire the attempts of SI tool wizards
and textbooks to present tasks as steps 1, 2, 3, I find the analogy of viewing
SI as an attempt to paint more consistent.
Will your painting come out right?
…will people understand it? …and,
most importantly, will the product “work”?
The practice of Signal Integrity requires the creative
application of a baseline set of skills.
It’s imperative to learn the principles and equations that govern SI as,
using the painting analogy, these are your colors. It’s also imperative to become skilled at
using models and simulators, these are your brushes. The project is your canvas. The challenge is to combine it all creatively
to produce something that makes sense and “works”. My advice to engineers climbing the SI
learning curve continues to be “hang onto your creativity” (2:25
And so there you have it:
Signal Integrity – as it has emerged in the realm of digital electronics
– has proven itself to be an essential and creative practice that offers
High-speed digital Signal Integrity has worked its way into
every corner of the electronics industry.
Whether you’re a PCB Fabricator, Test & Measurement Vendor, Contract
Manufacturer, Component Vendor, Standards Organization, or involved in any type
of digital product design, some amount of fluency in SI is required. Signal Integrity touches everyone, and is no
longer a fringe skill set.
Tune into my larger series on Signal
Integrity, In Practice to learn more about what SI “is” in terms of
specific skills and design tasks. In the
series I have endeavored to make the complex simple, boiling SI down to its
most essential aspects. I trust it will
be helpful, and look forward to your contribution as together we answer the
question: What is Signal Integrity?