Measurement Correlation

Wednesday, May 8, 2019

I believe what simulators tell me.  It’s funny because, throughout my career, I’ve met many who don’t.  How is it that I somehow learned to trust them?  Measurement Correlation.  Over the years, taking the time to build, measure, and compare what I simulated taught me what’s good – and what’s not so good – about simulation.  I developed confidence in my results and intuition that tells me when they’re suspect.  Said another way, I learned to get value out of the tools while juggling the fact that “All models are wrong, but some are useful”.

Measurement correlation is the process of cross-checking your pre-hardware simulation results against post-hardware measurements.  Sounds simple?  Well, yes and no.  Let’s dive in. 

The Tools of the Trade
High-frequency passive hardware measurements are made using two types of equipment:  Vector Network Analyzers (VNA) and Time Domain Reflectometers (TDR).  As the names suggest, VNAs work in the frequency domain while TDRs focus on the time domain.  This difference means they do a great job at helping us quantify and correlate the two things we need to solve:  Loss and Discontinuities (Steps 1 &2), respectively.  The good news is that if you measure your system using one tool/domain you can mathematically transform your data to look like it was measured in the other tool/domain.  Pretty handy. 

Active measurements use Oscilloscopes.  These measurements are a little more complex because you not only have to induce the signals to perform the targeted task, but you also have to figure out how to probe locations you have little access to.  Accurately probing a running system is challenging.  To make matters worse, serial links have brought us signals that are only distinguishable after Rx equalization deep inside the IC.  For this reason, many modern components offer utilities that measure what’s going on inside the chip.

Performing measurements is often out-sourced.  As the tools and techniques required to achieve accurate system measurement are complex and expensive, some companies have built a viable business offering this service.  For the value delivered, I’ve found the price reasonable.  This brings the task of measurement correlation within reach, so don’t be afraid to ask the next question. 

When Should I Correlate?
Answering this question is a lot like answering the question “When Should I Simulate?”.  The answer again lies in determining the “newness” of the technology at hand.  If it is new to you, your company, your fabricator, or the technology world in general, it is time to correlate.  Examples might be using a new dielectric material or a new PCB manufacturer.  Was everything actually built the way you simulated it?  Were the parameters used in your simulations reliable?  Think about the whole design, fabrication, and assembly chain.  It’s challenging to get everything to work right, particularly when it’s new, and measurement will help you isolate problems.

Cost, schedule, and complexity will force you to be selective with your measurements.  In practice you can’t measurement everything, so I like to focus on the “corners” – as you might do with your simulations.  For example, measure a couple long, short, and typical traces.  This creates a sort of binary search through the fabrication, exposing areas you may need to examine more closely with additional measurements.  It’s also important to measure structures with the greatest risk – perhaps those you have reason to believe the simulator and/or fabricator may not get right.

Once you’ve learned to answer the question of when you should correlate, you can then figure out when it’s not necessary.  Early in my career I was surprised when my manager suggested we not measure this time.  What?!  I had assembled a lab of equipment, developed checklists of what to measure, trained teams of people to execute it all, and built utilities to overlay simulations and measurements.  The idea of not measuring seemed absurd to me.  His point was we had measured that type of interface at that speed many times, and were not likely to learn anything new and simply checking the box.  The “newness” was gone, and our simulations had become reliable.  He was right.  So yes, I’ve learned to trust simulation.  But notice that measurement correlation is how I got there.

A Sample Channel Correlation
My previous discussion about Serial Link Simulation detailed a channel that spanned 3 PCBs, and explained how it was constructed and simulated.  Figure 1 below overlays a TDR simulation of that channel (green) with a TDR measurement of the channel after it was built (red).  Excellent correlation is seen in some factors, while others are not so good.  The largest discontinuities are the backplane vias, which were modeled accurately by the simulator.  Though via impedance is far from ideal, at least we were able to assess their impact on the larger system correctly.  The via’s time placements are also precise, suggesting both Tx and Backplane lengths and Dk values were correct.  Miss-correlation is seen in the slope on the Tx Card, suggesting something is awry with the accumulation of loss in the fabricated or simulated parameters.  On the backplane, the general slope is better yet the impedance is not so.  The connector models lack impedance irregularities, although their time delays are good.  Though the Rx Card correlation looks good here, if we plot the TDR from the other end we would see a clearer picture of the inaccuracies at that end.  Like the signal in the actual system, the bandwidth of the TDR plot rolls off as you move from Tx to Rx.

 

Figure 1:  Correlation of Simulated (green) and Measured (red) TDR of Channel

Figure 1 illustrates the value of measurement correlation.  You will find problems with your simulation and/or fabrication, and make the necessary corrections.  And you will always learn something.  Working through this process improves your sense of tolerances, what to believe when, and hence your overall engineering judgment.  And you need that, because Signal Integrity often requires you to make decisions amidst a plague of insufficient information. 

An Exercise in Miss-Correlation
With practice, TDR measurements provide valuable insight.  As Dr. Mike points out:  “There is a lot of information that escapes the casual observer but is readily apparent if one knows what to look for.”  So let’s test our skills with an exercise.  Figure 2 overlays channel simulations (red/blue) and measurement (green) of the channel shown in Figure 1.  Can you look at the simulation miss-correlation (red), determine what’s wrong, and then correct the simulation (blue) by adjusting only one PCB parameter?

 

Figure 2:  Channel Correlation Exercise.  Can you change red to blue with one parameter?
 

Comparing the red and green TDRs of Figure 2, the most obvious difference is the time shift seen in the two backplane vias at 2.4 and 4.0 nS.  This causes many observers to assume the simulated length on the Tx Card is too long.  While shortening the length would indeed align the vias, this is the wrong answer.  In practice, length is the simplest parameter to get right.  Looking closer at the Tx Card miss-correlation (0 to 1.5 nS), we see that the Tx impedance is also low.  So is there one PCB factor that influences both impedance and time delay?  …a parameter with a wider manufacturing variation than length? 

Thinking back to the exercises in our discussion about trace impedance, we recall that dielectric constant (Dk/Er) affects both time and impedance.  Time scales with Dk while impedance is inversely related.  If Figure 2’s manufactured Dk was low, measured impedance would be high and the time delay would be short.  And this is precisely the situation, and adjusting simulated Dk from 3.5 to 3.3 produces the change from red to blue.  This exercise illustrates is a typical miss-correlation, because manufactured Dk often varies within the tolerance shown in Table 3 on page 16.  When Dk is found to vary substantially, perhaps greater than 7%, you should contact your PCB fabricator and request an explanation. 

A Success Story
Every project has different design goals, tolerances and measures of success.  If you’re in a situation that requires tight control of variables, you can use the simulation-measurement correlation process to achieve it.  One such design task required control of a variety of discontinuities in a small space, and  new materials were being used.  Figure 3 shows measured performance across three design iterations, from red to blue to green.  I still remember the day we made it to green – funny the things that can make an engineer’s day.  The first iteration (red) exposed model problems.  Correcting those, we were in position to resolve manufacturing issues in the second iteration (blue) which allowed us to achieve the almost flat lines in the third iteration (green).  This story is told in greater detail on page 19, including eye diagrams that illustrate how the measured discontinuities change performance.

 

 

Figure 3:  Measured TDRs of Seven Discontinuities Across Three Design Iterations

Given the problems we encountered, I suppose this section could have been entitled “Failure Stories” instead of “Success Story”.  But I view it as “Success” because our process gave us a way to find, understand and correct failures.  And none of it would have been possible without measurement. 

In Conclusion
Electronics has entered the mainstream with products for every consumer, and now 40% of the world’s population uses a smartphone.  This change has radically altered and compressed design cycles, and “engineering” as I once knew it may never be the same.  One part of me thinks “of course engineers know they need to design, build, measure, learn, adjust – that goes without saying”.  Yet the other part of me has to acknowledge many may not take the time to measure, because before your product is built you are already working on the next project.  I believe it’s important to make the effort, because measurement correlation adds value to the product development process and builds simulation confidence that cannot be obtained in any other way.

By themselves, neither simulation nor measurement holds the keys to solving all your design challenges.  But deployed together they become quite powerful – a classic example of the sum being greater than the individual parts.  Throughout my career I’ve encountered those who only trust measurement and others who only simulate, yet time has taught me the value of the combination.  Accurate measurements can be readily obtained, and using these measurements to correlate and validate your simulations improves both your design and manufacturing processes. 

 

 

 

 

Donald Telian, SiGuys - Guest Blogger 5/8/2019

Add your comments:

Items in bold indicate required information.