Issue link: https://iconnect007.uberflip.com/i/275965
March 2014 • The PCB Design Magazine 29 Some systems are very repeatable and sit well within their factory specifications, but the specifications are there for a reason; do not assume because your measurements are very repeatable that they are more accurate than the published tolerance of the measurement system. " " TOLERANT OF TOLERANCE? continues feature This is why you will see, as designers push high speed digital to ever loftier transmission speeds, that insertion loss S21 or differential insertion loss Sdd21 is measured as a ratio of power into the line under test vs power out. The difference between the power in and out will be the in- sertion loss. Putting a tolerance on this type of measurement is tricky as the measurement of the loss will depend as much on the de- sign of the test pieces as on the me- trology instrument. It is interesting to look at this from a distance as, to a casual observer, accurate low- frequency measurements need a good meter, but the probes can be quite simple and low in cost. As the frequency ramps up, shielding is often required, along with coaxial or trans- mission line probes. Addition- ally, the interconnect between the probe and the measure- ment system becomes a criti- cal part of ensuring accuracy and repeatable measurements. At the highest frequencies, where ultimately a user may resort to a probe station, the probing system cost can be of a similar order to that of the measur- ing instrument. In fact, at high frequencies, all the parts of the measurement system need to be carefully crafted to ensure you can make ac- curate measurements within the desired toler- ance; typically, the higher the frequency, the more the achievable tolerance widens out. It is interesting to look at some VNA speci- fications regarding tolerance and accuracy: you will find that the measurement precision de- pends on a blend of factors, including the de- sign of the test piece. For example, if a coupon is designed with a very high loss (almost com- plete signal attenuation) or very low loss (al- most 100% transmission), it is hard to make an accurate measurement of loss per unit length. Samples therefore need to be designed with an appropriate loss depending on the base mate- rial used. Therefore, for low loss materials the coupons may need to be longer (loss is propor- tional to line length) than for a standard loss FR4 material. What increases your production tolerance? First—and, seemingly, one that people miss—you need to allow for the tolerance of your measurement system. Some systems are very repeatable and sit well within their factory specifications, but the specifications are there for a reason; do not assume because your measurements are very repeatable that they are more accurate than the pub - lished tolerance of the mea- surement system. Given that you have accepted this and you are working within the limits of your measurement system, let's take a look at con- trolled impedance traces and the possible offenders which may cause some of your prod- ucts to fall outside, or close to, their permitted tolerance. Cross-sections are important here. The primary drivers of impedance are dielectric sepa- ration and trace width. Trace thickness and dielectric con- stant are second-order effects. It follows then that variations in line width may cause traces to vary in their measured impedance. Taking several cross-sections can confirm if this is the case. As line width reduces, it becomes more difficult to achieve a given level of tolerance. Surprisingly (to some designers, at least), the same is true for dielectric separation. The board thickness does vary in different parts of the PCB—copper density is a key driver of this—and it is essential to ensure the density of Cu is well balanced to ensure there are not excessive varia- tions of height and therefore dielectric separa- tion. Local variation of dielectric constant can also have an effect—witness the many papers on fiber weave mitigation—though, interest- ingly, low-twist fibres and mechanically spread glass can help this and perhaps in future the ad- vances in glass design will reduce the need to deal with fiber weave variations in E r .