Troubleshooting Analog Circuits--Preventing Material and Assembly Problems: PC Boards and Connectors, Relays and Switches (part 1)

Home | Articles | Forum | Glossary | Books

PC Boards and Connectors, Relays and Switches

In addition to your choice of components, the materials you use to assemble your circuit will have an impact on how well it performs. This section covers what you need to know to solve the occasional problems caused by PC boards, solder, connectors, wire, and cable. Also covered is PC-board layout--a poor layout can cause more than occasional problems; it can determine completely how well your circuit works.

Some of the topics discussed in this troubleshooting series so far may have seemed obvious. But far too often it is this "obvious" information that engineers overlook, and it is this information that can make troubleshooting so much easier. So, be careful not to overlook the obvious. Don't assume that your PC-board materials or layout don't matter or that wire characteristics don't differ; you'll find that PC boards, connectors, wire, and cable can cause problems when you least expect it.

First of all, the use of the term "printed-circuit board" is a misnomer; these days, almost every board is an etched-circuit board. But I'll continue to use the abbreviation "PC board" because it's a familiar term. There are six basic troubles you can get into with PC boards:

The board is made of the wrong material.

The quality of the vendor's board is so bad that there are opens or shorts in the board or, worse yet, intermittent connections in the plated-through holes.

The foil starts to peel off the board because of mistreatment.

You were so concerned with cost that you neglected to specify a layer of solder mask; you ended up with a board full of solder shorts.

The surface of the board is leaky or contaminated.

Your circuit layout is such that signals leak and crosstalk to each other, or controlled-impedance lines are interfering, thus causing reflections and ringing.

Avoid PC-Board Problems at the Outset

The fixes for these problems, and ways to avoid them in the first place, are fairly straightforward.

These days, the G10 and G11 fiberglass-epoxy materials for PC boards are quite good and reasonably priced. Trying to use cheaper phenolic or "fishplate" is not economical in most cases. Conversely, a special high-temperature material or an exotic material or flexible substrate may be justifiable. If you don't have an expert on these materials, the PC-board maker or the manufacturer of the substrate material can usually provide some useful advice. (See Table 1 for a comparison of PC-board materials.) In some RF applications, phenolic material has advantages over glass epoxy: it has a lower dielectric constant and superior dimensional stability. And for ultra-broadband oscilloscope probes, some types of glass epoxy have a definite disadvantage due to mediocre dielectric absorption, especially if the epoxy has not been properly cured.

As for quality, there is almost never an excuse for buying your boards from a vendor whose products are of unknown quality. "Low cost" would be one poor alibi; "Can't get acceptable delivery time from our normal vendor" would be another. One time, to meet a rush contract, we had to build circuits on boards made in our own lab facilities. I had never had trouble before using these boards for prototyping. so I was surprised when I began the troubleshooting and found that an apparently good board occasionally had a short between two busses.

Close inspection with a magnifying glass showed a hairline short about 3 mils wide, which was caused by a hair that fell onto the artwork. You would never ask a printed-circuit foil that narrow to carry 20 mA, but this narrow short would carry 200 mA before blowing out. Similarly, we found hairline opens: The ground bus was broken in two or three places by a tiny 4-mil gap, just barely visible to the naked eye.

Of course these "opens" were caused by the image of a hair, during a negative process.

After several hours of fiddling around, opening shorts, and shorting opens, we vowed not to be caught by such poor workmanship again.

As for the third problem, don't let ham-fisted engineers or technicians beat up a good PC board with the overenthusiastic or misguided application of a soldering iron.

That's sure to lift the foil. Use an iron that's hot enough so you can get in and get our.

If the iron's not hot enough and it's taking too long, that's when the foil will lift . . .

Solder mask, the subject of the fourth problem, is almost always worthwhile, as many people have learned. Without it, the admirable tendency of solder to bridge things together, which is wonderful in most instances, becomes disastrous.

Table 1 PC-Board Laminate Materials

Figure 1. You can use this circuit to test your board for leakage current. The transistors are connected across the op amp's feedback path so that they act like a current-to-voltage detector, with a wide-range logarithmic characteristic.

Figure 2. You can calibrate the logarithmic-current-meter circuit of Figure 1 to sense currents between - 1 mA and - 1 pA and between + 1 pA and + 1 mA.

Leakage Can Be a Problem

When a PC board comes from its manufacturer, it is usually very clean and exhibits high impedance. Sometimes a board starts out leaky, but normally a board doesn't begin to leak until you solder it or wash it with a contaminated solvent-the fifth problem.

Looking for leakages

When you have a leaky board or a slightly less-than-infinite-impedance connector or insulator, how do you test for leakage? You can't just slap a DVM on it, because even on the highest range, (for example, 20 M-ohm) the display will just read Over range.

That's no help if you want to read 2000 M-ohm or 20,000 or 200,000 M-ohm even higher. Some DVMs or digital multimeters have a scale for microSiemens (conductance) which will let you resolve as high as 100 M-ohm. But this scale does not normally have more resolution than that.

Figure 3. The DVM approach is an alternative to the approach illustrated in Figure 1 for testing leakage. You can calculate the leakage current from Ohm's Law: Vs = I, X Rs, or IL = V1/R1.

There are basically two ways to measure leakage current. The approach I have used for many years is to connect a couple of transistors as a wide-range, logarithmic, current-to-voltage detector across the feedback path of a low-bias-current op amp. These days, I'm not using vacuum tubes--I have graduated to an LMC660, as shown in the circuit of Figure 1. I calibrated the meter with a hand-drawn scale to sense currents ranging from +1 pA to +1 mA and -1 pA to -1 mA (Figure 2). As long as the air conditioning doesn't break down, I know my calibration won't drift much more than 10 or 20%, which is adequate to tell me which decade of current I am working in. (The V,, does have some temperature sensitivity, but not enough to bother this circuit very much.) Because these transistors are, of course, quite nonlinear as current-to-voltage sensors, you do have to shield the summing point away from AC noises (60 Hz, 120 Hz, 1 MHz. etc.) to prevent rectification and false readings. So the whole test circuit and the unknown impedance are best located in a shallow metal box, grounded, with an optional metal cover.

In either case, if you put 15 V across 1,000,000 M-ohm and measure 15 PA, that is least 50,000 X higher resolution than most meters that can only measure up to 20 Ma. Whichever detector you use, apply a reasonable voltage across the unknown impedance and see where the leakage gets interesting. This method can also be used for diodes and transistor junctions. The op-amp circuit is not especially recommended for measuring the leakage of large-value capacitors; neither is the DVM approach because of the slow charging of the large capacitance, and because of soakage, or dielectric absorption, effects. But, if you're desperate and start out with a low value of R, you can eventually get some approximate measurements.

It's true that the DVM approach shown in Figure 3 has a little more accuracy and perhaps more resolution; but it too is easily fooled by noise, and the digital readout doesn't show trends well. And, if you want to cover a wide range of currents, you have to switch in different resistors or wait for the DVM to autorange, which is not my idea of fun. On the other hand, you can find a DVM almost anywhere, so this approach is easy to implement.

Recently, a customer had a problem with a simple basic design using an LM317 regulator in which the circuit's impedances were fairly low--just a few hundred ohms. After just a few minutes of operation, the output of the LM317 would start drifting badly. The cause turned out to be nor the LM317 or the resistors or the capacitors, but the flux build-up where the board hadn't ever been washed after soldering. In this case, the impedance of the scorched flux was as low as 500 ohm when measured across a 0.1 X 1-in. area of the PC board. The leakage was from the +V_in to the output, and it pulled the output voltage up out of regulation! So, even if you are not trying to achieve 10^12 ohm of leakage resistance, you should still observe rudimentary standards of cleanliness, or even your simplest circuits won't work right.

Similarly, one of our PC boards designed for a S/H circuit was yielding 10^11 R of leakage resistance, which was unacceptable. We tried cleaning the board with every organic solvent but had no luck. Finally, I took a few boards home and set them in the dishwasher along with the normal charge of Calgonite. After a full wash-and-rinse cycle, I pulled out the boards, banged them to shed most of the water beads, and set them in my oven to dry at 160F. The next day, they checked out at the more acceptable value of 10^13 ohm. I have used this technique several times on leaky PC boards and sockets, and it works surprisingly well. It can work when alcohol, TCE, and organic solvents are not helping at all.

After you get your board clean and dry, you'll want to keep it that way. For this purpose, you may want to use a coating such as the urethane, acrylic, or epoxy-types, sprays or dips. Humiseal is the pioneering name, and they have a broad catalog of different types for various production needs. In a similar vein, the guys up at Essex Junction, VT told me about some varnish made by John Armitage Co., which is a rather thick, heavy high-impedance coating. It takes a while to dry, but it's pretty durable and I like it. When I was building some little 1/3-ounce modules that some scientists were going to carry up to the top of Mt. Everest, I chose a couple well-baked coats of the "Armitage" to keep the modules clean and dry; it's much lighter than potting in epoxy, which is important when a guy has to carry a scientific package on his back up to 29,000 feet.

Of course, with any of these coatings, it is not trivial to cut in and repair the circuit or change components. So your choice of a durable coating should be tempered by the awareness of how much fun it is to go in and remove the coating and do your repairs.

When I was at Philbrick, we potted most of our products in epoxy, and it gave good reliability and security. If you get a good circuit in there, well potted in epoxy, it has an excellent chance to survive forever, with no moisture getting in, and with everything held isothermal, protected from shock and physical abuse. Of course, if somebody abuses the circuit electrically and damages it, it's substantially impossible to get inside to repair it. You may have to drill down to the PC board, just to do some troubleshooting. It's fun and a challenge to have to delve in and troubleshoot a potted circuit. Sometimes the potting material adds extra stresses to components: Squeezing the resistors and capacitors can change their values, and pouring epoxy around a circuit can add significant capacitance, too. If you pot a circuit that hasn't been baked and dried out thoroughly, the moisture may get sealed inside the potted module. Epoxy can cover up a multitude of sins, but there is no substitute for good workmanship and good engineering. Potting a piece of junk usually leads only to well-potted junk.

Make sure the designers who lay out your PC board keep a list of rules to avoid troubles. For example, if your circuit has a high-impedance point and you suspect leakage might be a problem, don't run that high-impedance trace beside a power-supply foil-guard it with a stripe of ground foil or "guard foil" between the two. A dozen times I have heard an engineer say, "The resistivity of this glass-epoxy material is 10^14 Ohm-cm, so you can't possibly expect to have a resistance of 10^12 R from your summing point to the rest of the world..." Then, I demonstrate that the measured impedance is typically a lot more than the specifications say, but I agree that I wouldn't dare count on that fact. So, I guard the summing point to ground with a grounded foil surrounding the critical nodes on both the top and bottom of the boards.

With the addition of these grounds, the circuit can perform well even under worst-case humidity conditions. After all, the internal volume of the glass-epoxy insulator is always dry, whereas the surface is the place where you can easily get a leakage problem due to dirt or moisture. That's where you have to prevent the leakage. Of course, crosstalk and high-frequency capacitive-coupling problems are caused by adjacent foil locations and are cured by the same guarding and shielding just discussed to prevent foil leakages. To help you plan a good layout, think about what dv/dt and di/dt will do in a poor layout.

Location, Location, and... Location

The field of PC-board layout is a subject unto itself. But there are some things you can do or add to a layout that will make testing the circuit much easier. Thoughtful designers have a store of these tricks, but I bet very few people write them down. In my world, the unwritten rules are the ones that are broken, so we are trying to write them down. I recommend that all designers write down a list of their good ideas.

Some layout tips from my list are:

Make sure that the signals you need access to, for troubleshooting or analysis, are easy to find and probe. Make a small hole in the solder mask for accessibility.

Include a silk-screen layer of labels in your layout artwork showing each component and its reference designation. It's also a good idea to label numbered test nodes and the correct polarity of diodes and electrolytic capacitors.

Arrange the signal paths so that if you are desperate, you can easily break a link and open a loop, be it analog or digital.

Many modem PC boards have multiple layers and sophisticated patterns of ground planes, power-supply busses, and signal flows. Troubleshooting such a board requires specialized techniques and skills and all sorts of "maps," so you don't get lost or confused. Make sure that all the board's nodes are accessible, not hidden or buried on an inside layer, or under a large component.

When possible, leave adequate space around components, especially ones that are more likely to fail and need to be replaced. Such components might be part of circuits that lead off the board and into the realm of ESD transients and lightning bolts, and thus might occasionally fail.

Locate delicate components away from the edge of boards, where they might be damaged by rough handling.

Beware of using eyelets to connect different layers of foil on your PC board.

Years ago, plated-through holes were considered risky, so we used eyelets to connect the top and bottom foils. When these eyelets went through temperature cycling, the thermal stresses would cause the eyelets to lift the foil right off the boards. Even in the last year, I have seen advertisements that sing out the praises of eyelets on PC boards, which scares the heck out of me. If you have to use an eyelet, don't count on it to connect the top foil to the bottom. These days, plated-through holes are quite reliable, but I still like to use two plated-through holes in parallel, whenever I have room, or to put a component's lead through the hole. It just makes me feel more comfortable. (See Section 13 for more comments on eyelets.)

More important than these layout conveniences is that your layout should not interfere with and, if possible, should enhance the expected performance of the circuit.

I try to make my layouts so the PC runs (and the metal runs on an IC chip) are as short and compact as possible, especially the sensitive ones that would receive a lot of noise or leakage or have a lot of capacitance, if they were longer. Otherwise, you have a lot of long wire runs that turn into a whole hank of spaghetti.

On the other hand, sometimes we have to locate some of the components in odd locations, for different reasons such as thermal or human interface-which leads to the "spaghetti syndrome." But sometimes you have to do it--it's a matter of engineering judgment, a matter of the trade-offs you have to make. Recently I was trying to design a 200-MHz counter, to start and stop in less than 5 ns (Ref. 1). I was able to lay out the fast 100,000-type ECL gates right close to each other, and with this close layout, I was able to gate the clock counter ON and OFF very quickly, faster than 3ns. Most designers of digital circuits are aware that, with high-speed logic, you can't just run the fast signals "any old way." You have to treat these signal paths as transmission lines and route them carefully. So, whether with printed-circuit foils or wire-wrapped conductors or IC metal runs, most digital-circuit designers have learned to design the wiring of fast digital circuits so that they work well--to avoid ringing and crosstalk.

However, I have seen cases in which an experienced digital designer had to add a few linear circuits into one comer of a mostly digital PC board. If the designer makes bad guesses about how to wire an opamp, the linear circuits may oscillate or exhibit bad cross-talk or work poorly. And, the availability of wire-wrap connections makes it tempting for the board designer to make a neat-looking layout with all the opamps and comparators and feedback resistors and capacitors in neat rows. Unfortunately, this nearness causes some critical interconnections to be a couple of inches apart, and causes other signals to be bussed too closely to each other. And then the designer is puzzled as to why the amplifiers and comparators are oscillating so badly.

So, PC-board designers should be made aware that the layout of linear ICs can be quite critical. The 2-inch spacing that you would never allow, for example, between a digital IC and its bypass capacitor is the same 2-inch spacing that makes an op amp unhappy when its inverting or noninverting input has to travel that distance to various resistors or capacitors. As I will explain in future sections on active-circuit troubleshooting, there are good reasons to keep those summing-point foils short, neat, and compact. Meanwhile it is not fair to assume that the PC board layout-designer will be able to guess which nodes are critical. The circuit-design engineer must provide a list of the nodes that are critical or sensitive-a list of large, noisy signals to keep away from delicate inputs, and so forth. It's only reasonable.

Some engineers like to use narrow PC foil runs, as narrow as possible. Others like to use wide foils and narrow stripes of insulation. Neither one is wrong, but you should be aware of the advantages of using large areas of foil when you are laying out a high-power transistor or IC. The collector of a TO-92 transistor can put a lot of heat down its (copper) collector lead, and an extra square inch of PC foil (on either or on both sides of the PC board) can spread that heat out and help keep the transistor cool. The same is true for high-power ICs: if you look at the data sheet of a medium-power IC such as the LM384, the curves show that 2 square inches of copper foil can help keep the IC much cooler than minimum foil, and 6 square inches is even better, when the device's 6 ground pins carry the heat out of the package into the foil. But some people point out that leaving too much foil on a PC board can cause warping after wave-soldering.

Engineers often assume that a printed-circuit trace has virtually no resistance and no IR drop. But, when you run large currents through a foil run, you will be unpleasantly surprised by the IR drop you'll see. The classic example is a layout where the signal ground for a preamplifier is mixed and shared with the ground-return path for the power supply's bridge and filter capacitor. This return path will see ampere-size surges 120 times per second. Needless to say, that preamplifier won't have "low noise" until the path for the current surges is essentially divorced from the preamp's ground. For precision work, your PC-board layout must include well-thought-out ground paths for your sensitive circuitry.

When you think about it, separating power-supply grounds that carry lots of current from voltage-sensing circuitry is similar to Kelvin connections, which are commonly used in test instruments. A Kelvin connection uses four wires: one pair of leads is meant to carry current and the other pair senses voltage across a device.

Keeping the idea of Kelvin connections in mind when designing your PC board will help you optimize your grounding scheme. In fact, when I draw up my schematic, I label each set of grounds separately. If it has a lot of dirty currents flowing, I keep that separate as Power Ground; if it has to be especially clean, that's a High-Quality ground, indicated with a triangular ground symbol. Then the grounds will be connected together at only one point.

I don't have very much to say about printed-circuit boards for surface-mount devices, because I have not worked on them, and I am not an expert about them. I hear that they are, of course, more challenging, and require more meticulous work in every way. In other words, if you are an expert at engineering ordinary PC boards, you can probably study really hard and do it well. If you are not an expert, well, it's not a good place to start. After you get your board built, one of the worst problems is that of thermal cycling and stresses. Packages such as LCCs (Leadless Chip Carriers) have problems because of their zero lead length-they have no mechanical compliance.

If the PC board material does not have the same thermal coefficient of expansion as the IC package, the soldered joints can be fatigued by the stresses of temp cycling, and may fail early. This is especially likely if you have to cycle it over a wide (military) temperature range--exactly where people wish they had perfect reliability. The commercial surface-mount ICs with gull-wing leads or J-leads are more flexible and more forgiving, and may cause less problems, but you have to do your homework. Too much solder and the leads get stiffened excessively; too little, and there's not enough to hold on.

cont. to part 2 >>

Top of Page

PREV.   NEXT Related Articles HOME