Troubleshooting Analog Circuits--Identifying and Avoiding Transistor Problems

Although transistors--both bipolars and MOSFETs--are immune to many problems, you can still have transistor troubles. Robust design methods and proper assumptions regarding their performance characteristics will steer you past the shoals of transistor vexation and the rocks of transistor disasters.

Transistors are wonderful--they're so powerful and versatile. With a handful of transistors, you can build almost any kind of high-performance circuit: a fast op amp, a video buffer, or a unique logic circuit.

On the other hand, transistors are uniquely adept at causing trouble. For example, a simple amplifier probably won't survive if you short the input to the power supplies or the output to ground. Fortunately, most opamps include forgiving features, so that they can survive these conditions. When the pA741 and the LM101 op amps were designed, they included extra transistors to ensure that their inputs and outputs would survive such abuse. But an individual transistor is vulnerable to damage by excessive forward or reverse current at its input, and almost every transistor is capable of melting. So it's up to us, the engineers, to design transistor circuits so that the transistors do not blow up, and we must troubleshoot these circuits when and if they do.

A simple and sometimes not-so-obvious problem is installing a transistor incorrectly.

Because transistors have three terminals, the possibility of a wrong connection is considerably greater than with a mere diode. Small-signal transistors are often installed so close to a printed-circuit board that you can't see if the leads are crossed or shorted to a transistor's can or to a PC trace. In fact, I recall some boards in which the leads were often crossed and about every tenth transistor was the wrong gender--pnp where an npn should have been, or vice versa. I've thought about it a lot, and I can't think of any circuits that work equally well whether you install a transistor of the opposite sex. So, mind your Ps and Qs. your Ps and Ns, your 2N1302s and 2N1303s, and your 2N3904s and 2N3906s.

In addition to installing a transistor correctly, you must design with it correctly.

First of all, unless they are completely protected from the rest of the world, transistors require input protection. Most transistors can withstand dozens of milliamperes of forward base current but will die if you apply "only a few volts" of forward bias.

One of my pet peeves has to do with adding protective components. MIL-HDBK-217 has always said that a circuit's reliability decreases when components are added. Yet when you add resistors or transistors to protect an amplifier's input or output, the circuit's reliability actually improves. It just goes to show that you can't believe everything you read in a military specification. For detailed criticism of the notion of computing reliability per MIL-HDBK-217, see Ref. 1.

Similarly, if you pump current out of the base of a transistor, the base-emitter junction will break down or "zener." This reverse current--even if it's as low as nano-amperes or very brief in duration-tends to degrade the low-current beta of the transistor, at least on a temporary basis. So in cases where accuracy is important, find a way to avoid reverse-biasing the inputs. Bob Widlar reminded me that the high-current beta of a transistor is generally not degraded by this zenering, so if you are hammering the VEB of a transistor in a switch-mode regulator, that will not necessarily do it any harm, nor degrade its high-current beta.

Transistors are also susceptible to ESD-electrostatic discharge. If you walk across a rug on a dry day, charge yourself up to a few thousand volts, and then touch your finger to an npn's base, it will probably survive because a forward-biased junction can survive a pulse of a few amperes for a small part of a microsecond. But, if you pull up the emitter of a grounded-base NPN stage, or the base of a PNP, you risk reverse-biasing the base-emitter junction. This reverse bias can cause significant damage to the base-emitter junction and might even destroy a small transistor.

When designing an IC, smart designers add clamp diodes, so that any pin can survive a minimum of + and -2000 V of ESD. Many IC pins can typically survive two or three times this amount. These ESD-survival design goals are based on the "human-body" model, in which the impedance equals about 100 pF in series with 1500a. With discrete transistors, whose junctions are considerably larger than the small geometries found in ICs, ESD damage may not be as severe. But in some cases, ESD damage can still happen. Delicate RF transistors such as 2N918s, 2N4275s, and 2N2369s sometimes blow up "when you just look at 'em" because their junctions are so small.

Other transistor-related problems arise when engineers make design assumptions.

Every beginner learns that the VBE of a transistor decreases by about 2 mV per degree Celsius and increases by about 60 mV per decade of current. Don't forget about the side effects of these rules, or misapply them at extreme temperatures. Don't make sloppy assumptions about VBEs. For instance, it's not fair to ask a pair of transistors to have well-matched VBEs if they're located more than 0.1 in. apart and there are heat sources, power sources, cold drafts, or hot breezes in the neighborhood.

Matched pairs of transistors should be glued together for better results. Of course, for best results, monolithic dual transistors like the LM394 give the best matching.

I've seen people get patents on circuits that don't even work-based on misconceptions of the relationships between VBE and current. It's fair to assume that two matched transistors with the same VBE at the same small current will have about the same temperature coefficient of VBE. But you wouldn't want to make any rash assumptions if the two transistors came from different manufacturers or from the same manufacturer at different times. Similarly, transistors from different manufacturers will have different characteristics when going into and coming out of saturation, especially when you're driving the transistors at high speeds. In my experience, a components engineer is a very valuable person to have around and can save you a lot of grief by preventing unqualified components from confusing the performance of your circuits.

Another assumption engineers make has to do with a transistor's failure mode. In many cases, people say that a transistor, like a diode, fails as a short circuit or in a low-impedance mode. But unlike a diode, the transistor is normally connected to its leads with relatively small lead-bond wires; so if there's a lot of energy in the power supply, the short circuit will cause large currents to flow, vaporizing the lead bonds.

As the lead bonds fail, the transistor will ultimately fail as an open circuit.

More Beta--More Better?

It's nice to design with high-beta transistors, and, "if some is good, more's better." But, as with most things in life, too much can be disastrous. The h-parameter, h_rb, is equal to delta_V_BE / delta_V_CB with the base grounded. Many engineers have learned that as beta rises, so does h_rb. As beta rises and h, rises, the transistor's output impedance decreases; its Early voltage falls; its voltage gain decreases; and its common-emitter breakdown voltage, BV_CEO, ay also decrease. (The Early voltage of a transistor is the amount of V, that causes the collector current to increase to approximately two times its low-voltage value, assuming a constant base drive. V_Early is approximately equal to 26 mV X (1/h_rb)). So, in many circuits there is a point where higher beta simply makes the gain lower, not higher.

Another way to effectively increase "beta" is to use the Darlington connection: but the voltage gain and noise may degrade, the response may get flaky, and the base current may decrease only slightly. When I was a kid engineer, I studied the ways that Tektronix made good use of the tubes and transistors in their mainframes and plug-ins. Those engineers didn't use many Darlingtons. To this day, I keep learning more and more reasons not to use Darlingtons or cascaded followers. For many years, it's been more important (in most circuits) to have matched betas than to have sky-high betas. You can match betas yourself, or you can buy monolithic dual matched transistors like the LM394, or you can buy four or five matched transistors on one monolithic substrate, such as an LM3045 or LM3086 monolithic transistor array.

One of the nice things about bipolar transistors is that their transconductance, g",. is quite predictable. At room temperature, g, = 38.6 X IC (This is much more consistent than the forward conductance of diodes, as mentioned in the previous section.) Since the voltage gain is defined as Av = g, X ZL, computing it is often a trivial task.

You may have to adjust this simple equation in certain cases. For instance, if you include an emitter-degeneration resistor, Re, the effective transconductance falls to l/(Re + g,-'). Av is also influenced by temperature changes, bias shifts in the emitter current, hidden impedances in parallel with the load, and the finite output impedance of the transistor. Remember--higher beta devices can have much worse output impedance than normal.

Also be aware that although the transconductance of a well-biased bipolar transistor is quite predictable, beta usually has a wide range and is not nearly as predictable. So you have to watch out for adverse shifts in performance if the beta gets too low or too high and causes shifts in your operating points and biases.

Field Effect Transistors

For a given operating current, field-effect transistors normally have much poorer g, than bipolar transistors do. You'll have to measure your devices to see how much lower. Additionally, the V,, of FETs can cover a very wide range, thus making them harder to bias than bipolars.

JFETs (Junction Field-Effect Transistors) became popular 20 years ago because you could use them to make analog switches with resistances of 30 R and lower.

JFETs also help make good op amps with lower input currents than bipolar devices, at least at moderate or cool temperatures. The BiFET process made it feasible to make JFETs along with bipolars on a monolithic circuit. It's true that the characteristic of the best BiFET inputs are still slightly inferior to the best bipolar ones in terms of V, temperature coefficient, long-term stability, and voltage noise. But these BiFET characteristics keep improving because of improved processing and innovative circuit design. As a result, BiFETs are quite close to bipolar transistors in terms of voltage accuracy. and offer the advantage of low input currents. at room temperature.

FIG. 1. Using equations to analyze circuits can sometimes help you define a problem. But if the equations are inapplicable, they do a lot more harm than good.

JFETs can have a larger gate current when current flows through the source than when no current flows (which is called I&. When I discovered this, and discussed it with Joel Cohen at Crystalonics back 20 years ago, we called it the "Pease-Cohen Effect." I thought it was caused by imperfect ohmic contacts, but other engineers showed that it was actually caused by impact ionization, or "hot carriers." Either way, the gate current has a tendency to increase as a linear function of source current, with an exponential dependence on high drain-source voltages.

I recall working on a hybrid circuit that had some JFETs whose gate connection was supposed to be through the back of the die. I found that some of the dice didn't have proper metallurgical processing, which caused some strange behavior. Initially, the gate acted as if it really were connected to the metal under the die, and would act that way for a long period of time. Then, after a while the gate would act like an open circuit with as much as 1V of error between the actual gate and the bottom of the die.

The amplifier's V,, would grow as large as 1V! The gate would remain disconnected until a voltage transient restored the connection for another week! The intermittency was awful because nothing would speed up the 1-week cycle-to-failure time. So, we had to go back and add definite lead bonds to the gate's bond pad on the top of the chip, which we had been told was unnecessary. Ouch! When designing hybrids, you need to make sure to connect the substrate of a chip to the correct DC level. The bottom of a FET chip is usually tied to the gate, but the connection may be through a large and unspecified impedance. You have to be a pretty good chemist or metallurgist to be sure that you don't have to add that bond to the gate's metal bonding-pad, on the top of the die, just to get a good gate connection.

The substrate of a discrete bipolar transistor's die is the collector. Most linear and digital IC substrates are tied to the negative supply. Exceptions include the LM117 and similar adjustable positive regulators-their substrate is tied to V,,,. The LM196 voltage regulator's substrate is tied to the positive supply voltage, +Vs, as are the substrates of the MM74HC00 family of chips, the NSC LMC660 and LPC660 family, and most of the dielectrically isolated op amps from Harris. So, be aware of your IC's substrate connection. If an LM101AH op amp's metal can should happen to bump against ground or +V,, you have a problem. Similarly, you shouldn't let an HA2525's case bump against ground or -Vs.

MOSFETs are widely used in digital ICs but are also very popular and useful in analog circuits, such as analog switches. The quad switches such as CD4016 and CD4066 are popular because of their low (typical) leakages and low price. Op amps with MOSFET inputs are starting to do well in the general-purpose op-amp market.

MOSFETs used to have a bad reputation for excessive noise, but new IC devices, such as the LMC662 dual op amp, demonstrate that clean processing can cure the problem, thus making MOSFETs competitive with BiFETs. They offer an advantage of a 1000:1 improvement in input current, decreased from 10 pA down to 10 fA. Just be careful not to let ESD near the inputs. Most MOSFET-input linear ICs do have protection diodes and may be able to withstand 600 V, but they usually can't survive 2000 V. If you work with unprotected MOSFETs, such as the 3N160, you must keep the pins securely shorted until the device is soldered into its PC board in which the protection diodes are already installed. I do all of that and wash the transistor package with both an organic solvent and soap and water. And, I keep the sensitive gate circuits entirely off the PC board by pulling the gate pin up in the air and using point-to-point wiring. Air, which is a superior dielectric, is also a good insulator (Ref. 2). So far, I haven't had any blown inputs or bad leakages-at least nothing as bad as 10 fA.

On the other hand, when using CMOS digital ICs, I always plug them into live sockets; I never use conductive foam; and I never wear a ground strap on my wrist.

And I've almost never had any failures-with one exception. One time I shuffled across a carpeted floor and pointed an accusatory finger at a CMOS IC. There was a small crack of ESD--probably 5000 V-followed by a big snap as the IC blew out and crow-barred the entire power supply. Since ESD testing is usually done with the power OFF, then if you did some tests with the power ON, you might get some messy failure modes like the one I just mentioned. Always be wary of any devices that manufacturers claim are safe from ESD.

One reader reminded me that in some cases, if you abuse CMOS ICs with ESD, they may not fail instantly, but they may become unreliable and fail at a later time. So, I must caution you that fooling around with CMOS ICs while you are not properly grounded might cause latent unreliability problems. If you do have to do troubleshooting of CMOS ICs while you are not grounded, if you decide to plug in CMOS ICs while the power busses are hot, just be aware that you might in some cases do some long-lasting harm to an occasional IC. But you have to use your judgment and trade off that possibility against the advantages of more free-swinging troubleshooting approaches.

FIG. 2. When you hit a component or circuit with a pulse of real ESD, you can never be sure what kind of trouble you'll get-unless you've already tested it with an ESD simulator.

Power Transistors May Hog Current

As you build a bipolar transistor bigger and bigger, you may be tempted to go to extremes and make a huge power transistor. But there are practical limitations. Soon, the circuit capacitances cause oppressive drive requirements, and removing the heat is difficult. Still, no matter how big you build power transistors, people will find a use for them. Their most serious limitation on just building transistors bigger and bigger is secondary breakdown, which is what happens when you drive a transistor outside its "safe operating area." When you operate a power transistor at very high currents and low voltages, the distributed emitter resistance of the device-which includes the resistance of the emitter metal and the inherent emitter resistivity-can cause enough I x R drop to force the entire emitter and its periphery to share the current. Now, let's halve the current and double the voltage: The amount of dissipation is the same, but the I x R drop is cut in half. Now continue to halve the current and double the voltage. Soon you'll reach a point where the ballasting (FIG. 3) won't be sufficient, and a hot spot will develop at a high-power point along the emitter. The inherent decrease of VBE will cause an increase of current in one small area. Unless this current is turned OFF promptly, it will continue to increase unchecked. This "current hogging" will cause local overheating, and may cause the area to melt or crater--this is what happens in "secondary breakdown." By definition you have exceeded the secondary breakdown of the device. The designers of linear ICs use ballasting, cellular layouts, and thermal-limiting techniques, all of which can prevent harm in these cases (Ref. 3). Some discrete transistors are beginning to include these features.

Fortunately, many manufacturers' data sheets include permitted safe-area curves at various voltages and for various effective pulse-widths. So, it's possible to design reliable power circuits with ordinary power transistors. The probability of an unreliable design or trouble increases as the power level increases, as the voltage increases, as the adequacy of the heat sink decreases, and as the safety margins shrink. For example, if the bolts on a heat sink aren't tightened enough, the thermal path degrades and the part can run excessively hot.

FIG. 3. Ballast resistors, also known as sharing resistors, are often connected to the emitters of a number of paralleled transistors (a) to help the transistors share current and power. In an integrated circuit (b), the ballast resistors are often integrated with adjacent emitters. (Photo of National Semiconductor Corp's LM 138.) High temperature per se does not cause a power transistor to fail. But, if the drive circuitry was designed to turn a transistor ON and only a base-emitter resistor is available to turn it OFF, then at a very high temperature, the transistor will turn itself ON and there will be no adequate way to turn it OFF. Then it may go into secondary breakdown and overheat and fail. However, overheating does not by itself cause failure. I once applied a soldering iron to a 3-terminal voltage regulator-I hung it from the tip of the soldering iron-and then ran off to answer the phone. When I came back the next day, I discovered that the TO-3 package was still quite hot: +300 C, which is normally recommended for only 10 seconds. When I cooled it off, the regulator ran fine and met spec. So, the old dictum that high temperature will necessarily degrade reliability is not always true. Still, it's a good practice to not get your power transistors that hot, and to have a base drive that can pull the base OFF if they do get hot.

You can also run into problems if you tighten the screws on the heat sink too tight, or if the heat sink under the device is warped, or if it has bumps or burrs or foreign matter on it. If you tighten the bolt too much, you'll overstress and warp the tab and die attach. Overstress may cause the die to pop right off the tab. The insulating washer under the power transistor can crack due to overstress or may fail after days or weeks or months. Even if you don't have an insulating washer, over-torqueing the bolts of plastic-packaged power transistors is one of the few ways a user can mistreat and kill these devices. Why does the number 10 inch-pounds max, 5 typ, stick in my head? Because that's the spec the Thermalloy man gave me for the 6/32 mounting bolts of TO-220 packages. For any other package, make sure you have the right spec for the torque. Don't hire a gorilla to tighten the bolts.

Apply the 5-Second Rule

Your finger is a pretty good heat detector--just be careful not to burn it with high voltages or very hot devices. A good rule of thumb is the 5-second rule: If you can hold your finger on a hot device for 5 seconds, the heat sink is about right, and the case temperature is about 85 C. If a component is hotter than that, too hot to touch, then dot your finger with saliva and apply it to the hot object for just a fraction of a second. If the moisture dries up quickly, the case is probably around 100 C; if it sizzles instantaneously, the case may be as hot as 140 C. Alternatively, you can buy an infrared imaging detector for a price of several thousand dollars, and you won't bum your fingers. You will get beautiful color images on the TV screen, and contour maps of isothermal areas. You will learn a lot from those pictures. About twice a year, I wish I could borrow or rent one.

FIG. 4. When using high-power amplifiers, there are certain problems you just never have if you use a big-enough heat sink. This heat sink's thermal resistance is lower than 0.5 C/W.

Fabrication Structures Make a Difference

Another thing you should know when using bipolar power transistors is that there are two major fabrication structures: the epitaxial base, and the planar structure pioneered by Fairchild Semiconductor (FIG. 5) (Ref. 4). (See my comments a couple paragraphs down concerning the obsolete single-diffused transistors.) Transistors fabricated with the epi-base structure are usually more rugged and have a wider safe operating area. Planar devices feature faster switching speeds and higher frequency response, but aren't as rugged as the epi-base types. You can compare the two types by looking at the data sheets for the Motorola 2N3771 and the Harris 2N5039. The 2N.5039 planar device has a current-gain bandwidth 10 times greater than the 2N3771 epi-base device. The 2N5039 also has a switching speed faster than the 2N3771 when used as a saturated switch, but the 2N3771 has a considerably larger safe area if used for switching inductive loads. You can select the characteristics you prefer, and order the type you need ....

But be careful. If you breadboard with one type and then start building in production with the other, you might suddenly find that the bandwidth of the transistor has changed by a factor of 10 (or a factor of 0.1) or that the safe area doesn't match that of the prototypes. Also be aware that the planar power devices, like the familiar 2N2222 and 2N3904, are quite capable of oscillating at high frequencies in the dozens of megahertz when operated in the linear region, so you should plan to use beads in the base and/or the emitter, to quash the oscillation. The slower epi-base devices don't need that help very often.

When I first wrote these articles on troubleshooting back in 1988, you could still buy the older "single-diffused'' transistors such as 2N3055H and the old 2N3771. I wrote all about how these devices had even more Safe Operating Area (SOA) than the epi-base device, so you might want to order these if you wanted a "really gutsy" transistor for driving inductive loads. Unfortunately, these transistors were obsolescent and obsolete; they were slow (perhaps 0.5 MHz of fa), had a large die area, and were expensive. For example, although these transistors required only one diffusion, in some cases this diffusion had to run 20 hours. Because of all these technical reasons, sales shrank until, in the last 2 years, all the single-diffused power transistors have been discontinued.

FIG. 5. The characteristics of power transistors depend on their fabrication structure. The epitaxial base structure (a) takes advantage of the properties of several different epitaxial layers to achieve good beta, good speed, low saturation, small die size, and low cost. This structure involves mesa etching, which accounts for the slopes at the die edges. Planar power transistors (b) can achieve very small geometries, small base-widths, and high-frequency responses, but they're less rugged than epitaxial-base types, in terms of Forward-Biased SOA.

So it's kind of academic to talk about the old single-diffused parts, (see FIG. 6) but I included a mention here just for historical interest. Also, I included it because if you looked in my old EDN write-up and then tried to buy the devices I recommended.

you would meet with incredulity. You might begin to question the sanity of yourself, or the salesman, or of Pease. When I inquired into the availability of these parts, I talked to many sales people who had no idea what I was talking about. Finally when I was able to talk to technical people, they explained why these transistors were not available-they admitted that I was not dreaming, but that the parts had been discontinued recently. These engineers at some of the major power-transistor manufacturers were quite helpful as they explained that newer geometries helped planar power transistors approach the safe area of the other older types without sacrificing the planar advantages of speed. Also, power MOSFETs had even wider amounts of SOA, and their prices have been dropping, and they were able to take over many new tasks where the planars did not have enough SOA. So the puzzle all fits together.

There is still one tricky problem. Originally the old 2N3771 was a single-diffused part. If you wanted to buy an epi-base part, that was the MJ3771. But now if you order a 2N3771, you get the epi-base part, which does meet and exceed the JEDEC 2N3771 specs. It just exceeds them a lot more than you would expect-like, the current-gain-bandwidth is 10 or 20x higher. So, if you try to replace an old 2N3771 with a new 2N3771, please be aware that they are probably not very similar at all.

FIG. 6. In the old single-diffused structure, n-type dopants were diffused simultaneously into the front and back of a thin p-type wafer. This structure produced rugged transistors with wider Safe Operating Areas than the more modern epitaxial-base transistor types, in terms of Forward Biased SOA. However, this fabrication has been obsoleted.

Power-Circuit Design Requires Expertise

For many power circuits, your transistor choice may not be as clear-cut as in the previous examples. So, be careful. Design in this area is not for the hotshot just out of school-there are many tricky problems that can challenge even the most experienced designers. For example, if you try to add small ballasting resistors to ensure current sharing between several transistors, you may still have to do some transistor matching. This matching isn't easy. You'll need to consider your operating conditions; decide what parameters, such as beta and VBE, you'll match; and figure out how to avoid the mix-and-match of different manufacturers' devices. Such design questions are not trivial. When the performance or reliability of a power circuit is poor, it's probably not the fault of a bad transistor. Instead, it's quite possibly the fault of a bad or marginal driver circuit or an inadequate heat sink. Perhaps a device with different characteristics was inadvertently substituted in place of the intended device. Or perhaps you chose the wrong transistor for the application.

A possible scenario goes something like this. You build 10 prototypes, and they seem to work okay. You build 100 more, and half of them don't work. You ask me for advice. I ask, "Did they ever work right?" And you reply, "Yes." But wait a minute. There were 10 prototypes that worked, but the circuit design may have been a marginal one. Maybe the prototypes didn't really work all that well. If they're still around, it would be useful to go back and see if they had any margin to spare. If the 10 prototypes had a gain of 22,000, but the current crop of circuits has gains of 18,000 and fails the minimum spec of 20,000, your new units should not be called "failures." It's not that the circuit isn't working at all, it's just that your expectations were unrealistic.

After all, every engineer has seen circuits that had no right to work, but they did work--for a while. And then when they began to fail, it was obviously just a hopeless case. So, which will bum you quickest, a marginal design or marginal components? That's impossible to say. If you build in some safety margin, you may survive some of each. But you can't design with big margins to cover every possibility, or your design will become a monster. That's where experience and judgment must be invoked....

An old friend wrote to me from Japan, "Why do you talk about having to troubleshoot 40% of the units in a batch of switching regulators? In Japan that would be considered a bad design. , . ." I replied that I agreed that it sounds like a problem, but until you see what is the cause of the problems, it is unfair to throw any blame around. What if it was a bad workmanship problem? Then that does not sound like a bad design-unless the design was so difficult to execute that the assembly instructions could not be followed. Or maybe a bad part was put in the circuit. Or maybe it was a marginally bad design and part of the circuit does need to be changed-perhaps an extra test or screening of some components-before the circuit can run in production. But you cannot just say that if there is ever trouble, it is the design engineer's fault. What if the design engineer designed a switching regulator that never had any problems in production-never ever-but it only puts out 1W per 8 cubic inches, and all the parts are very expensive, and then there is a lot of expensive testing on each component before assembly, to prove that there is a good safety margin. Is that a good design? I doubt it. Because if you tried to build a plane with too big a safety factor, it might be bigger than a 747, but able to carry only 10 passengers.

Every circuit should be built with an appropriate safety factor. If you only use a transistor that is always SURE to work well, that may be an uneconomic safety factor. Judgment is required to get the right safety factor.

MOSFETS Avoid Secondary Breakdown

When it comes to power transistors, MOSFETs have certain advantages. For many years, MOSFETs have been available that switch faster than bipolar transistors, with smaller drive requirements. And MOSFETs are inherently stable against secondary breakdown and current hogging because the temperature coefficient of IDS vs. V,, is inherently stable at high current densities. If one area of the power device gets too hot, it tends to carry less current and thus has an inherent mechanism to avoid running away. This self-ballasting characteristic is a major reason for the popularity of MOSFETs over bipolar transistors. However, recent criticism points out that when you run a MOSFET at high-enough voltages and low current, the current density gets very small, the temperature coefficient of IDS vs. I_GS reverses, and the device's inherent freedom from current hogging may be lost (Ref. 5). So at high voltages and low current densities, watch out for this possibility. When the V_DS gets high enough, MOSFETS can exhibit current hogging and "secondary breakdown" similar to that of bipolars!

The newer power MOSFETs are considerably more reliable and less expensive than the older devices. Even though you may need a lot of transient milliamps to turn the gate ON or OFF quickly, you don't need a lot of amps to hold it ON like you do with a bipolar transistor. You can turn the newer devices OFF quicker, too, if you have enough transient gate drive current available.

However, MOSFETs are not without their problem areas. If you persist in dissipating too many watts into a MOSFET, you can melt it just as you can melt a bipolar device. If you don't overheat a MOSFET, the easiest way to cause a problem is to forget to insert a few dozen or hundred ohms of resistance (or a ferrite bead) right at the gate lead of the device. Otherwise, these devices have such high bandwidths that they can oscillate at much higher frequencies than bipolar transistors.

For example, the first high-fidelity, all-MOSFET audio amplifier I ever saw blew up. It worked okay in the lab, but some misguided engineer decided that if a band width of 5 Hz to 50 kHz was good, then 0.5 Hz to 500 kHz was better. Consequently, when the speaker cables were extended from 10 feet to 20 feet for a demonstration, the amplifier broke into a megahertz-region scream and promptly went up in smoke because of the lack of damping at the sources. I was told that after a minor redesign the amplifier was perfectly reliable. The redesign involved cutting the bandwidth down to a reasonable level, adding some ballasting in the sources, and tying antisnivet resistors directly to the gate pins. (Note: A snivet is a nasty, high-frequency oscillation originally found in vacuum-tube TV sets-an oscillation similar to the oscillation of a MOSFET with no resistance in series with its gate.) As with bipolar transistors, MOSFETs are very reliable if you don't exceed their voltage, current, or temperature ratings. Dissatisfaction with a device's reliability or performance usually stems from the drivers or the related circuitry. Most MOSFETs have a maximum VGs rating of just 20 or 25 V. A MOSFET may temporarily survive operation with 30 or 50 V on the gate, but it's not safe to run it up there forever. If you apply excessive gate voltage, gradual gain or threshold degradation may occur.

So-please-don't. Also, power MOSFETs are not quite as rugged as bipolars when it comes to surviving ESD transients. A common precaution is to add a little decoupling, clamping, or current-limiting circuitry, so that terminals accessible to the outside world can withstand ESD.

DMOS FETs are so easy to apply that we usually forget about the parasitic bipolar transistor that lurks in parallel with them. If dV/dt is too large at the drain, if the drain junction is avalanched at too high a current and voltage, or if the transistor gets too hot, the bipolar device turns ON and dies an instant death due to current hogging or an excursion from its safe operating area.

But I'm spoiled rotten. I'm accustomed to linear ICs, which have protection transistors built right in, so the user rarely has a problem. (But most of the transistor troubles are left to the IC designer!) Discrete designs are appropriate and cost-effective for many applications, but the availability of linear ICs--especially opamps--can simplify your design task considerably, at the same time as it improves reliability.

Next time, we'll discuss the ins and outs and innards of opamps.

References

1. Leonard, Charles, "Is reliability prediction methodology for the birds?" Power Conversion and Intelligent Motion, November 1988, p. 4.

2. Pease, Robert A., "Picoammeter/calibrator system eases low-current measurements," EDN, March 31, 1982, p. 143.

3. "A 150W IC Op Amp Simplifies Design of Power Circuits," R. J. Widlar and M. Yamatake, AN-446, National Semiconductor Corp, Santa Clara, CA.

4. Applications Engineering Staff, PowerTech Inc., "Speed-up inductor increases switching speed of high current power transistors," Power Electronics, May 1989, p. 78.

5. Passafiume, Samuel J., and William J. Nicholas, "Determining a MOSFET's real FBSOA," Powertechnics Magazine, June 1989, p. 48.